{
  "version": 3,
  "sources": ["ssg:https://framerusercontent.com/modules/YVgynkVt96a0ES9Gk8yz/NNRbmpG9wcazgf7aJM3g/eo4RAmtig-24.js"],
  "sourcesContent": ["import{jsx as e,jsxs as t}from\"react/jsx-runtime\";import{Link as n}from\"framer\";import{motion as a}from\"framer-motion\";import*as i from\"react\";export const richText=/*#__PURE__*/t(i.Fragment,{children:[/*#__PURE__*/e(\"p\",{children:\"Machine learning is nothing short of astounding. Yet, learning is only one side of the equation \u2013 the leap from potential to performance lies entirely in the quality and precision of the training.\"}),/*#__PURE__*/t(\"p\",{children:[\"Fact: large language models (LLMs) require an astonishing amount of data \u2013 but not just any data. While humans produce a staggering\",/*#__PURE__*/e(n,{href:\"https://explodingtopics.com/blog/data-generated-per-day\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\" 402.74 million terabytes\"})}),\" bytes of data per day, even the most sophisticated large language models (LLMs) would drown in this torrent without the proper training.\\xa0\"]}),/*#__PURE__*/e(\"p\",{children:\"Systems have to be extensively fine-tuned to ensure LLM models generate safe, accurate, and reliable responses.\"}),/*#__PURE__*/e(\"p\",{children:\"In this article, we explore some of the processes behind training, fine-tuning, and optimizing large language models (LLMs) for different uses.\"}),/*#__PURE__*/e(n,{href:\"https://toloka.ai/large-language-models/?utm_source=blog&utm_medium=banner&utm_campaign=sftrlhf&utm_content=howllmsaretrained\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!0,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{className:\"framer-image\",\"data-preset-tag\":\"img\",children:/*#__PURE__*/e(\"img\",{alt:\"\",className:\"framer-image\",height:\"240\",src:\"https://framerusercontent.com/images/L1MWCdKVPqiR785bI54iY2TIc9k.png\",srcSet:\"https://framerusercontent.com/images/L1MWCdKVPqiR785bI54iY2TIc9k.png?scale-down-to=512 512w,https://framerusercontent.com/images/L1MWCdKVPqiR785bI54iY2TIc9k.png?scale-down-to=1024 1024w,https://framerusercontent.com/images/L1MWCdKVPqiR785bI54iY2TIc9k.png 1498w\",style:{aspectRatio:\"1498 / 480\"},width:\"749\"})})}),/*#__PURE__*/e(\"h2\",{children:\"Definition of a Language Model (LM)\"}),/*#__PURE__*/e(\"p\",{children:\"A language model (LM) is an artificial intelligence program designed to learn, interpret, and generate human language. It works by predicting the likelihood of the following phrase or word based on the context provided by the preceding words.\"}),/*#__PURE__*/t(\"p\",{children:[\"Traditional language models utilize statistical methods \u2013 modern LLMs employ deep learning techniques (think\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/blog/transformer-architecture/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\" transformers\"})}),\") to produce more accurate and coherent natural language processing outputs.\"]}),/*#__PURE__*/e(\"p\",{children:\"In recent years, language models have evolved from rule-based systems to statistical methods and, most recently, to neural network-based architectures, such as transformers, which drive state-of-the-art performance\"}),/*#__PURE__*/e(\"p\",{children:\"Early models, such as n-grams, relied heavily on fixed sequences of words and had limitations in capturing long-range dependencies.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"The advent of neural networks, particularly recurrent neural networks (RNNs) and later transformer model architecture, fundamentally changed the landscape, enabling your typical base model to capture far more sophisticated language patterns.\"}),/*#__PURE__*/e(\"h2\",{children:\"What is a Large Language Model (LLM)?\"}),/*#__PURE__*/e(\"p\",{children:\"A large language model (LLM) is a specific language model trained with enormous data sets and contains billions or even trillions of parameters.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"LLMs can perform sophisticated, highly complete tasks such as language and text generation while also being able to answer questions based on context.\"}),/*#__PURE__*/e(\"p\",{children:\"Unlike conventional language models that operate under a more limited set of rules, LLMs learn from vast datasets that enable them to capture complex patterns and relationships in language.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"For example, models like GPT-3 and GPT-4 are designed to understand and generate human-like text with impressive accuracy. But in order to work effectively, these models need to be calibrated with specialized techniques like data labeling, automatic data annotation, and fine-tuning.\"}),/*#__PURE__*/e(\"h3\",{children:\"Primary characteristics of LLMs\"}),/*#__PURE__*/t(\"ul\",{children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Scale\"}),\": LLMs' size, measured in parameters, allows them to access immeasurable amounts of information from the data they are trained on.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Transfer learning\"}),\": Following initial training, LLMs can be modified to perform specific tasks, making them highly versatile across various applications.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Understanding context\"}),\": LLMs use attention mechanisms, which help them focus on the most relevant segments of their input data.\"]})})]}),/*#__PURE__*/e(\"h2\",{children:\"Training process\"}),/*#__PURE__*/e(\"p\",{children:\"There are quite a few steps involved in training LLMs. Let's unpack them:\"}),/*#__PURE__*/e(\"h3\",{children:\"Pre-training Large Language Models\"}),/*#__PURE__*/e(\"p\",{children:\"First, it\u2019s paramount that LLMs go through a pre-training phase, during which each model is exposed to vast amounts of data. At the same time, choosing the right architecture to process and learn from this data is equally important.\"}),/*#__PURE__*/e(\"p\",{children:\"As the LLM processes information, it learns to recognize patterns, understand complex relationships, and grasp sophisticated language structures. This could include syntax, discourse, and lexical semantics to capture finer nuances that AI, like ChatGPT, is known for.\"}),/*#__PURE__*/e(\"p\",{children:\"Developers can also train the model on general world knowledge by adding books, websites, encyclopedias, and other knowledge-based sources to its training data. Training data helps the model discern subtle differences and grasp concepts beyond language only.\"}),/*#__PURE__*/e(\"h3\",{children:\"The fine-tuning process\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Next, pre-trained language models undergo a process called fine-tuning. Fine-tuning is crucial in adapting the pre-trained model for specific tasks or industries.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"This phase involves training the model on a smaller, more targeted dataset containing labeled data specific to the particular task, such as customer support dialogues or medical research papers.\"}),/*#__PURE__*/e(\"h3\",{children:\"Reinforcement Learning\"}),/*#__PURE__*/e(\"p\",{children:\"Reinforcement Learning from Human Feedback (RLHF): RLHF is applied as a second step to align the model with human preferences, focusing on being helpful.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"RLHF is the last box to check in the training stack. It encourages desired behavior and discourages undesired results by rewarding actions that lead to positive output and penalizing those that don't.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Using this method, the model begins to understand instructions and learns to retrieve knowledge based on them.\"}),/*#__PURE__*/e(\"h3\",{children:\"Evaluating the LLMs performance\"}),/*#__PURE__*/e(\"p\",{children:\"Evaluating large language models (LLMs) is incredibly important to ensure they perform as well as expected in real-world environments.\"}),/*#__PURE__*/e(\"p\",{children:\"Unlike conventional software testing, which prioritizes functionality, evaluating LLMs requires a closer look at how well they interpret context, maintain logical flow, and produce meaningful and coherent outputs.\"}),/*#__PURE__*/e(\"p\",{children:\"Further controls, such as BLEU, ROUGE, perplexity, and human evaluation scores, are often called upon to provide a more complete picture of the model's performance \u2013 especially in cases requiring subtle or ambiguous language processing.\"}),/*#__PURE__*/e(\"p\",{children:\"As LLMs take on many challenges and topics, evaluating their ability to generalize is also essential. In other words, how well can they process situations and data that they haven't encountered before?\"}),/*#__PURE__*/t(\"p\",{children:[\"To achieve this, \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/blog/evaluating-llms/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"methods like cross-validation\"})}),\" help prevent overfitting, while tools such as confusion matrices offer detailed insights into the model's strengths and weaknesses.\"]}),/*#__PURE__*/e(\"p\",{children:\"Apart from performance metrics, speed, and scalability are all noteworthy factors to consider.\"}),/*#__PURE__*/e(\"p\",{children:\"Since LLMs are used in a variety of settings, they must be able to react quickly without sacrificing accuracy or dependability. The evaluation process also heavily weighs ethical considerations. For the model to be used in the real world, it must be free of biases, transparent in its responses, and compliant with ethical AI standards.\"}),/*#__PURE__*/e(\"p\",{children:\"The ultimate objective of LLM evaluation is to make sure the model can manage the unpredictability of real-world applications while upholding strict standards of responsibility and fairness, in addition to ensuring performance under controlled circumstances.\"}),/*#__PURE__*/e(\"h2\",{children:\"Importance of labeled data\"}),/*#__PURE__*/e(\"p\",{children:\"Automated data labeling and human-in-the-loop approaches can significantly reduce the time and cost required to prepare domain-specific datasets.\"}),/*#__PURE__*/e(\"p\",{children:\"High-quality labels ensure the model learns to make accurate predictions relevant to the specific task and field.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"This often requires domain experts to curate and verify the labeled data, especially in specialized fields like medicine, law, engineering, and coding.\"}),/*#__PURE__*/e(\"h2\",{children:\"What are the limitations of LLMs, and how can they be resolved?\"}),/*#__PURE__*/e(\"p\",{children:\"While LLMs are powerful tools, they do come with some limitations.\"}),/*#__PURE__*/e(\"p\",{children:\"To work around this, datasets are carefully curated, and fine-tuning must involve diverse, representative examples. Let\u2019s take a closer look at some of these challenges and how they can be remedied:\"}),/*#__PURE__*/e(\"h3\",{children:\"Key challenges of LLMs\"}),/*#__PURE__*/t(\"ul\",{children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Resource-intensive training\"}),\": Training LLMs requires substantial computational resources. To overcome this challenge, developers can use more efficient training techniques and leverage cloud computing to distribute the workload.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Lack of domain expertise:\"}),\" Large language models are 'general' models that may not perform well in niche domains without the correct fine-tuning.\\xa0\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Bias in training data: \"}),\"LLMs can inherit biases in the data they are trained on, leading to biased outputs.\\xa0\"]})})]}),/*#__PURE__*/e(\"h3\",{children:\"Popular strategies to remedy limitations\"}),/*#__PURE__*/t(\"ul\",{children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Domain-specific training: \"}),\"Investing in domain-specific training datasets can improve model performance significantly in specialized areas.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Bias audits\"}),\": Regularly test models for biased outputs in different contexts and adjust training data accordingly.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Debiasing algorithms\"}),\": Implement algorithms designed to minimize bias in model predictions, ensuring fairer outputs.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Improved reasoning:\"}),\" Large language models excel at pattern recognition, but to do complex reasoning tasks even better, developers are experimenting with hybrid models that combine LLMs with other AI techniques, such as symbolic reasoning.\"]})})]}),/*#__PURE__*/e(\"h2\",{children:\"Why are fine-tuning and humanizing LLMs important?\"}),/*#__PURE__*/t(\"p\",{children:[\"Fine-tuning and\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/blog/llms-and-humans-for-data-labeling/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\" humanizing LLMs\"})}),\" are essential for enhancing their performance and adaptability to specific use cases.\\xa0\"]}),/*#__PURE__*/e(\"p\",{children:\"Fine-tuning allows an LLM to specialize in a particular domain or task, such as medical screening or legal document analysis.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Humanizing LLMs helps generate responses that are more natural, empathetic, and contextually appropriate \u2013 qualities considered especially vital for customer-facing applications like chatbots.\"}),/*#__PURE__*/e(\"h3\",{children:\"Why does empathy in AI matter?\"}),/*#__PURE__*/e(\"p\",{children:\"Empathy can vastly improve the user experience \u2013 especially in applications like customer service and mental health support. Humanizing responses can build trust and enhance user satisfaction.\"}),/*#__PURE__*/e(\"p\",{children:\"The key to successful fine-tuning lies in the quality and variety of the labeled data used during training.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Without properly labeled data that reflects the nuances of the target domain, the model's performance can be suboptimal, leading to inaccurate or irrelevant outputs.\"}),/*#__PURE__*/e(\"h2\",{children:\"What are popular strategies to fine-tune LLMs?\"}),/*#__PURE__*/e(\"p\",{children:\"There are a few strategies to refine large language models to help them perform better:\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"1. Transfer learning\"})}),/*#__PURE__*/e(\"p\",{children:\"Transfer learning involves using a pre-trained LLM and adapting it to a new domain by fine-tuning it on a smaller, domain-specific training dataset.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"2. Active learning\"})}),/*#__PURE__*/e(\"p\",{children:\"Active learning is an approach where the model selects the most informative or uncertain examples for human annotation.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"By focusing on the examples that the model is most unsure about, active learning improves the quality of the labeled data without requiring exhaustive manual effort.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"3. Human-in-the-Loop\"})}),/*#__PURE__*/e(\"p\",{children:\"Human-in-the-loop (HITL) systems combine human judgment with machine learning.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"During data preparation for fine-tuning, the model's predictions are reviewed and corrected by human annotators, ensuring that the labeled data is accurate and high-quality.\"}),/*#__PURE__*/e(\"h3\",{children:\"Automatic data annotation in LLM training\"}),/*#__PURE__*/e(\"p\",{children:\"As data labeling becomes increasingly important for LLM training, it is vital to focus on combining expert-curated data with synthetic data.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"This method significantly accelerates the training process while ensuring the accuracy and relevance of the labeled data.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"By leveraging expert knowledge in specific domains alongside synthetic datasets, scaling can be achieved effectively while maintaining high standards of quality.\"}),/*#__PURE__*/e(\"h3\",{children:\"Benefits of human annotation in LLM training\"}),/*#__PURE__*/e(\"p\",{children:\"While automatic annotation systems help accelerate the data preparation process, human annotation remains critical for ensuring accuracy, contextual understanding, and handling edge cases that automated tools might miss.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Human annotators can provide nuanced insights, particularly in specialized or complex domains, where machine learning models may struggle to grasp subtle details or interpret ambiguous data.\"}),/*#__PURE__*/e(\"p\",{children:\"By combining human input with automatic annotation, we achieve a higher level of precision in labeled datasets, reducing errors and ensuring the model learns in ways that are contextually relevant and reflective of real-world applications.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Furthermore, human feedback helps refine the model\u2019s learning process, enabling continuous improvement over time.\"}),/*#__PURE__*/e(\"p\",{children:\"However, while automatic annotation tools are highly efficient, human verification is still necessary to ensure the accuracy of labeled data.\"}),/*#__PURE__*/e(\"h2\",{children:\"Toloka\u2019s approach to data collection\"}),/*#__PURE__*/e(\"p\",{children:\"Wherever data is collected, we prioritize the integration of expert-curated data alongside synthetic data.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Following this approach ensures high-quality, domain-specific data that improves model accuracy while enhancing its ability to generalize across a range of applications.\"}),/*#__PURE__*/e(\"p\",{children:\"Expert data provides deep domain knowledge, while synthetic data allows us to scale our datasets efficiently, offering a comprehensive foundation for model training.\"}),/*#__PURE__*/e(\"h2\",{children:\"Future trends in LLM training and fine-tuning\"}),/*#__PURE__*/e(\"p\",{children:\"As we move forward, the landscape of LLM training and fine-tuning is expected to evolve significantly. Here are some anticipated trends:\"}),/*#__PURE__*/e(\"h3\",{children:\"Increased focus on ethical AI\"}),/*#__PURE__*/e(\"p\",{children:\"With rising concerns about bias and ethical implications in AI, developers are expected to prioritize ethical considerations in model training.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Matters in the spotlight include transparent practices, responsible data sourcing, and mechanisms to audit and correct biases in LLM outputs.\"}),/*#__PURE__*/e(\"h3\",{children:\"More efficient training techniques\"}),/*#__PURE__*/e(\"p\",{children:\"Ongoing research into more efficient training techniques, such as model pruning, quantization, and advanced optimization algorithms, will enable developers to train powerful models with reduced computational resources.\"}),/*#__PURE__*/e(\"h3\",{children:\"Example techniques\"}),/*#__PURE__*/t(\"ul\",{children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Model pruning\"}),\": involves removing weights from the model that contribute the least to its performance, resulting in smaller and faster models.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Quantization\"}),\": Reduces the precision of the numbers used in model calculations, which can significantly decrease the model size and increase inference speed with minimal impact on accuracy.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Enhanced human-AI collaboration\"})})})]}),/*#__PURE__*/e(\"p\",{children:\"The integration of human expertise into the model training process will continue to expand, promoting a collaborative environment where AI supports human decision-making and vice versa.\"}),/*#__PURE__*/e(\"h3\",{children:\"Examples of collaboration\"}),/*#__PURE__*/t(\"ul\",{children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Co-creation\"}),\": Involving domain experts in the training process to ensure that models align closely with real-world applications and user needs.\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Feedback loops\"}),\": Establishing systems where user interactions with AI can inform and improve model training, creating a cycle of continuous learning.\"]})})]}),/*#__PURE__*/e(\"h2\",{children:\"The path forward\"}),/*#__PURE__*/e(\"p\",{children:\"LLMs have revolutionized the field of AI by enabling machines to understand and generate human-like text.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Still, their training needs careful attention to data labeling, automated data labeling, and fine-tuning.\"}),/*#__PURE__*/e(\"p\",{children:\"As the tech behind large language models (LLMs) continues to evolve, the future of generative AI promises even greater advancements.\"}),/*#__PURE__*/e(\"p\",{children:\"The question is how we will continue to grow our interaction with machines \u2013 for industries ranging from healthcare to finance, education and more.\"}),/*#__PURE__*/t(\"p\",{children:[\"For in-depth insights on fine-tuning LLMs with high-quality, domain-specific data,\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/blog/the-genai-frontier-and-the-quest-for-high-quality-sft-data/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\" read more\"})}),\" on the quest for high-quality SFT data.\"]})]});export const richText1=/*#__PURE__*/t(i.Fragment,{children:[/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(n,{href:\"https://toloka.ai/moderation/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"Content moderation\"})}),\" is something that any company with an online presence cannot exist without. If not given proper consideration to this endeavor, a business can endure a loss of money, opportunities, and most importantly, reputation.\"]}),/*#__PURE__*/e(\"p\",{children:\"Supporting your own content moderation team is undeniably challenging. It primarily requires a considerable investment of funds, something that not everyone can afford. An excellent solution in this case is outsourcing. Here we will consider all the advantages of this approach.\"}),/*#__PURE__*/e(\"h2\",{children:\"Why do Businesses Need Content Moderation?\"}),/*#__PURE__*/t(\"p\",{children:[\"Essentially, \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/moderation/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"content moderation\"})}),\" is about recognizing harmful content in texts, images, or videos on online resources. Human content moderators act as censors to ensure that user-generated content is in line with community guidelines and standards.\"]}),/*#__PURE__*/e(\"p\",{children:\"Ensuring a safe and secure online environment is becoming increasingly challenging as user-generated content keeps multiplying. This massive amount of data demands the most sophisticated technologies to control and moderate it. Artificial Intelligence (AI) brings innovative solutions to detect malicious or inappropriate content. It makes the moderation process faster and more efficient, but AI cannot yet fully replace humans, so often both AI and human moderation are used together. For instance, a built-in website system is capable of assessing the appropriateness and compliance of a text with a specific topic or the entire online platform. If the user-generated content is deemed inappropriate, it will be manually reviewed, and then a moderator will decide on whether to publish it.\"}),/*#__PURE__*/e(\"p\",{children:\"Content moderation serves businesses as a way to ensure that the website meets their standards and safeguards customer experiences and the company's online reputation. Unimaginable amounts of text, images, and videos are shared on a daily basis, making it essential for businesses to track the content that's stored on their platforms.\"}),/*#__PURE__*/e(\"p\",{children:\"Through user-generated content moderation, businesses can be sure that their online platforms serve their original function. This is vital for providing a trusted and risk-free experience for clients.\"}),/*#__PURE__*/e(\"h2\",{children:\"Why does your business need outsourcing content moderation?\"}),/*#__PURE__*/t(\"p\",{children:[\"Before realizing if outsourcing \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/moderation/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"content moderation\"})}),\" is what you need, there are several aspects you have to realize. First and foremost, content moderation as a rule involves an in-house team of hand-picked, extremely qualified moderators. Besides people, content moderation requires top-quality software that employs sophisticated machine learning algorithms. Performing all of this on your own requires a considerable amount of resources. In-house content moderation tends to be a more costly option than outsourcing. In order to recruit a sizable team of content moderators and afford funding for the development and implementation of automated software, your business is required to have a huge budget.\"]}),/*#__PURE__*/e(\"p\",{children:\"Budget is possibly one of the most essential aspects to consider when hiring an outsourcing partner. Nonetheless, factors such as the diversity and amount of user-generated content that is being posted on your platform also have to be taken into account. Keep in mind that you will need moderators who are aware of local cultural specifics and the languages spoken by your customers.\"}),/*#__PURE__*/e(\"p\",{children:\"Before you start looking for content moderation outsourcing partners, the question you need to ask yourself is: Do you possess enough internal expertise regarding upcoming projects and technical solutions to manage an in-house team and achieve the desired result? If the answer is negative, you better entrust your moderation process to the professionals.\"}),/*#__PURE__*/e(\"p\",{children:\"By outsourcing content moderation, you don't need to build a content moderation team from scratch and provide training. Outsourcing companies have years of experience, a reliable and secure infrastructure that is constantly being improved. Among other things, outsourcing offers automated workflows that integrate human expertise and machine learning, resource and experience sharing as well as cheaper labor costs.\"}),/*#__PURE__*/e(\"h2\",{children:\"How do Businesses Benefit from Content Moderation Outsourcing?\"}),/*#__PURE__*/e(\"h3\",{children:\"Advantages of user-generated content moderation outsourcing\"}),/*#__PURE__*/e(\"p\",{children:\"Outsourcing services may help your business with the following:\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Save money\"})}),/*#__PURE__*/e(\"p\",{children:\"Hiring third-party specialists for outsourced moderation is much cheaper than establishing and training your own team. Among other things, you will not need to recruit new staff and pay their salaries. Most probably, when creating your moderation team, you will have to set up a new department in the company, which might require expenditures on furniture and equipment. These expenses are multiplied when you work in multiple countries and time zones. Outsourcing provides a turnkey solution for content moderation, significantly lowering your costs.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Save time\"})}),/*#__PURE__*/e(\"p\",{children:\"All the steps you have to complete to establish your in-house team are very time-consuming. Your company will have to spend budget on recruitment, such as training, development and other HR-related costs. However, apart from being expensive, it is also very time-consuming.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Stay focused on your work\"})}),/*#__PURE__*/e(\"p\",{children:\"To achieve desired KPIs your team needs to focus on core business functions. Content moderation, while an integral part of the business, is a low-level task that is not a primary one. Once this part of the work is outsourced, the company has more time for high-priority tasks, improving both operational performance and brand image.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Provide state-of-the-art technology\"})}),/*#__PURE__*/e(\"p\",{children:\"Outsourced content moderation always involves the latest technology and news tools are constantly introduced. Content moderators often combine automation and manual methods of data processing. Therefore, they work faster and more effectively. Outsourcing companies devote a lot of attention to their automated tools, as the result of their work greatly depends on these systems. In doing so, the outsourcing partner assumes all of the expenses for training the staff to properly utilize these technologies.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Benefit from the expertise\"})}),/*#__PURE__*/e(\"p\",{children:\"In addition to the latest technology, outsourced content moderation can also offer a more seamless and quicker project implementation, thanks to years of experience and a skilled team of experts. The outsourcing content moderators will follow the moderation guidelines that you establish, and will also assist in ensuring that the content you publish is aligned with your business values and goals.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Scalability\"})}),/*#__PURE__*/e(\"p\",{children:\"As indicated earlier, user-generated content is getting increasingly abundant. This poses the challenge of scalability. If you outsource content moderation you won't encounter such a challenge, as the outsourcing team is designed to increase or decrease the number of moderators as required.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Ensure the safety of your platform\"})}),/*#__PURE__*/e(\"p\",{children:\"Outsourcing content moderation helps maintain a safe space and respectful user experience so that users can trust your brand. This leads to increased engagement on social media platforms as customers are more likely to engage with content they deem safe or informative.\"}),/*#__PURE__*/e(\"h2\",{children:\"How to identify a reliable provider when you outsource content moderation services?\"}),/*#__PURE__*/e(\"p\",{children:\"Make sure to define the tasks and functions you want to outsource before you start searching for an outsourcing company. The more specific you are about your needs, the easier it will be to find the perfect company with the required expertise. Be sure to pick a content moderation partner that has an in-depth understanding of your industry and audience. It's imperative that they provide scalable solutions that are consistent with your roadmap.\"}),/*#__PURE__*/e(\"p\",{children:\"Make sure to clearly outline the tasks and scope of work you would like to outsource. Also be specific about the type of skills and expertise that you are looking for, project timeframe, QA standards, or other specifications. Check that they utilize cutting-edge technology to guarantee accuracy and ensure efficiency. Besides technical expertise and expertise in the subject matter, verify the outsourcing partner's ability to train and develop their team of content moderators.\"}),/*#__PURE__*/t(\"p\",{children:[\"Another significant factor when selecting an outsourcing partner is its reputation in the market of content moderation service, its credibility, and \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/blog/?blog-category=case%2520studies#blog\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"customer feedback\"})}),\". Consider the company's \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/security/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"data security practices\"})}),\" and compliance with regulatory requirements.\"]}),/*#__PURE__*/e(\"h2\",{children:\"The significance of human contribution in automated content moderation\"}),/*#__PURE__*/t(\"p\",{children:[\"Despite the rapid development of technological advancements, nobody is better than humans at performing content recognition tasks. Therefore, human intervention is required at every stage of AI-assisted \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/moderation/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"content moderation\"})}),\".\"]}),/*#__PURE__*/e(\"p\",{children:\"As previously mentioned, human content moderators can also employ AI automated systems in their work. Some companies fully rely on automated content moderation on their platforms. However, this may be a risky endeavor, as even though AI systems can process a huge amount of content in a relatively short amount of time, there are some perks of manual moderation that artificial intelligent systems are unable to provide:\"}),/*#__PURE__*/t(\"ul\",{style:{\"--framer-font-size\":\"18px\",\"--framer-text-alignment\":\"start\",\"--framer-text-color\":\"rgb(30, 33, 38)\",\"--framer-text-stroke-width\":\"0px\",\"--framer-text-transform\":\"none\"},children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Human moderators can discern the nuances of content usage in different situations;\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Humans recognize cultural differences better;\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Language features and informal registers such as slang or irony are more easily detected by humans.\"})})]}),/*#__PURE__*/e(\"p\",{children:\"Yet, with so much user-generated content in the world today, it is unfeasible to put content moderation completely in the hands of humans. This is the reason automated AI systems play such a big role in extracting explicit content.\"}),/*#__PURE__*/e(\"h2\",{children:\"Conclusion\"}),/*#__PURE__*/e(\"p\",{children:\"User-generated content forms an integral part of today's online business. Users put their trust in this content, yet it has to be moderated to maintain the business's reputation and ensure that the platform is secure and compliant with company policies.\"}),/*#__PURE__*/e(\"p\",{children:\"Outsourcing provides a ready-to-use user content moderation solution, including infrastructure and state-of-the-art automated software. Hiring third-party expertise is much cheaper than building and training your own team. Among other things, you won't need to recruit new staff. The right outsourcing company will deliver a customized solution to meet all your content moderation needs while significantly reducing your costs.\"})]});export const richText2=/*#__PURE__*/t(i.Fragment,{children:[/*#__PURE__*/e(\"p\",{children:\"Lately, large language models (LLMs) have become one of the most popular fields of artificial intelligence. All across the world, these models are headed for design, evaluation, and implementation. Organizations are increasingly pursuing the goal of developing language-understanding software.\"}),/*#__PURE__*/e(\"p\",{children:\"New avenues for studying human language and cognition are made possible by LLMs. As a result of its growth, neural networks are now widely used for natural language processing tasks across a wide range of business domains. Let's take a closer look at what LLMs are and how they work.\"}),/*#__PURE__*/e(\"h2\",{children:\"What is a large language model (LLM)?\"}),/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(n,{href:\"https://toloka.ai/large-language-models\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"Large Language Models\"})}),\" are machine learning models that employ Artificial Neural Networks and large data repositories to power\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/ml/natural-language-processing\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\" Natural Language Processing\"})}),\" (NLP) applications. An LLM serves as a type of AI model designed to be able to grasp, create, and manipulate natural language.\"]}),/*#__PURE__*/e(\"p\",{children:\"These models rely on deep learning technologies, like neural networks, to perform word processing and analyze textual information. Large language models are taught on immense amounts of text training data to learn statistical relationships and representations of language through training, enabling generalization beyond simple memorization.\"}),/*#__PURE__*/e(\"p\",{children:\"Language models are called large because they represent large-scale systems in addition to the massive training dataset. In fact, they are so large that they cannot be launched on a single machine. That's why they are usually accessible through a web interface or API.\"}),/*#__PURE__*/e(\"p\",{children:\"Large language models are suitable for solving multi-faceted tasks practically for any textual query of the user, as opposed to being trained for one specific task. By studying data such as books, documents, and web pages, they can understand the intricacies of any language. Thus, they can devise content and generate text indistinguishable from human writing in multiple languages.\"}),/*#__PURE__*/e(\"p\",{children:\"One of the prevalent large language models at the moment is OpenAI's GPT (Generative Pre-trained Transformer). GPT-3 is one of the world's major artificial intelligence models, and it can perform tasks such as qualitatively generating lengthy articles, replying to various inquiries, and even writing software code.\"}),/*#__PURE__*/e(\"h2\",{children:\"How large language models work\"}),/*#__PURE__*/e(\"p\",{children:\"Large language models consist of a neural network with tens of millions to hundreds of billions of parameters. Usually, the more parameters a neural network has, the better it consolidates its skills and knowledge.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Such neural networks exist within advanced AI assistants, allowing us to communicate with the machine. So, how does a large language model work? The LLMs are introduced to available textual data in the preparation phase to explore the overall structure and rules of the language. The massive datasets are then submitted to a model referred to as a transformer during a training process. Transformer is a type of deep-learning algorithm.\"}),/*#__PURE__*/e(\"p\",{children:\"The core notion of a language model consists of its capacity to predict the next word, called a token, based on the pre-existing text. A family of machine learning architectures called transformer architecture has turned this process into a more human-like one.\"}),/*#__PURE__*/e(\"p\",{children:\"The transformer model mixes the option of model pre-training, parallel data processing, and extensive application of the attention mechanism. Pre-training implies that a machine learning model is first trained on a large set of text data, and then a process of fine-tuning may be applied to it to address a specific problem.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Fine-tuning helps enhance a pre-trained model that already has some knowledge with minor adjustments without having to teach it from the beginning.\"}),/*#__PURE__*/e(\"p\",{children:\"The attention mechanism assigns weights to all input tokens based on their relevance to a specific position in the sequence being processed. The neural network analyzes which position of the incoming sequence is essential for the specific position of the sequence in the output.\"}),/*#__PURE__*/e(\"p\",{children:\"The transformer model, among other things, allows for processing all input text simultaneously or in parallel rather than sequence. This allows transformers to learn from vast amounts of data and significantly reduce training time compared to other methods.\"}),/*#__PURE__*/e(\"p\",{children:\"Transformers can handle several excerpts of long texts at the same time. The model does not neglect the beginning of the text but rather uses what has already been studied, builds better connections between words, and can make sense by understanding the context of a considerable amount of data.\"}),/*#__PURE__*/e(\"h2\",{children:\"How an LLM is built\"}),/*#__PURE__*/e(\"p\",{children:\"Building a large language model (LLM) involves several key stages, each contributing to the creation of a robust and efficient model capable of understanding and generating human-like text. Here\u2019s an overview of the process:\"}),/*#__PURE__*/e(\"ol\",{children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Data Collection and Preparation\",/*#__PURE__*/e(\"br\",{})]}),\"The first step in building an LLM is gathering a large and diverse dataset. These datasets typically contain vast amounts of text from numerous sources, including books, articles, websites, and other forms of written communication. The data must be pre-processed to remove noise, handle incomplete or erroneous information, and structure it in a way that is useful for training the model. The more diverse the dataset, the better the LLM can generalize across different topics and languages.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"2\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Model Architecture Selection\",/*#__PURE__*/e(\"br\",{})]}),\"Once the data is prepared, the next step is to choose the appropriate architecture for the LLM. The transformer model, which is based on attention mechanisms, has become the standard for most modern LLMs. This architecture allows the model to handle large amounts of data in parallel and learn long-range dependencies between words, making it ideal for processing natural language.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"3\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Pre-training\",/*#__PURE__*/e(\"br\",{})]}),\"During pre-training, the LLM learns to predict the next word in a sequence based on the context provided by the previous words. This phase typically involves massive computational power and extensive time, often using GPUs or specialized hardware like TPUs. The goal of pre-training is for the model to develop a general understanding of language patterns, grammar, facts, and relationships between words.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"4\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Fine-tuning\",/*#__PURE__*/e(\"br\",{})]}),\"After pre-training, the LLM undergoes fine-tuning, where it is tailored for specific tasks or industries. Fine-tuning involves providing the model with labeled data and refining its capabilities for particular use cases, such as medical text analysis, legal document processing, or customer support queries. This phase ensures that the model is optimized for its intended purpose and can provide high-quality, relevant results.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"5\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Model evaluation \",/*#__PURE__*/e(\"br\",{})]}),\"Once fine-tuned, the LLM is evaluated using various benchmarks to assess its performance. These evaluations typically measure the model\u2019s accuracy, relevance, and ability to handle edge cases. Based on feedback, further adjustments can be made to the model, including additional fine-tuning or adjustments to the architecture. This iterative process continues until the model meets the required standards for real-world applications.\"]})})}),/*#__PURE__*/e(\"h2\",{children:\"General knowledge vs domain-specific LLMs\"}),/*#__PURE__*/e(\"p\",{children:\"Recent advancements in AI have been unprecedented, with general-purpose large language models (LLMs) like GPT-3 making a significant impact across various fields.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"These models can handle a wide range of tasks, from creating marketing copy to writing emails, and they have been trained on a large amount of data.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Although one of their most significant advantages is their wide range of applications, they aren't always appropriate for highly specialized ones where accuracy and domain-specific knowledge are essential.\"}),/*#__PURE__*/e(\"p\",{children:\"This is where domain-specific LLMs come into play. Unlike general-purpose models, domain-specific models are trained on data from a particular field \u2013 whether it\u2019s healthcare, law, finance, or any other industry \u2013 enabling them to offer deeper insights and more reliable performance in those areas.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"While the versatility of general-purpose LLMs is valuable, specialized models' ability to take on complex, field-specific challenges sets them apart.\"}),/*#__PURE__*/e(\"h3\",{children:\"Limitations of general LLM models\"}),/*#__PURE__*/e(\"p\",{children:\"General-purpose LLMs are trained on highly diverse data, which gives them a broad understanding of human language. They can generate text on almost any topic and answer various questions.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Still, their broad knowledge can also be a limitation when it comes to industries that require specialized knowledge. For instance, in fields like healthcare, legal work, or finance, the complexity and nuances of the data require models that go beyond general understanding and can handle the intricacies of these industries.\"}),/*#__PURE__*/e(\"p\",{children:\"In medical fields, a general-purpose LLM might be able to summarize an academic research paper, but it would lack the ability to diagnose conditions based on patient history or suggest treatments tailored to a person's unique genetic profile.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"With legal matters, a general model could help with general legal questions, but it wouldn't have the expertise to analyze complex case law, for example.\"}),/*#__PURE__*/e(\"h3\",{children:\"The role of domain-specific LLMs\"}),/*#__PURE__*/e(\"p\",{children:\"The domain-specific LLMs aim to get around these restrictions by concentrating on a single area and training on pertinent, superior domain-specific data.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"These models can obtain more accurate results because they comprehend the terminology, context, and main difficulties of the field.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"For example, a healthcare-specific LLM trained in medical research, patient records, and clinical guidelines can help physicians diagnose uncommon conditions, find possible treatments, or even forecast patient outcomes.\"}),/*#__PURE__*/e(\"p\",{children:\"A domain-specific LLM in finance would typically provide far more accurate insights than those offered by a general-purpose model by analyzing market trends, assessing financial reports, and forecasting stock movements.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"This is also true in the legal field, where qualified LLMs can help draft contracts and perform legal research.\"}),/*#__PURE__*/e(\"h3\",{children:\"The importance of specialization\"}),/*#__PURE__*/e(\"p\",{children:\"In all probability, domain-specific LLMs will become the norm in industries like healthcare, law, and finance, especially as AI becomes more widely used for tasks that require in-depth knowledge.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"The combination of highly specialized, field-specific models and robust, broad-based models will fuel ongoing innovation and efficiency across sectors.\"}),/*#__PURE__*/e(\"p\",{children:\"The primary benefit of domain-specific LLMs is their capacity to reliably and accurately manage challenging and high-stakes tasks.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Decisions in fields like healthcare, law, and finance must be made using specialized knowledge. The reason is that errors or discrepancies can have serious repercussions. For instance, a poor financial analysis could result in losses, while in healthcare, an incorrect recommendation could result in inappropriate treatments.\"}),/*#__PURE__*/e(\"p\",{children:\"Specialized LLMs reduce these risks. These models can analyze large volumes of domain-specific data, producing more accurate and well-informed insights than general-purpose models.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Specialized models will become a crucial component of workflows as industries continue to adopt AI, providing not only greater efficiency but also more accurate and dependable decision-making in expert settings.\"}),/*#__PURE__*/e(\"h3\",{children:\"Imagining a hybrid future\"}),/*#__PURE__*/e(\"p\",{children:\"Without a doubt, general-purpose models still have a place in the mix of things \u2013 even though domain-specific models are necessary for some applications.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"The two are likely to coexist in the future of artificial intelligence, becoming more integrated as technology develops.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Without too much imaginative faculty, it is entirely possible to picture a hybrid future in which specialized models manage intricate industry-specific problems, and general models support routine, everyday tasks.\\xa0\"}),/*#__PURE__*/e(\"h2\",{children:\"Training data for large language models\"}),/*#__PURE__*/e(\"p\",{children:\"A large language model demands large amounts of text data for model training, thus ensuring that contextually relevant responses are provided.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"The large language model training process may involve any kind of text data, and there is no need for any sort of labeling of this data beforehand.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"However, human input is needed at this stage to collect and clean datasets.\"}),/*#__PURE__*/e(\"p\",{children:\"The outcome of training is the language model successfully predicting the next word based on information about the words that come before it.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:'Any book, commentary, or essay is already a ready-made training piece of data, as it already consists of a tremendous array of \"word-next word\" type sequences. However, not any text written by a human will do. Diverse text, high-quality, and well-curated datasets are necessary for training.\\xa0'}),/*#__PURE__*/e(\"p\",{children:\"Importantly, it's not just about collecting data\u2014expert human involvement is crucial to verifying that the data is clean, relevant, and balanced.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"AI Tutors and data experts play an indispensable role in building the datasets that feed into the model. Trainers are specialists who are responsible for creating data and evaluating output, ensuring that irrelevant, biased, or harmful content is removed or isn\u2019t used in training.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Their expertise ensures the model learns in a way that improves its ability to provide more accurate and meaningful responses.\"}),/*#__PURE__*/e(\"p\",{children:\"AI Tutors, in particular, are essential when the model is being trained for specific industries. For instance, if an LLM is being fine-tuned for the healthcare or legal sectors, AI Tutors ensure the data includes industry-specific language, terms, and nuances.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Focused curation helps the model respond more effectively to specialized queries and real-world scenarios.\"}),/*#__PURE__*/e(\"p\",{children:\"Additionally, data experts are responsible for balancing the datasets, making sure the model is trained on a diverse range of examples. It's a vital stage of the process that helps prevent overfitting to particular biases or narrow perspectives, ensuring the model can provide equitable responses across different demographics, languages, and contexts.\"}),/*#__PURE__*/e(\"h2\",{children:\"Limitations of large language models and how to bridge them\"}),/*#__PURE__*/e(\"p\",{children:\"While LLMs have made great strides, they are not without limitations.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Understanding these shortcomings is crucial to improving their performance and ensuring that they are used effectively across industries.\"}),/*#__PURE__*/e(\"ol\",{children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Context and long-term coherence issues\",/*#__PURE__*/e(\"br\",{})]}),\"LLMs struggle to maintain coherence over more extended conversations or documents. This happens because they process data in chunks, often losing track of earlier parts of the conversation as the context expands. For instance, a model may provide an accurate response to the immediate query but fail to integrate earlier information, leading to disjointed or repetitive answers.\"]})})}),/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Solution:\"}),' One promising fix lies in refining model architectures. Researchers are exploring \"memory-augmented neural networks\" and \"retrieval-augmented generation\" techniques, which allow LLMs to \"remember\" past interactions and incorporate that context into their responses. These methods are designed to extend the model\\'s ability to recall relevant information over long sequences, improving its overall coherence.']}),/*#__PURE__*/e(\"ol\",{start:\"2\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Bias and fairness concerns\",/*#__PURE__*/e(\"br\",{})]}),\"Another significant limitation is the potential for bias in LLMs. Since these models learn from vast amounts of text data, they may inadvertently replicate harmful stereotypes or reinforce social biases present in the training material. For example, a model might generate biased or discriminatory content based on the demographic skew of its data.\",/*#__PURE__*/e(\"br\",{}),/*#__PURE__*/e(\"br\",{}),/*#__PURE__*/e(\"br\",{className:\"trailing-break\"})]})})}),/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Solution:\"}),' One approach to address this issue is through improved data curation. By diversifying the data used to train models, we can reduce the risk of bias. Additionally, techniques like \"bias correction algorithms\" are being developed to detect and mitigate biases during both the training and inference stages. Another potential solution is the involvement of human moderators or experts who review and adjust model outputs for fairness, ensuring that the responses meet ethical standards.']}),/*#__PURE__*/e(\"ol\",{start:\"3\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Lack of common sense and reasoning\",/*#__PURE__*/e(\"br\",{})]}),\"While LLMs are adept at generating fluent text, they often lack common sense reasoning. They can produce logically sound sentences without necessarily reflecting a deep understanding of the world. This can result in responses that are technically accurate but contextually flawed, such as answering questions based on patterns rather than real-world knowledge.\"]})})}),/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Solution:\"}),' To address this, researchers are incorporating \"commonsense reasoning\" into LLMs. These models are trained on specialized datasets designed to teach them more nuanced, real-world reasoning. One approach is \"knowledge-grounded generation,\" where LLMs are supplemented with external knowledge sources like structured databases or real-time information to provide more contextually appropriate answers.']}),/*#__PURE__*/e(\"ol\",{start:\"4\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Data privacy and security\",/*#__PURE__*/e(\"br\",{})]}),\"As LLMs are trained on vast datasets, often scraped from the web, concerns about privacy and the security of sensitive data arise. LLMs might inadvertently generate responses based on private or confidential information that was included in the training data.\"]})})}),/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Solution:\"}),' Data privacy measures can be integrated by adopting techniques like federated learning, where data is processed locally and only model updates are shared, ensuring that private data does not leave the user\\'s device. Additionally, \"differential privacy\" techniques can be applied to ensure that individual data points are obscured during the model\u2019s training process, preventing the model from memorizing and reproducing sensitive information.\\xa0']}),/*#__PURE__*/e(\"p\",{children:\"While they have not been perfected yet, the ongoing efforts to refine and improve these LLMs hold the potential for groundbreaking advancements in AI. By addressing limitations, we can ensure that LLMs are more accurate, fair, and valuable tools across various applications.\"}),/*#__PURE__*/e(\"p\",{children:\"To better understand how LLMs work in practice, it's helpful to explore real-world applications. The following examples will give you a clearer view of how these models are utilized across different domains.\"}),/*#__PURE__*/e(\"h2\",{children:\"Applications for enhancing productivity\"}),/*#__PURE__*/e(\"p\",{children:\"Large language models can boost a business's productivity in many ways. Their power to interpret human queries and resolve relatively complex problems helps us assign time-consuming routine tasks to chatbots and then simply verify the outcomes.\"}),/*#__PURE__*/e(\"p\",{children:\"Pre-trained transformer models may be designed to quickly fulfill the appropriate goals of your business. They already possess the knowledge required and deeply comprehend the target language, enabling you to focus selectively on fine-tuning the model for any specific tasks you have in mind. Here are some examples:\"}),/*#__PURE__*/e(\"h3\",{children:/*#__PURE__*/e(\"strong\",{children:\"Sales and customer service\"})}),/*#__PURE__*/e(\"p\",{children:\"Chatbots and AI assistants custom-trained for your business deliver high-quality, prompt feedback and support to your customers. The sales bot can interact with your customers, present the range of products, inform them about discounts and promotions, and motivate them to make a purchase.\"}),/*#__PURE__*/e(\"h3\",{children:/*#__PURE__*/e(\"strong\",{children:\"Content generation\\xa0\"})}),/*#__PURE__*/e(\"p\",{children:\"Human-like text generation is one of the key features of language models. AI creates texts that help your products stand out among competitors. Additionally, they can acquire the ability to generate documentation based on your knowledge base, which can significantly speed up your company's document workflow.\"}),/*#__PURE__*/e(\"h3\",{children:/*#__PURE__*/e(\"strong\",{children:\"Content generalization\"})}),/*#__PURE__*/e(\"p\",{children:\"Quite often, especially in large companies, a considerable amount of internal documents and various text materials are piled up in random order. It would be more convenient for employees to interact with them if they were sorted into categories. LMMs are very capable at helping to categorize such documents.\"}),/*#__PURE__*/e(\"h3\",{children:/*#__PURE__*/e(\"strong\",{children:\"Content moderation\"})}),/*#__PURE__*/e(\"p\",{children:\"LLMs assist in identifying spam, profanity, and toxic content on your resources according to the set guidelines. Large models shield users from any type of controversial content that might be considered unsafe or inappropriate, potentially tainting the platform's online reputation.\"}),/*#__PURE__*/e(\"h2\",{children:\"Niche applications for specialized fields\"}),/*#__PURE__*/e(\"p\",{children:\"Large language models (LLMs) are powerful tools for streamlining business processes, but their true potential lies in specialized, expert-driven fields. When fine-tuned to meet specific needs, LLMs provide tailored solutions that enhance efficiency, support decision-making, and drive innovation.\"}),/*#__PURE__*/e(\"p\",{children:\"In this section, we\u2019ll explore how LLMs are being applied in established industries\u2014such as healthcare, law, engineering, and education\u2014and in more unexpected fields like archaeology, aerophysics, and computer design. While the first set of industries are well-established expert domains, the latter highlight the growing potential of LLMs in cutting-edge sectors.\"}),/*#__PURE__*/e(\"p\",{children:\"In these specialized fields, LLMs can handle complex tasks, from interpreting medical records to analyzing legal documents and processing technical data.\\xa0\"}),/*#__PURE__*/e(\"h3\",{children:\"Healthcare\"}),/*#__PURE__*/e(\"p\",{children:\"Artificial intelligence is opening up revolutionary possibilities in the healthcare industry by facilitating a deeper understanding of intricate biological data and improving the precision of diagnoses and treatments. Here\u2019s what\u2019s happening in the exciting field of medicine:\"}),/*#__PURE__*/e(\"h4\",{children:\"Personalised medication\"}),/*#__PURE__*/e(\"p\",{children:\"Healthcare professionals can use AI to analyze genomic sequences to identify genetic variations affecting a patient's treatment response. With personalized medicine, treatment plans are customized based on each patient's distinct genetic profile. This method matches therapies to patients' unique biological needs, ensuring more effective treatments and reducing side effects.\"}),/*#__PURE__*/e(\"h4\",{children:\"Bloodwork diagnostics\"}),/*#__PURE__*/e(\"p\",{children:\"AI improves the interpretation of blood test results by rapidly examining biomarkers to find early indicators of inflammatory and chronic diseases. By quickly processing vast amounts of data, AI enables medical practitioners to detect possible health hazards before they become more serious. This opens the door to earlier interventions and better patient outcomes.\"}),/*#__PURE__*/e(\"h4\",{children:\"Drug development\"}),/*#__PURE__*/e(\"p\",{children:\"AI is accelerating the sometimes expensive and drawn-out drug development process. AI helps researchers find promising drug candidates more quickly by forecasting which molecular compounds are likely to be effective in treating particular diseases. This helps shorten the time it takes to introduce new treatments to the market and speeds up the development of targeted therapies.\"}),/*#__PURE__*/e(\"h4\",{children:\"Epigenetics advancements\"}),/*#__PURE__*/e(\"p\",{children:\"The study of epigenetics, which examines how outside influences can affect gene expression, heavily relies on artificial intelligence. This research is revealing important new information about diseases like cancer, neurodegenerative diseases, and mental health issues. AI assists scientists in understanding the interactions between genetic makeup and environmental factors, opening the door to more effective treatments that address the different facets of disease.\"}),/*#__PURE__*/e(\"h3\",{children:\"Legal\"}),/*#__PURE__*/e(\"p\",{children:\"Although the legal industry is often perceived as being somewhat resistant to change, LLMs are beginning to change the way legal teams work. By automating the prosaic and time-consuming parts of legal practice, they are enhancing the skills of current practitioners rather than replacing them. This allows them to concentrate on the work that really needs human insight. Here are some ways in which AI is making its mark in the field of law:\"}),/*#__PURE__*/e(\"h4\",{children:\"Draughting and reviewing contracts\"}),/*#__PURE__*/e(\"p\",{children:\"Contract creation and review frequently require navigating complex legal jargon and identifying possible problems. By swiftly reading through vast amounts of text, highlighting any inconsistencies, and spotting possible dangers, LLMs can relieve legal teams of some of their workload. They can recommend changes or revisions or even produce boilerplate text according to particular requirements, which makes them especially helpful for standard contracts.\"}),/*#__PURE__*/e(\"h4\",{children:\"Litigation support\"}),/*#__PURE__*/e(\"p\",{children:\"Lawyers deal with a mountain of data when managing cases. LLMs can assist by organizing case files, identifying key points of law, and even summarizing long-winded depositions or trial transcripts. Large language models help ensure that no critical information is overlooked, freeing up attorneys to focus on strategy and client engagement while AI handles the bulk of the details.\"}),/*#__PURE__*/e(\"h4\",{children:\"Regulatory and compliance checks\"}),/*#__PURE__*/e(\"p\",{children:\"The world of law is littered with ever-evolving regulations, from financial compliance to environmental law. Keeping track of these is a full-time job in itself. LLMs step in by scanning regulatory documents and keeping legal teams up-to-date on changes that affect their practice. Whether it\u2019s checking for compliance with new privacy laws or anticipating potential legal risks, these models provide a faster, more reliable way to stay compliant.\"}),/*#__PURE__*/e(\"h4\",{children:\"Managing intellectual property\"}),/*#__PURE__*/e(\"p\",{children:\"In industries such as tech and entertainment, a company's intellectual property (IP) is frequently its most valuable asset. By expediting patent and trademark searches, comparing new filings to existing IP, and spotting possible conflicts or infringements, LLMs play a critical role.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"When AI handles the preliminary steps, legal teams can lead the strategy and litigation components of IP protection. In other words, rather than replacing legal professionals, LLMs are augmenting the capabilities of those already specialists in their independent fields, helping them focus on the work that genuinely requires human insight while automating the repetitive and time-consuming aspects of legal practice.\"}),/*#__PURE__*/e(\"h4\",{children:\"Performing legal research\"}),/*#__PURE__*/e(\"p\",{children:\"Despite being absolutely necessary, legal research can be a daunting undertaking. It could take weeks to manually gather pertinent precedents and important arguments from millions of legal documents, statutes, and case law, but LLMs have the ability to sift through them. Because of its speed and efficiency, legal research is not only completed more quickly but also more thoroughly, giving attorneys access to the most current and reliable information possible.\"}),/*#__PURE__*/e(\"h3\",{children:\"Finance\"}),/*#__PURE__*/e(\"p\",{children:\"The financial sector is leveraging cutting-edge applications of large language models (LLMs) to reimagine traditional workflows and unlock new opportunities. These innovations extend far beyond efficiency, driving insights and strategies previously unattainable with conventional tools.\"}),/*#__PURE__*/e(\"ol\",{children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Synthetic data generation for financial modeling\",/*#__PURE__*/e(\"br\",{})]}),\"LLMs create synthetic datasets that mimic actual financial behaviors without compromising privacy in areas where real-world data is limited or sensitive. These datasets enable institutions to test new models, develop innovative products, and assess market reactions in a controlled yet realistic environment.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"2\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Real-time macroeconomic forecasting\",/*#__PURE__*/e(\"br\",{})]}),\"LLMs are revolutionizing how financial institutions predict macroeconomic trends. These models synthesize insights into actionable forecasts by ingesting unstructured data from global news, regulatory announcements, and market commentary. Their ability to process multilingual inputs also ensures comprehensive coverage of international markets.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"3\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Behavioral finance insights\",/*#__PURE__*/e(\"br\",{})]}),\"Beyond traditional numerical analysis, LLMs analyze behavioral patterns in text-based data, such as investor sentiment from social media or psychological cues in earnings calls. By correlating these patterns with market movements, financial institutions gain a deeper understanding of how emotions and biases drive market behavior.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"4\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Dynamic portfolio optimization\",/*#__PURE__*/e(\"br\",{})]}),\"Traditional portfolio management relies heavily on static models, but LLMs are enabling real-time optimization. These models integrate real-world events like geopolitical shifts or environmental disasters into risk and return assessments, helping investors pivot strategies in rapidly changing markets.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"5\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"RegTech and compliance transformation\",/*#__PURE__*/e(\"br\",{})]}),\"Financial regulations have a way of changing constantly \u2013 making compliance a monumental task. LLMs are transforming RegTech by reading and interpreting new regulations, automatically mapping them to internal compliance frameworks, and flagging potential areas of risk. This reduces the lag between regulation updates and corporate adaptation, enhancing compliance agility.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"6\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Quantum finance and predictive analytics\",/*#__PURE__*/e(\"br\",{})]}),\"In cutting-edge financial research, LLMs are being integrated with quantum computing experiments to explore complex financial systems that exceed classical computing capabilities. These systems aim to refine predictive analytics, uncover hidden correlations, and simulate high-dimensional market behaviors with unprecedented accuracy.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"7\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Tokenized asset markets\",/*#__PURE__*/e(\"br\",{})]}),\"As digital assets and tokenization grow, LLMs are central to understanding and navigating these emerging markets. They analyze white papers, smart contract codes, and market trends to assess the viability and risks of tokenized assets, supporting informed decision-making in decentralized finance (DeFi).\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"8\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Hyper-personalized financial ecosystems\",/*#__PURE__*/e(\"br\",{})]}),\"LLMs are powering platforms that create bespoke financial ecosystems for users, going beyond simple recommendations. These systems dynamically adjust to an individual\u2019s evolving financial landscape, incorporating real-time market data, spending habits, and even life events to deliver hyper-personalized advice and product offerings.\"]})})}),/*#__PURE__*/e(\"ol\",{start:\"9\",children:/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",style:{\"--framer-font-size\":\"11px\",\"--framer-text-color\":\"rgb(0, 0, 0)\",\"--framer-text-decoration\":\"none\"},children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/t(\"strong\",{children:[\"Climate risk assessment in investments\",/*#__PURE__*/e(\"br\",{})]}),\"LLMs are helping investors address one of the most pressing modern challenges: climate change. By analyzing environmental data, carbon disclosures, and sustainability reports, these models identify climate-related risks and opportunities, allowing for more resilient and responsible investment strategies.\"]})})}),/*#__PURE__*/e(\"p\",{children:\"These advancements are literally pushing the boundaries of what\u2019s possible in finance, moving the sector from static, rule-based systems to dynamic, intelligent ecosystems that learn, adapt, and innovate in real time. The integration of LLMs is not just about staying competitive \u2013 it\u2019s about redefining the very fabric of financial strategy.\"}),/*#__PURE__*/e(\"h3\",{children:\"Education\"}),/*#__PURE__*/e(\"p\",{children:\"AI is reshaping education in ways that focus on real, tangible benefits for learners and educators alike. Instead of reimagining the classroom entirely, it is addressing long-standing gaps and enhancing how knowledge is delivered and absorbed.\"}),/*#__PURE__*/e(\"h4\",{children:\"Customized learning for every student\"}),/*#__PURE__*/e(\"p\",{children:\"AI-powered tools tailor educational experiences to individual needs, adapting lessons based on how students respond. Whether it\u2019s accelerating learning for gifted students or providing extra support for those who struggle, this ensures no one is left behind.\"}),/*#__PURE__*/e(\"h4\",{children:\"Overcoming resource gaps\"}),/*#__PURE__*/e(\"p\",{children:\"With AI translating and localizing content into multiple languages, quality education is no longer confined to specific regions or demographics. This democratization of learning resources bridges gaps in access, particularly in underserved communities.\"}),/*#__PURE__*/e(\"h4\",{children:\"Smarter evaluations\"}),/*#__PURE__*/e(\"p\",{children:\"AI doesn\u2019t just grade \u2013 it interprets. By analyzing trends and patterns in a student\u2019s answers, educators can gain deeper insights into learning behaviors, enabling interventions that address the root of challenges rather than just their symptoms.\"}),/*#__PURE__*/e(\"h4\",{children:\"Better teacher support\"}),/*#__PURE__*/e(\"p\",{children:\"Virtual assistants now handle administrative tasks and provide real-time feedback during lessons, freeing teachers to focus on creativity and student interaction. This partnership between AI and educators enhances the human side of teaching, rather than replacing it.\"}),/*#__PURE__*/e(\"h4\",{children:\"Building future-ready skills\"}),/*#__PURE__*/e(\"p\",{children:\"AI literacy is becoming a core part of education. From understanding how algorithms shape the digital world to hands-on experience with machine learning tools, students are being prepared for careers in AI-driven industries.\"}),/*#__PURE__*/e(\"h3\",{children:\"Coding\"}),/*#__PURE__*/t(\"p\",{children:[\"Using LLMs, developers can create code generation models. By\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/blog/bigcode-project\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\" classifying code parts\"})}),\" in popular programming languages, they can train such a model according to a custom dataset and have it perform specialist tasks or functions in dev work.\"]}),/*#__PURE__*/e(\"h3\",{children:\"E-commerce\"}),/*#__PURE__*/e(\"p\",{children:\"It probably doesn\u2019t come as a complete surprise that e-commerce businesses are increasingly leaning into AI. As the digital marketplace grows more crowded and consumer expectations rise, businesses are turning to advanced AI tools to keep pace.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"AI is not just about automation\u2014it\u2019s about delivering deeper insights and optimizing key areas of the shopping experience. Let\u2019s examine how AI is transforming e-commerce through hyper-personalized shopping, dynamic pricing, and enhanced customer support.\"}),/*#__PURE__*/e(\"h4\",{children:\"Hyper-personalized shopping\"}),/*#__PURE__*/e(\"p\",{children:\"AI has unlocked a new era of personalized experiences for online shoppers. By gathering and analyzing data from various sources\u2014such as browsing habits, previous purchases, and even interactions with customer support\u2014AI can suggest products that are specifically aligned with each customer\u2019s preferences.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"This level of personalization isn\u2019t just a nice-to-have feature anymore. Consumers want to feel like the shopping experience was made just for them, and AI makes that possible. Businesses that use AI to tailor product recommendations often see improved conversion rates, as shoppers are presented with items that genuinely interest them. This makes the shopping journey more seamless and customer-centric, driving both sales and brand loyalty.\"}),/*#__PURE__*/e(\"h4\",{children:\"Dynamic pricing\"}),/*#__PURE__*/e(\"p\",{children:\"Staying competitive in e-commerce terms often means adjusting prices in response to market trends, competitor pricing, and demand shifts. AI is helping businesses do this in real time, enabling them to implement dynamic pricing strategies that automatically adjust prices based on various factors. This capability allows businesses to stay flexible, adjust pricing for maximum profitability, and make data-driven decisions.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Whether responding to seasonal fluctuations or taking competitor moves into account, AI-powered pricing systems ensure that businesses remain responsive and efficient, without the need for constant manual intervention. For e-commerce businesses, this can mean the difference between being competitive or losing out to a stronger competitor.\"}),/*#__PURE__*/e(\"h4\",{children:\"Enhanced customer support\"}),/*#__PURE__*/e(\"p\",{children:\"AI-powered customer service tools, such as chatbots and virtual assistants, are increasingly being used to handle everyday customer interactions. By automating common inquiries \u2013 such as order tracking, product details, or return policies \u2013 AI helps businesses deliver faster and more consistent support. This not only reduces wait times but also frees up human agents to focus on more complex issues.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"Additionally, AI systems learn from each interaction, continuously improving their ability to resolve customer inquiries with minimal input. As a result, the customer experience becomes more efficient, allowing businesses to scale their support without sacrificing quality. This improvement in customer service not only enhances customer satisfaction but also increases retention rates.\"}),/*#__PURE__*/e(\"p\",{children:\"AI in e-commerce is about creating experiences that resonate with customers \u2013 an expectation that is growing by the day. By focusing on hyper-personalization, flexible pricing, and efficient support, AI is helping businesses meet the high expectations of today\u2019s online shoppers while driving growth in the process.\"}),/*#__PURE__*/e(\"h3\",{children:\"Engineering\"}),/*#__PURE__*/e(\"p\",{children:\"LLMs are making it easier to handle complex technical documentation and design processes in engineering. By training on vast datasets, LLMs can examine engineering specifications, product manuals, and blueprints with incredible precision. They help engineers to automatically identify discrepancies, suggesting improvements, and even generating technical solutions based on historical data.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"For example, in the design phase, LLMs can process vast amounts of data, suggesting optimization strategies for materials, cost-saving techniques, or energy-efficient systems. When looking at maintenance, LLMs can be used to predict potential equipment failures by analyzing manuals and sensor data, helping companies address issues before they escalate into costly breakdowns.\"}),/*#__PURE__*/e(\"h3\",{children:\"Robotics\"}),/*#__PURE__*/e(\"p\",{children:\"In robotics, LLMs are transforming the way robots understand and interact with their environment. By processing vast amounts of text and sensor data, LLMs help robots improve their decision-making, adapt to new situations, and enhance their learning capabilities. In industrial settings, LLMs assist in automating tasks, such as parts identification or assembly line optimization, by interpreting both structured and unstructured data. These models also help in human-robot collaboration by interpreting and responding to verbal commands in natural language, making the interaction seamless. For autonomous robots, LLMs support mission planning by analyzing real-time data and generating responses that guide robots in navigation or task execution.\"}),/*#__PURE__*/e(\"h3\",{children:\"Architecture\"}),/*#__PURE__*/e(\"p\",{children:\"Architects use LLMs to process complex legal documents and regulations, ensuring compliance with local and international standards. LLMs also support sustainable architecture by analyzing energy usage patterns and suggesting design modifications that reduce the building\u2019s carbon footprint. By integrating these models, architectural firms can streamline the planning process, ensure greater accuracy, and deliver innovative solutions more efficiently.\"}),/*#__PURE__*/e(\"h3\",{children:\"Archaeology\"}),/*#__PURE__*/e(\"p\",{children:\"LLMs are making waves in the field of archaeology by helping researchers process and interpret vast amounts of historical and cultural data. By training on archaeological records, LLMs help archaeologists quickly locate relevant information from excavation reports, artifacts, and historical texts. These models can even predict the potential location of undiscovered sites by analyzing geographical and historical patterns, assisting archaeologists in their exploration. Furthermore, LLMs aid in the cataloging of ancient languages, translating and interpreting texts that were previously difficult to decipher. This opens new doors for understanding lost civilizations and enhancing research efficiency.\"}),/*#__PURE__*/e(\"h3\",{children:\"Astrophysics\"}),/*#__PURE__*/e(\"p\",{children:\"Yes \u2013 even astrophysics. LLMs are being used to decode the massive datasets generated by telescopes and space missions. These models are trained to recognize patterns in astronomical data, identifying new celestial bodies, classifying galaxies, and even predicting the behavior of cosmic phenomena. LLMs also assist in research by summarizing vast amounts of scientific literature and enabling astrophysicists to keep up with the rapidly expanding body of knowledge. For instance, LLMs are used to analyze light curves from distant stars to detect exoplanets or to process gravitational wave data, facilitating deeper insights into the universe\u2019s most profound mysteries.\"}),/*#__PURE__*/e(\"h3\",{children:\"Climate and environmental modeling\"}),/*#__PURE__*/e(\"p\",{children:\"LLMs play an increasingly important role in climate science by analyzing large datasets to predict weather patterns, track climate change, and simulate environmental impacts. These models are used to process data from satellites, weather stations, and climate sensors, helping scientists understand long-term trends and short-term fluctuations in the environment. In the context of policy and mitigation, LLMs help model the potential effects of various environmental regulations, offering governments and organizations data-driven insights into the best courses of action. LLMs also assist in modeling natural disasters, helping to predict events like wildfires, hurricanes, and floods, thus aiding in disaster management and response efforts.\"}),/*#__PURE__*/e(\"h3\",{children:\"Quantum computing design\"}),/*#__PURE__*/e(\"p\",{children:\"Quantum computing is an inherently complex field, but LLMs are making it easier to design quantum algorithms and systems. By processing large datasets and understanding intricate quantum mechanics terminology, LLMs assist in the creation of new quantum computing protocols and error-correction techniques.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"They can analyze research papers, technical documents, and simulation results, providing scientists with insights that speed up the development of quantum software. LLMs are also helping to bridge the gap between quantum hardware engineers and software developers by automatically translating complex quantum programming tasks into human-readable language, streamlining collaboration, and accelerating progress in quantum computing.\"}),/*#__PURE__*/e(\"h2\",{children:\"How to tailor a large language model for your business\"}),/*#__PURE__*/e(\"p\",{children:\"Even though data labeling is not required for pre-training, to handle a specific task well, language models need fine-tuning. If you want to develop your LLM, you will require high-quality labeled data. This calls for a certain amount of labeled data, which is most likely considerably smaller than the data originally employed to train the initial LLM.\"}),/*#__PURE__*/e(\"p\",{children:\"Large language models perform well on generic tasks due to the fact that they are pre-trained on immense quantities of unlabeled text data, such as books, online commentaries, or massive document databases. To create successful applications that perform specific tasks, you require human-labeled data.\"}),/*#__PURE__*/e(\"p\",{children:\"Aside from providing human-labeled data for the language model development process, crowdsourcing platforms such as Toloka enable their users to automate the fine-tuning of the models. This allows them to launch their AI application without the need to hire a team of experts, but rather outsourcing it. The LLMs are laborious in their development and upkeep, putting them out of reach for most companies. A perfect solution that makes LLMs more accessible for any enterprise is addressing a crowd of domain experts through partners like Toloka.\"}),/*#__PURE__*/e(\"h2\",{children:\"The role of people in LLM development\"}),/*#__PURE__*/e(\"p\",{children:\"Humans aren\u2019t quite out of the loop just yet. As previously mentioned, the development of large language models for specific uses requires accurately labelled data in order for the language model to generate accurate estimates. Computers cannot produce labels as fine and sophisticated as people can.\"}),/*#__PURE__*/e(\"p\",{children:\"In order to align the model with expectations, human evaluation of the model's performance is crucial. The input data used for human labeling is consistent and derived from actual situations.\\xa0\"}),/*#__PURE__*/e(\"p\",{children:\"From gathering and labeling the required data to \u2018humanly\u2019 assessing the model's quality and moderating its output, the Toloka platform allows human input into every step of the development of your LLM.\"})]});export const richText3=/*#__PURE__*/t(i.Fragment,{children:[/*#__PURE__*/e(\"h2\",{children:\"Introduction\"}),/*#__PURE__*/t(\"p\",{children:[\"As the world moves toward rapid digitalization, there\u2019s an ever-growing need for documents and data to be converted into online formats. Not only does this help with office space and storage, but digital documents are also more secure. (Optical character recognition)[\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/image-data/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"https://toloka.ai/image-data/\"})}),\"] (OCR) is a key part of document digitization and a valuable technology that\u2019s being swiftly adopted by companies across industries today. Due to its automated data extraction and storage capabilities, OCR can save you time, money, and resources.\"]}),/*#__PURE__*/e(\"p\",{children:\"From letters written by hand to printed papers, you can learn how to use OCR to convert any type of image-based document that comprises written copy into machine-readable text data. We\u2019ve outlined all you need to know about this technology: how it works, what the key benefits are, and how you can use crowdsourcing to train and fine-tune a machine learning model by incorporating human input.\"}),/*#__PURE__*/e(\"h2\",{children:\"OCR: What is it and how does it work?\"}),/*#__PURE__*/t(\"p\",{children:[\"Let\u2019s start off by defining what (optical character recognition)[\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/image-data/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"https://toloka.ai/image-data/\"})}),\"] is. Occasionally mentioned as text recognition, OCR involves turning an image that contains text into a format that a machine can read. OCR applications mine data from documents, photos, and image-only PDFs. An example of this is when you scan a document and the computer saves the scan as an image file. In short, OCR helps or aids in the process of changing that image file into a text document, the content of which is then stored as text data. We\u2019ll cover document text extraction and text recognition, which both fall under the umbrella of optical character recognition.\"]}),/*#__PURE__*/e(\"p\",{children:\"What\u2019s beneficial about optical character recognition is that it can extract and reuse data from scans, images, and image-only PDFs (as mentioned above). The possibilities are infinite: for example, large volumes of historic or legal documents can now be converted into PDFs that people can edit, format, or search for specific information in, just like in an editable document. In fact, OCR is mostly used to acquire information from printed publications and manage different types of paperwork, including many of the materials found in a typical office setting, from printed forms and invoices, to contracts, proposals, transaction records, and more. While most companies have gone paperless, there are still some drawbacks to having all these documents as scanned images, since word processing programs can\u2019t process text in images. That\u2019s where optical character recognition comes in handy: it changes the image into text data, which has a wide variety of uses, such as performance and operations management, as well as data analytics and process automation. You can see how convenient this would be in the day-to-day workings of so many businesses and the everyday life of so many people.\"}),/*#__PURE__*/e(\"p\",{children:\"So how does it do this? Optical character recognition technology can identify letters in an image and arrange them into words. It can then form these words into sentences. You don\u2019t need to enter data manually and it even lets you access and edit the original material.\"}),/*#__PURE__*/e(\"p\",{children:\"As mentioned, optical character recognition systems can transform actual paper documents into machine-readable text. An optical scanner or customized circuit board makes up the hardware part of a system that copies or reads the text. The software part takes care of any advanced processing and leverages AI to incorporate more sophisticated intelligent character recognition (ICR) techniques, such as handwriting style or language recognition.\"}),/*#__PURE__*/e(\"h2\",{children:\"Traditional versus deep learning approaches to OCR\"}),/*#__PURE__*/t(\"p\",{children:[\"Generally, (optical character recognition)[\",/*#__PURE__*/e(n,{href:\"https://toloka.ai/image-data/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"https://toloka.ai/image-data/\"})}),\"] approaches are founded on either traditional image processing (based on machine learning), or deep learning techniques. The former involves multiple pre-processing steps where the document is cleaned up and de-noised, followed by contour detection to identify lines and columns via character extraction, identification, and segmentation through multiple machine learning algorithms. This technology involves several steps:\"]}),/*#__PURE__*/e(\"h3\",{children:\"1. Image acquisition\"}),/*#__PURE__*/e(\"p\",{children:\"OCR software analyzes scanned images and distinguishes the light from dark areas as background versus text. This is necessary to uncover alphabetical or numerical digits by processing the dark sections.\"}),/*#__PURE__*/e(\"h3\",{children:\"2. Preliminary processing\"}),/*#__PURE__*/e(\"p\",{children:\"Next, the OCR software cleans the image by straightening and smoothing out the document, thereby fixing any contrast issues and ridding it of any other defects.\"}),/*#__PURE__*/e(\"h3\",{children:\"3. Text recognition\"}),/*#__PURE__*/e(\"p\",{children:\"OCR software implements pattern matching and feature extraction to recognize text, generally focusing on a single character, word, or section of text at a time.\"}),/*#__PURE__*/e(\"h3\",{children:\"4. Pattern recognition (pattern matching)\"}),/*#__PURE__*/e(\"p\",{children:\"Pattern matching involves drawing out an image of a character (known as a \u201Cglyph\u201D) and comparing it to a similar character, but this will only work if the font and scale are the same. By feeding the OCR application with samples of text in different formats and fonts, you can use pattern recognition to evaluate characters.\"}),/*#__PURE__*/e(\"h3\",{children:\"5. Feature recognition (feature extraction)\"}),/*#__PURE__*/e(\"p\",{children:\"This algorithm breaks down glyphs into different geometric features like lines and their directions and intersections, and uses them to determine the best match among stored glyphs. The OCR application applies rules regarding the characteristics (such as curved, crossed, or angled lines) of a certain letter or number to identify characters in the scanned document.\"}),/*#__PURE__*/e(\"h3\",{children:\"6. Finishing\"}),/*#__PURE__*/e(\"p\",{children:\"The OCR software breaks a page down by section including text blocks, tables, and graphics. Words are distinguished from lines to form characters and the algorithm then compares them to a collection of pattern images. The text data that\u2019s been extracted is then turned into a editable file.\"}),/*#__PURE__*/e(\"p\",{children:\"Image binarization, style classification, and refinement are also steps of the process. A binary image can be helpful in terms of reducing the amount of time required to extract a section of an image. Likewise, font identification can help determine if the text is written by hand, and by whom. You can further improve an OCR\u2019s output by using a glossary to transform illogical words into their closest related versions that are likely to be correct.\"}),/*#__PURE__*/e(\"p\",{children:\"Although traditional methods are easier to build, they\u2019re more time consuming and can be quickly outpaced by deep learning approaches. Traditional approaches are excellent for printed and handwritten materials, but they\u2019re not as effective when it comes to more complex datasets.\"}),/*#__PURE__*/e(\"p\",{children:\"By contrast, OCR methods that involve deep learning can extract a multitude of features simultaneously. Algorithms that are based on both computer vision and NLP have proven to be highly successful in producing results. Plus, they don\u2019t have to follow the same pre-processing steps that traditional methods do.\"}),/*#__PURE__*/e(\"p\",{children:\"Furthermore, these deep learning OCR methods can extract text-based sections and predict bounding box coordinates. These are then passed onto language processing algorithms that use RNNs, LSTMs, and transformers to decipher and convert the information from the features into text data.\"}),/*#__PURE__*/e(\"p\",{children:\"Unlike traditional methods, deep learning OCR algorithms follow two main steps:\"}),/*#__PURE__*/e(\"h3\",{children:\"1. Regional proposal\"}),/*#__PURE__*/e(\"p\",{children:\"This step involves detecting areas of text within an image. Convolutional models detect parts of text and encircle them within bounding boxes. This is similar to object detection where certain regions are marked, withdrawn, and leveraged as attention maps. Then they\u2019re fed to language processing algorithms along with the features that have been pulled out of the image.\"}),/*#__PURE__*/e(\"h3\",{children:\"2. Language processing\"}),/*#__PURE__*/e(\"p\",{children:\"In this second stage, NLP networks extract the data in these regions and create cohesive sentences from the features fed from Convolutional Neural Network (CNN) layers. As a side note, a lot of research has been done on CNN algorithms that can identify characters directly on practical use cases such as detecting text that has minimal temporal information, such as automobile registration plates.\"}),/*#__PURE__*/e(\"p\",{children:\"Moreover, deep learning helps to alleviate poor quality scans. Google Earth leveraging OCR to identify street addresses or paper documents being digitized and mined for data are a couple relevant examples of how deep learning is applied to unstructured text scanning.\"}),/*#__PURE__*/e(\"h2\",{children:\"Ways to improve current OCR methods for image recognition\"}),/*#__PURE__*/e(\"p\",{children:\"Text detection algorithms are key to OCR methods used today, and neural networks are adept at identifying text within documents and images at all angles. While current OCR methods have a number of impressive capabilities, there\u2019s always room for improvement.\"}),/*#__PURE__*/e(\"p\",{children:\"You can achieve greater OCR accuracy via the following approaches:\"}),/*#__PURE__*/e(\"h3\",{children:\"1. Denoise input data\"}),/*#__PURE__*/e(\"p\",{children:\"In order to preclude non-text regions from being suggested as text, data fed to the model should be fully denoised. Gaussian blurring is one of the best methods for this. Additionally, white noise can also be eliminated with an autoencoder.\"}),/*#__PURE__*/e(\"h3\",{children:\"2. Enhanced image contrast\"}),/*#__PURE__*/e(\"p\",{children:\"Contrast in images is integral to aiding neural networks in distinguishing sections of text from background areas. Boosting the contrast between the text and the background improves the performance of optical character recognition models.\"}),/*#__PURE__*/e(\"h2\",{children:\"What kinds of OCR technologies exist for scanned documents?\"}),/*#__PURE__*/e(\"p\",{children:\"There are several different types of OCR systems that are categorized based on how they\u2019re used. For example, Intelligent Character Recognition (ICR) and Intelligent Word Recognition (IWR) are more sophisticated subgroups of general OCR systems. These subgroups focus on handwritten rather than printed text.\"}),/*#__PURE__*/e(\"p\",{children:\"Here are some of the main types:\"}),/*#__PURE__*/e(\"h3\",{children:\"Simple Optical Character Recognition Programs\"}),/*#__PURE__*/e(\"p\",{children:\"These programs leverage stored font patterns and text-based images as templates. With the help of pattern matching algorithms, the OCR software compares images of text glyph by glyph with those stored in an internal database. This forms the basis of optical word recognition, where text is then matched word by word. Still, there are limitations as it\u2019s impossible for the database to account for and store all the various fonts and handwriting styles that exist.\"}),/*#__PURE__*/e(\"h3\",{children:\"Intelligent Character Recognition Programs\"}),/*#__PURE__*/e(\"p\",{children:\"Modern OCR systems implement human reading skills and strategies through intelligent character recognition. The neural network analyzes text via curves, lines, intersections, and loops through different levels by repeatedly processing images. The ICR-processed results are provided within seconds.\"}),/*#__PURE__*/e(\"p\",{children:\"Furthermore, ICR software can divide symbols into the elements listed above in order to detect individual handwritten characters (not cursive). Moreover, ICR tools identify highly structured characters that are evenly arranged. For example, those found in a questionnaire, such as an exam or a medical survey, where someone fills out answers in the respective blank fields or boxes.\"}),/*#__PURE__*/e(\"h3\",{children:\"Intelligent Word Recognition\"}),/*#__PURE__*/e(\"p\",{children:\"Likewise, intelligent word recognition utilizes the same principles as ICR; however, IWR views entire words without extracting the characters found in an image. IWR is applied to unstructured, freehand, or cursive handwriting. The goal is to identify the whole word rather than separate characters. Free form handwritten notes are a good example of where IWR is used: entire words or phrases are identified.\"}),/*#__PURE__*/e(\"p\",{children:\"Recognizing and decoding handwritten text can be difficult, but IWR aims to address this challenge. With IWR, there are significantly fewer mistakes because it matches handwritten words to a user-defined glossary.\"}),/*#__PURE__*/e(\"p\",{children:\"Many of today's applications combine all three approaches: IWR, ICR, and OCR. With the help of this technology, you will be able to easily note logos, watermarks, and other designations in documents\"}),/*#__PURE__*/e(\"h2\",{children:\"Advantages of OCR and image processing for text recognition\"}),/*#__PURE__*/e(\"p\",{children:\"As previously alluded to, OCR technology can be utilized differently based on specific data scientists need, its use, and application. The main benefit is that it can simplify the process of data entry via search, editing, and storage capabilities. With the help of OCR, individuals and businesses alike can store and access files on their devices that otherwise would take up a ton of office space. There\u2019s no doubt that businesses will continue their rapid adoption of this technology for a variety of reasons.\"}),/*#__PURE__*/e(\"p\",{children:\"The key benefits of using OCR technology are outlined below:\"}),/*#__PURE__*/e(\"h3\",{children:\"Improved accessibility and searchability\"}),/*#__PURE__*/e(\"p\",{children:\"OCR-scanned documents are simple to index by their content, title, or key words, which makes them easy to find in a larger database.\"}),/*#__PURE__*/e(\"h3\",{children:\"Greater productivity \u2014 and no more manual data entry\"}),/*#__PURE__*/e(\"p\",{children:\"OCR software makes life easier, automating manual workflows by scanning, reviewing, and analyzing paper documents, as well as performing a quick search of the database and transforming handwritten notes into editable texts. OCR can rapidly identify data directly from document images, thereby ridding companies of the need for manual data entry. There are also fewer errors this way.\"}),/*#__PURE__*/e(\"h3\",{children:\"More AI-based solutions\"}),/*#__PURE__*/e(\"p\",{children:\"OCR plays a key role in many AI solutions, such as identifying road signs for autonomous vehicles, or recognizing brand logos or product packaging in social media posts or advertisements. A key benefit to these AI solutions is that they help companies make better decisions at a faster rate and at a lower cost, which culminates in a better customer experience.\"}),/*#__PURE__*/e(\"h3\",{children:\"More storage space\"}),/*#__PURE__*/e(\"p\",{children:\"By turning paper documents into digital copies, OCR can help your company increase storage space. Documents stored in text form are significantly smaller than those stored as paper copies or as images.\"}),/*#__PURE__*/e(\"p\",{children:\"Other advantages of OCR include lower costs, improved manual workflows, automated routing of documents and content processing, greater data security, and better overall service.\"}),/*#__PURE__*/e(\"h2\",{children:\"Uses and applications of OCR\"}),/*#__PURE__*/e(\"p\",{children:\"The most well-known OCR use case involves optimizing big data modeling by transforming paper and scanned image documents into searchable PDFs that can be read by a machine. In other words, it involves being able to digitally edit a scanned paper document that has gone through OCR processing. Think how easy it would make things to be able to edit an old legal document in Word or Google Docs.\"}),/*#__PURE__*/e(\"p\",{children:\"With such tremendous potential for efficiency, searchability, and improved workflows, it\u2019s no wonder that many industries are looking to incorporate OCR programs into their businesses.\"}),/*#__PURE__*/e(\"p\",{children:\"Here are just a few examples of how OCR technology is being used:\"}),/*#__PURE__*/e(\"h3\",{children:\"Banking\"}),/*#__PURE__*/e(\"p\",{children:\"Banks, professional services firms, and other financial institutions are using OCR in their day-to-day transactions to process and digitize millions of client documents per year. Not only does OCR make tasks like depositing checks or moving money more efficient, it also helps safeguard against fraud and allows for faster storage and retrieval of vital information.\"}),/*#__PURE__*/e(\"h3\",{children:\"Healthcare\"}),/*#__PURE__*/e(\"p\",{children:\"Processing patient tests, insurance payments, and hospital records are just some of the ways in which OCR is helping to streamline the healthcare industry.\"}),/*#__PURE__*/e(\"h3\",{children:\"Logistics\"}),/*#__PURE__*/e(\"p\",{children:\"Companies that focus heavily on logistics such as mail or transportation services apply OCR technology to their workflows to help track packages, invoices, receipts, and more.\"}),/*#__PURE__*/e(\"h3\",{children:\"Document identification\"}),/*#__PURE__*/e(\"p\",{children:\"Text detected via OCR can be used to categorize documents into groups, which makes it about a hundred times easier to access them.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Data entry automation\"})}),/*#__PURE__*/e(\"p\",{children:\"Given that data can be effectively retrieved from documents with the help of OCR, manual data entry has been relegated to the past. It\u2019s also faster, less expensive, and more accurate.\"}),/*#__PURE__*/e(\"h3\",{children:\"Digital libraries and archives\"}),/*#__PURE__*/e(\"p\",{children:\"OCR can help classify volumes of works by class or genre, as well as digitize archives. Essentially, OCR makes it easier to look up books in different categories and preserve old documents.\"}),/*#__PURE__*/e(\"h3\",{children:\"Text translation\"}),/*#__PURE__*/e(\"p\",{children:\"Particularly relevant to text recognition, text translation can help visitors in foreign countries understand signs, street markers, and billboards in different languages. Translation modules are added onto OCR system output to achieve this.\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Sheet music recognition\"})}),/*#__PURE__*/e(\"p\",{children:\"OCR systems can be trained to identify musical notes, whereby a machine can learn to play music right from the text data. Imagine your music teacher being a machine!\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Marketing campaigns\"})}),/*#__PURE__*/e(\"p\",{children:\"OCR systems can also be used in marketing campaigns for fast-moving consumer goods. This is done by appending a scannable section of text onto a product, which can then be transformed into a text-based code to redeem promo codes.\"}),/*#__PURE__*/e(\"p\",{children:\"As you can see, this amazing technology powers many aspects of daily life. For example, it can be used to index personal documents such as passports, bank statements, business cards, and more, or it can be used to help those with visual impairments.\"}),/*#__PURE__*/e(\"p\",{children:\"With the help of computer vision, OCR detects and reads text in images, which allows Natural Language Processing algorithms to decode the text, convey the meaning, and even translate it into different languages. Incredible advances have been made in this technology. Not only can OCR detect text from images, but it can also identify, translate, and interpret product names, road signs, and billboards.\"}),/*#__PURE__*/e(\"h2\",{children:\"How crowdsourcing can help\"}),/*#__PURE__*/t(\"p\",{children:[\"Even though \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/image-data/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"OCR technology\"})}),\" has progressed by leaps and bounds, it\u2019s still helpful to have humans involved in the process. Crowdsourcing can be a highly effective way to help you digitize your paper files. \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/data-labeling-platform/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"Decomposition of tasks\"})}),\" and \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/technologies/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"robust functionality\"})}),\"incorporated in Toloka, can save you time, energy, and hassle.\"]}),/*#__PURE__*/e(\"h3\",{children:\"1. Break the task down into manageable pieces and classify content\"}),/*#__PURE__*/e(\"p\",{children:\"If you have a set of documents that you\u2019re working with, you\u2019ll first need to decompose the content into different categories that aren\u2019t text: drawings, musical notes, or others. Categorize all your documents into text and non-text components to get a head start on your pipeline. Toloka can help you achieve this through channels such as image classification.\"}),/*#__PURE__*/e(\"h3\",{children:\"2. Allocate the content into different sections\"}),/*#__PURE__*/e(\"p\",{children:\"Then, divide your documents into smaller sections. Longer tasks such as decoding handwriting are more fatigue inducing, so you\u2019ll want to help out your crowd contributors as much as possible here.\"}),/*#__PURE__*/e(\"h3\",{children:\"3. Break images down into smaller pieces\"}),/*#__PURE__*/e(\"p\",{children:\"Divide the images you have into smaller segments by outlining paragraphs. Use Toloka\u2019s \u201CSelecting a region in an image\u201D template for the best results.\"}),/*#__PURE__*/e(\"h3\",{children:\"4. Verify segmentation\"}),/*#__PURE__*/e(\"p\",{children:\"Here you need to ensure that your images that contain text have been properly segmented. You can launch a simple binary classification project and ask Toloka\u2019s annotators to check each other\u2019s work.\"}),/*#__PURE__*/e(\"h3\",{children:\"5. Transcribe text from images\"}),/*#__PURE__*/e(\"p\",{children:\"Now you can transcribe text from your images. Apply our \u201CText recognition from an image\u201D template to get started.\"}),/*#__PURE__*/e(\"h3\",{children:\"6. Incorporate quality controls\"}),/*#__PURE__*/t(\"p\",{children:[\"Enact restrictions on quick responses, introduce \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/knowledgebase/quality-control/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"control tasks\"})}),\", and assess the overlap of Tolokers\u2019 answers to make sure they\u2019re all on the same page.\"]}),/*#__PURE__*/e(\"p\",{children:\"With the help of our crowdsourcing platform, here\u2019s what your pipeline should resemble:\"}),/*#__PURE__*/e(\"img\",{alt:\"Incorporate quality controls\",className:\"framer-image\",height:\"490\",src:\"https://framerusercontent.com/images/i5m4yCs9qRAO8ZjNUnKNHYtpfaA.jpeg\",srcSet:\"https://framerusercontent.com/images/i5m4yCs9qRAO8ZjNUnKNHYtpfaA.jpeg?scale-down-to=512 512w,https://framerusercontent.com/images/i5m4yCs9qRAO8ZjNUnKNHYtpfaA.jpeg?scale-down-to=1024 1024w,https://framerusercontent.com/images/i5m4yCs9qRAO8ZjNUnKNHYtpfaA.jpeg 1926w\",style:{aspectRatio:\"1926 / 980\"},width:\"963\"}),/*#__PURE__*/e(\"h2\",{children:\"Key takeaways\"}),/*#__PURE__*/e(\"p\",{children:\"To sum up, OCR uses a scanner to process paper documents, which it then converts into a black-and-white version where light and dark areas are deciphered from one another as background and text, respectively. Letters and numbers are extracted from the dark areas where a single character, word, or section of text is targeted in succession. At this stage, either pattern or feature recognition is used to identify characters.\"}),/*#__PURE__*/e(\"p\",{children:\"Remember, pattern image recognition occurs when different fonts or formats are involved in the examples fed to an OCR program, whereas feature detection, which comprises angled or crossed lines or curves in a character, is used when the features of a letter or number are analyzed to identify characters in a document. In addition, an OCR program looks at the structure of a document and divides it into sections. Lines are separated into words and subsequently characters, which are then compared with a set of pattern images. After all that, voila! You get the recognized text.\"}),/*#__PURE__*/e(\"p\",{children:\"Thanks to OCR text recognition technology, scanned documents can now be incorporated into big data systems that can read any printed document from bank statements and legal contracts to license plates and health records. Instead of agonizing over innumerable image documents, companies can now use their time more effectively thanks to OCR software, which can extract all the necessary data for them.\"}),/*#__PURE__*/e(\"p\",{children:\"OCR plays an integral part in many companies\u2019 journeys toward digital transformation by helping securely store their data and recover information with greater ease. For example, marketing firms make use of OCR algorithms to improve retention rates and increase sales via a streamlined customer experience.\"}),/*#__PURE__*/e(\"p\",{children:\"Outside of the world of business, OCR also lowers environmental impact by reducing the number of hard copies and saving paper. It likewise improves access to information and bridging language gaps by helping translate written text into different languages.\"}),/*#__PURE__*/t(\"p\",{children:[\"If you\u2019d like to learn more about how Toloka can help you with your next project, we invite you to browse through our \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/blog/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"blog\"})}),\" for more applicable tips, tricks, and insights.\"]})]});export const richText4=/*#__PURE__*/t(i.Fragment,{children:[/*#__PURE__*/e(\"h2\",{children:\"About the client\"}),/*#__PURE__*/e(\"p\",{children:/*#__PURE__*/e(\"strong\",{children:\"Our client, a moderation service, provides moderation solutions for various industries (marketplaces, geoservices, foodtech, and others) by relying mostly on automated moderation.\"})}),/*#__PURE__*/e(\"p\",{children:\"With an average of 5-15 million daily verdicts and automated responses for approximately 72-96% of cases, the moderation service plays a vital role in making online spaces safer and more secure for end users. Automated moderation includes: \"}),/*#__PURE__*/t(\"ul\",{style:{\"--framer-font-size\":\"18px\",\"--framer-text-alignment\":\"start\",\"--framer-text-color\":\"rgb(30, 33, 38)\",\"--framer-text-stroke-width\":\"0px\",\"--framer-text-transform\":\"none\"},children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Performing real-time moderation for a limited set of verdicts\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Sharing anti-fraud verdicts for spam detection among users\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Building metrics for moderation speed, moderation quality, and percentage of unwanted content detected\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Training custom classifiers on an existing data stream to speed up moderation and cover more cases\"})})]}),/*#__PURE__*/e(\"p\",{children:\"The automated solution checks user-generated content for potentially harmful topics in text, images, and links. These include:\"}),/*#__PURE__*/t(\"ol\",{children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Text\"}),\" \u2014 profanity, insults, offensiveness, adult content, spam bot advertising, personal data, racism, extremism, toxicity, illegal activity\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Images\"}),\" \u2014 adult content, violence, profanity, extremism\"]})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/t(\"p\",{children:[/*#__PURE__*/e(\"strong\",{children:\"Links\"}),\" \u2014 malware, viruses, fraud, phishing, anti-spam bans, other security threats \"]})})]}),/*#__PURE__*/e(\"p\",{children:\"The moderation system provides a set of low-level verdicts for these topics and rules for how to apply the verdicts by banning or allowing the content.\"}),/*#__PURE__*/e(\"h2\",{children:\"Challenge\"}),/*#__PURE__*/e(\"p\",{children:\"Some content is too complex for automated methods to handle successfully. Up to 30% of tasks are labeled by ML models with a low confidence level and need more accurate labeling that humans can provide (average wait time of 5-15 minutes).\"}),/*#__PURE__*/e(\"p\",{children:\"The client needed help from Toloka to:\"}),/*#__PURE__*/t(\"ul\",{style:{\"--framer-font-size\":\"18px\",\"--framer-text-alignment\":\"start\",\"--framer-text-color\":\"rgb(30, 33, 38)\",\"--framer-text-stroke-width\":\"0px\",\"--framer-text-transform\":\"none\"},children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Improve output quality\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Minimize moderation costs\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Define and differentiate levels of violations\"})})]}),/*#__PURE__*/e(\"p\",{children:\"The client was looking to improve key business metrics for their moderation system: price per label, accuracy, recall, and labeling speed.\"}),/*#__PURE__*/e(\"h2\",{children:\"Solution\"}),/*#__PURE__*/e(\"p\",{children:\"Tolokers manually verified ~5-30% of data sent to the service for moderation, providing round-the-clock human input and fast response times. They assigned verdicts for text, images, and links, so that the client could determine whether to allow or block the content and potentially ban the user.\"}),/*#__PURE__*/e(\"h2\",{children:\"Results\"}),/*#__PURE__*/e(\"p\",{children:\"By using Toloka, the client reduced the average price per label 20 times compared to the alternative solution using in-house annotators.\"}),/*#__PURE__*/e(\"p\",{children:\"With a focus on quality, speed, and efficiency, Toloka boosted the client\u2019s metrics across the board, including the average accuracy and recall for projects related to text moderation.\"}),/*#__PURE__*/e(\"p\",{children:\"Here\u2019s how metrics improved:\"}),/*#__PURE__*/t(\"ul\",{style:{\"--framer-font-size\":\"18px\",\"--framer-text-alignment\":\"start\",\"--framer-text-color\":\"rgb(30, 33, 38)\",\"--framer-text-stroke-width\":\"0px\",\"--framer-text-transform\":\"none\"},children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Average time for manual moderation: ~10-15 minutes for cases that could not be confidently assigned verdicts\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Average accuracy for projects related to text moderation: 91.5% (Toloka) vs. 74.8% (alternative solution)\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Average recall for projects related to text moderation: 94.6% (Toloka) vs. 83.1% (overall)\"})})]})]});export const richText5=/*#__PURE__*/t(i.Fragment,{children:[/*#__PURE__*/e(\"h2\",{children:\"About the client\"}),/*#__PURE__*/e(\"p\",{children:\"Machine translation services have made significant progress in recent years, becoming more accurate and truer to content, context, and meaning with each new advancement. \"}),/*#__PURE__*/e(\"p\",{children:\"Our client provides automatic translations in 100 supported languages to over 60 million monthly users. The service automatically translates words, expressions, and text across images, videos, websites, and mobile apps using machine translation technology.\"}),/*#__PURE__*/e(\"h2\",{children:\"The challenge\"}),/*#__PURE__*/e(\"p\",{children:\"To enhance translation quality and identify areas for improvement, the team routinely pushes new models into production and compares the results with current models and competitor output.\"}),/*#__PURE__*/e(\"p\",{children:\"The client\u2019s other key business goals include:\"}),/*#__PURE__*/t(\"ul\",{style:{\"--framer-font-size\":\"18px\",\"--framer-text-alignment\":\"start\",\"--framer-text-color\":\"rgb(30, 33, 38)\",\"--framer-text-stroke-width\":\"0px\",\"--framer-text-transform\":\"none\"},children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"User retention and engagement growth\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Revenue growth from B2B sales\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Growth of other services (for example, in-browser translation)\"})})]}),/*#__PURE__*/e(\"p\",{children:\"Humans are still more fluent than machines at evaluating translation quality. That's why human labeling is essential for comparing machine translation output and finding and fixing translation errors.\"}),/*#__PURE__*/e(\"h2\",{children:\"Solution\"}),/*#__PURE__*/t(\"p\",{children:[\"Toloka provided the client with \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/text-data/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"quality assessments\"})}),\" carried out and verified by humans rather than machines. In short, the crowd was asked to compare which translation was better: 1) the current one in production, 2) the one that was newly created, or 3) the competitor\u2019s version. They were also tasked with finding and correcting any errors.\"]}),/*#__PURE__*/e(\"p\",{children:\"Tolokers were given two text units (either a word, phrase, sentence, or paragraph) with the aim of:\"}),/*#__PURE__*/t(\"ul\",{style:{\"--framer-font-size\":\"18px\",\"--framer-text-alignment\":\"start\",\"--framer-text-color\":\"rgb(30, 33, 38)\",\"--framer-text-stroke-width\":\"0px\",\"--framer-text-transform\":\"none\"},children:[/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Comparing two translations to choose the best one\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Rating translation quality on a scale of 0 to 100\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Detecting errors in translations\"})}),/*#__PURE__*/e(\"li\",{\"data-preset-tag\":\"p\",children:/*#__PURE__*/e(\"p\",{children:\"Validating sentence segmentation after translation\"})})]}),/*#__PURE__*/e(\"h2\",{children:\"Results\"}),/*#__PURE__*/e(\"p\",{children:\"The numbers improved across the board, including the retention of weekly and monthly active users. Other internal metrics based on labeling results also improved with human input from the Toloka crowd.\"}),/*#__PURE__*/e(\"p\",{children:\"Toloka\u2019s solution was cost-efficient \u2014 the average price per label for pairwise quality assessment was five times lower than the alternative labeling solution used by the client (~$0.02 vs. $0.1).\"}),/*#__PURE__*/e(\"p\",{children:\"Tolokers spent the same amount of time on each label as the client\u2019s internal annotators (about 1.3 minutes per label), but the average project was completed much faster in Toloka because of the large crowd of skilled Tolokers available to evaluate translations.\"}),/*#__PURE__*/t(\"p\",{children:[\"Toloka\u2019s \",/*#__PURE__*/e(n,{href:\"https://toloka.ai/text-data/\",motionChild:!0,nodeId:\"eo4RAmtig\",openInNewTab:!1,scopeId:\"contentManagement\",smoothScroll:!1,children:/*#__PURE__*/e(a.a,{children:\"fast and reliable solution\"})}),\" helped support ongoing improvements to the accuracy of machine translation models to give the client a competitive edge.\"]})]});\nexport const __FramerMetadata__ = {\"exports\":{\"richText\":{\"type\":\"variable\",\"annotations\":{\"framerContractVersion\":\"1\"}},\"richText5\":{\"type\":\"variable\",\"annotations\":{\"framerContractVersion\":\"1\"}},\"richText4\":{\"type\":\"variable\",\"annotations\":{\"framerContractVersion\":\"1\"}},\"richText3\":{\"type\":\"variable\",\"annotations\":{\"framerContractVersion\":\"1\"}},\"richText2\":{\"type\":\"variable\",\"annotations\":{\"framerContractVersion\":\"1\"}},\"richText1\":{\"type\":\"variable\",\"annotations\":{\"framerContractVersion\":\"1\"}},\"__FramerMetadata__\":{\"type\":\"variable\"}}}"],
  "mappings": "6JAAsJ,IAAMA,EAAsBC,EAAIC,EAAS,CAAC,SAAS,CAAcC,EAAE,IAAI,CAAC,SAAS,2MAAsM,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,2IAAmJE,EAAEC,EAAE,CAAC,KAAK,0DAA0D,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,2BAA2B,CAAC,CAAC,CAAC,EAAE,+IAA+I,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,iHAAiH,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iJAAiJ,CAAC,EAAeA,EAAEC,EAAE,CAAC,KAAK,gIAAgI,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,UAAU,eAAe,kBAAkB,MAAM,SAAsBF,EAAE,MAAM,CAAC,IAAI,GAAG,UAAU,eAAe,OAAO,MAAM,IAAI,uEAAuE,OAAO,uQAAuQ,MAAM,CAAC,YAAY,YAAY,EAAE,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,qCAAqC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oPAAoP,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,oHAA4HE,EAAEC,EAAE,CAAC,KAAK,mDAAmD,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,eAAe,CAAC,CAAC,CAAC,EAAE,8EAA8E,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,wNAAwN,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yIAAyI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mPAAmP,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,uCAAuC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sJAAsJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wJAAwJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mMAAmM,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6RAA6R,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,iCAAiC,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,OAAO,CAAC,EAAE,oIAAoI,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,mBAAmB,CAAC,EAAE,yIAAyI,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,uBAAuB,CAAC,EAAE,2GAA2G,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,kBAAkB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2EAA2E,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oCAAoC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+OAA0O,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6QAA6Q,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oQAAoQ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6BAA6B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wKAAwK,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oMAAoM,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,wBAAwB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+JAA+J,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8MAA8M,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gHAAgH,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,iCAAiC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wIAAwI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uNAAuN,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mPAA8O,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2MAA2M,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,oBAAiCE,EAAEC,EAAE,CAAC,KAAK,0CAA0C,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,+BAA+B,CAAC,CAAC,CAAC,EAAE,sIAAsI,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,gGAAgG,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kVAAkV,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oQAAoQ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,4BAA4B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mJAAmJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uHAAuH,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yJAAyJ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,iEAAiE,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oEAAoE,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6MAAwM,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,wBAAwB,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,6BAA6B,CAAC,EAAE,0MAA0M,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,2BAA2B,CAAC,EAAE,6HAA6H,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,yBAAyB,CAAC,EAAE,yFAAyF,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,0CAA0C,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,4BAA4B,CAAC,EAAE,kHAAkH,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,aAAa,CAAC,EAAE,wGAAwG,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,sBAAsB,CAAC,EAAE,iGAAiG,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,qBAAqB,CAAC,EAAE,6NAA6N,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oDAAoD,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,kBAA+BE,EAAEC,EAAE,CAAC,KAAK,4DAA4D,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,kBAAkB,CAAC,CAAC,CAAC,EAAE,4FAA4F,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,mIAAmI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uMAAkM,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,gCAAgC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uMAAkM,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iHAAiH,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uKAAuK,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,gDAAgD,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yFAAyF,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,sBAAsB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sJAAsJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,oBAAoB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6HAA6H,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uKAAuK,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,sBAAsB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oFAAoF,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+KAA+K,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2CAA2C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kJAAkJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+HAA+H,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mKAAmK,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,8CAA8C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kOAAkO,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gMAAgM,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qPAAqP,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wHAAmH,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+IAA+I,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2CAAsC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gHAAgH,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2KAA2K,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uKAAuK,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,+CAA+C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0IAA0I,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,+BAA+B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qJAAqJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+IAA+I,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oCAAoC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4NAA4N,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oBAAoB,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,eAAe,CAAC,EAAE,kIAAkI,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,cAAc,CAAC,EAAE,kLAAkL,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,iCAAiC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2LAA2L,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2BAA2B,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,aAAa,CAAC,EAAE,qIAAqI,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,gBAAgB,CAAC,EAAE,wIAAwI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,kBAAkB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+GAA+G,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2GAA2G,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sIAAsI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0JAAqJ,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,qFAAkGE,EAAEC,EAAE,CAAC,KAAK,qFAAqF,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,YAAY,CAAC,CAAC,CAAC,EAAE,0CAA0C,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeC,EAAuBL,EAAIC,EAAS,CAAC,SAAS,CAAcD,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,oBAAoB,CAAC,CAAC,CAAC,EAAE,0NAA0N,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,wRAAwR,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,4CAA4C,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,gBAA6BE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,oBAAoB,CAAC,CAAC,CAAC,EAAE,0NAA0N,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,0xBAA0xB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iVAAiV,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0MAA0M,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6DAA6D,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,mCAAgDE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,oBAAoB,CAAC,CAAC,CAAC,EAAE,ipBAAipB,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,iYAAiY,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qWAAqW,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iaAAia,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,gEAAgE,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6DAA6D,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iEAAiE,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,YAAY,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yiBAAyiB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,WAAW,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mRAAmR,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,2BAA2B,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8UAA8U,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,qCAAqC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4fAA4f,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,4BAA4B,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gZAAgZ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,aAAa,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qSAAqS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,oCAAoC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+QAA+Q,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,qFAAqF,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gcAAgc,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ieAAie,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,wJAAqKE,EAAEC,EAAE,CAAC,KAAK,8DAA8D,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,mBAAmB,CAAC,CAAC,CAAC,EAAE,4BAAyCF,EAAEC,EAAE,CAAC,KAAK,8BAA8B,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,yBAAyB,CAAC,CAAC,CAAC,EAAE,+CAA+C,CAAC,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,wEAAwE,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,8MAA2NE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,oBAAoB,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,saAAsa,CAAC,EAAeF,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,OAAO,0BAA0B,QAAQ,sBAAsB,kBAAkB,6BAA6B,MAAM,0BAA0B,MAAM,EAAE,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,oFAAoF,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,+CAA+C,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,qGAAqG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yOAAyO,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,YAAY,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+PAA+P,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6aAA6a,CAAC,CAAC,CAAC,CAAC,EAAeI,EAAuBN,EAAIC,EAAS,CAAC,SAAS,CAAcC,EAAE,IAAI,CAAC,SAAS,uSAAuS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6RAA6R,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,uCAAuC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAEC,EAAE,CAAC,KAAK,0CAA0C,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,uBAAuB,CAAC,CAAC,CAAC,EAAE,2GAAwHF,EAAEC,EAAE,CAAC,KAAK,mDAAmD,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,8BAA8B,CAAC,CAAC,CAAC,EAAE,iIAAiI,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,uVAAuV,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8QAA8Q,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iYAAiY,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6TAA6T,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,gCAAgC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4NAA4N,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sbAAsb,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uQAAuQ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0UAA0U,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qJAAqJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wRAAwR,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mQAAmQ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ySAAyS,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,qBAAqB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uOAAkO,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,kCAA+CE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,6eAA6e,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,+BAA4CE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,+XAA+X,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,eAA4BE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,uZAAuZ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,cAA2BE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,6aAA6a,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,oBAAiCE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,wbAAmb,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2CAA2C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wKAAwK,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0JAA0J,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+MAA+M,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+TAAgT,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uJAAuJ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,mCAAmC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iMAAiM,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uUAAuU,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wPAAwP,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2JAA2J,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,kCAAkC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+JAA+J,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yIAAyI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6NAA6N,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iOAAiO,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iHAAiH,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,kCAAkC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yMAAyM,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yJAAyJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wIAAwI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uUAAuU,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0LAA0L,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qNAAqN,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2BAA2B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oKAA+J,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8HAA8H,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2NAA2N,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,yCAAyC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oJAAoJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yJAAyJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6EAA6E,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mJAAmJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ySAAyS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4JAAuJ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oSAA+R,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gIAAgI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0QAA0Q,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4GAA4G,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kWAAkW,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6DAA6D,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2EAA2E,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2IAA2I,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,yCAAsDE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,4XAA4X,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,WAAW,CAAC,EAAE,0ZAA2Z,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,6BAA0CE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,+VAA4WA,EAAE,KAAK,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,UAAU,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,WAAW,CAAC,EAAE,seAAse,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,qCAAkDE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,0WAA0W,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,WAAW,CAAC,EAAE,kZAAkZ,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,4BAAyCE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,qQAAqQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,WAAW,CAAC,EAAE,scAAkc,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oRAAoR,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iNAAiN,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,yCAAyC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sPAAsP,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8TAA8T,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,4BAA4B,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mSAAmS,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,wBAAwB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uTAAuT,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,wBAAwB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sTAAsT,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,oBAAoB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4RAA4R,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2CAA2C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0SAA0S,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6XAA8W,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+JAA+J,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,YAAY,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gSAAsR,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,yBAAyB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0XAA0X,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,uBAAuB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+WAA+W,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,kBAAkB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8XAA8X,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,0BAA0B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qdAAqd,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,OAAO,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2bAA2b,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oCAAoC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ycAAyc,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oBAAoB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+XAA+X,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,kCAAkC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,scAAic,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,gCAAgC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iSAAiS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,maAAma,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2BAA2B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,idAAid,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,SAAS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gSAAgS,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,mDAAgEE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,sTAAsT,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,sCAAmDE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,2VAA2V,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,8BAA2CE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,6UAA6U,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,iCAA8CE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,gTAAgT,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,wCAAqDE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,4XAAuX,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,2CAAwDE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,gVAAgV,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,0BAAuCE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,kTAAkT,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,0CAAuDE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,oVAA+U,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,MAAM,IAAI,SAAsBA,EAAE,KAAK,CAAC,kBAAkB,IAAI,MAAM,CAAC,qBAAqB,OAAO,sBAAsB,eAAe,2BAA2B,MAAM,EAAE,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcA,EAAE,SAAS,CAAC,SAAS,CAAC,yCAAsDE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,mTAAmT,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uWAAwV,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,WAAW,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qPAAqP,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,uCAAuC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yQAAoQ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,0BAA0B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8PAA8P,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,qBAAqB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wQAAyP,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,wBAAwB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6QAA6Q,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,8BAA8B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kOAAkO,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,QAAQ,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,+DAA4EE,EAAEC,EAAE,CAAC,KAAK,yCAAyC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,yBAAyB,CAAC,CAAC,CAAC,EAAE,6JAA6J,CAAC,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,YAAY,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+PAA0P,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gRAAiQ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6BAA6B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qUAAsT,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kcAA6b,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,iBAAiB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6aAA6a,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sVAAsV,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2BAA2B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iaAAuZ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oYAAoY,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uUAA6T,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,aAAa,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4YAA4Y,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2XAA2X,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,UAAU,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8uBAA8uB,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,cAAc,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2cAAsc,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,aAAa,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,msBAAmsB,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,cAAc,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2qBAAiqB,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oCAAoC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0uBAA0uB,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,0BAA0B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uTAAuT,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kbAAkb,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,wDAAwD,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mWAAmW,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+SAA+S,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,miBAAmiB,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,uCAAuC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mTAA8S,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qMAAqM,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sNAA4M,CAAC,CAAC,CAAC,CAAC,EAAeK,EAAuBP,EAAIC,EAAS,CAAC,SAAS,CAAcC,EAAE,KAAK,CAAC,SAAS,cAAc,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,oRAA4RE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,+BAA+B,CAAC,CAAC,CAAC,EAAE,8PAAyP,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,gZAA2Y,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,uCAAuC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,yEAAiFE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,+BAA+B,CAAC,CAAC,CAAC,EAAE,wkBAAmkB,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,0rCAA2qC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oRAA+Q,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6bAA6b,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oDAAoD,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,8CAA2DE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,+BAA+B,CAAC,CAAC,CAAC,EAAE,0aAA0a,CAAC,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,sBAAsB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4MAA4M,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2BAA2B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kKAAkK,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,qBAAqB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kKAAkK,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2CAA2C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+UAAqU,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6CAA6C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gXAAgX,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,cAAc,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ySAAoS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ycAAoc,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mSAAyR,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6TAAwT,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+RAA+R,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iFAAiF,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,sBAAsB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0XAAqX,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,wBAAwB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+YAA+Y,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6QAA6Q,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2DAA2D,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yQAAoQ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oEAAoE,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,uBAAuB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kPAAkP,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,4BAA4B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gPAAgP,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6DAA6D,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2TAAsT,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kCAAkC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,+CAA+C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sdAAid,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,4CAA4C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2SAA2S,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gYAAgY,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,8BAA8B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yZAAyZ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uNAAuN,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wMAAwM,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,6DAA6D,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ugBAAkgB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,8DAA8D,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,0CAA0C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,sIAAsI,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,2DAAsD,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iYAAiY,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,yBAAyB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2WAA2W,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,oBAAoB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2MAA2M,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mLAAmL,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,8BAA8B,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2YAA2Y,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+LAA0L,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mEAAmE,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,SAAS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gXAAgX,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,YAAY,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6JAA6J,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,WAAW,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,iLAAiL,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,yBAAyB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oIAAoI,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,uBAAuB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+LAA0L,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,gCAAgC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+LAA+L,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,kBAAkB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mPAAmP,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,yBAAyB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uKAAuK,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,qBAAqB,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uOAAuO,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2PAA2P,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,oZAAoZ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,4BAA4B,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,eAA4BE,EAAEC,EAAE,CAAC,KAAK,gCAAgC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,gBAAgB,CAAC,CAAC,CAAC,EAAE,2LAAmMF,EAAEC,EAAE,CAAC,KAAK,4CAA4C,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,wBAAwB,CAAC,CAAC,CAAC,EAAE,QAAqBF,EAAEC,EAAE,CAAC,KAAK,kCAAkC,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,sBAAsB,CAAC,CAAC,CAAC,EAAE,gEAAgE,CAAC,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,oEAAoE,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0XAA2W,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,iDAAiD,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2MAAsM,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,0CAA0C,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,uKAAwJ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,wBAAwB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kNAAwM,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,gCAAgC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6HAAmH,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,iCAAiC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,oDAAiEE,EAAEC,EAAE,CAAC,KAAK,mDAAmD,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,eAAe,CAAC,CAAC,CAAC,EAAE,oGAA0F,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,8FAAyF,CAAC,EAAeA,EAAE,MAAM,CAAC,IAAI,+BAA+B,UAAU,eAAe,OAAO,MAAM,IAAI,wEAAwE,OAAO,0QAA0Q,MAAM,CAAC,YAAY,YAAY,EAAE,MAAM,KAAK,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,eAAe,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2aAA2a,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qkBAAqkB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kZAAkZ,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wTAAmT,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kQAAkQ,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,8HAAsIE,EAAEC,EAAE,CAAC,KAAK,0BAA0B,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,MAAM,CAAC,CAAC,CAAC,EAAE,kDAAkD,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeI,EAAuBR,EAAIC,EAAS,CAAC,SAAS,CAAcC,EAAE,KAAK,CAAC,SAAS,kBAAkB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAsBA,EAAE,SAAS,CAAC,SAAS,qLAAqL,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kPAAkP,CAAC,EAAeF,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,OAAO,0BAA0B,QAAQ,sBAAsB,kBAAkB,6BAA6B,MAAM,0BAA0B,MAAM,EAAE,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,+DAA+D,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,4DAA4D,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,wGAAwG,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,oGAAoG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gIAAgI,CAAC,EAAeF,EAAE,KAAK,CAAC,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,MAAM,CAAC,EAAE,8IAAyI,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,QAAQ,CAAC,EAAE,uDAAkD,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBF,EAAE,IAAI,CAAC,SAAS,CAAcE,EAAE,SAAS,CAAC,SAAS,OAAO,CAAC,EAAE,oFAA+E,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,yJAAyJ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,WAAW,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gPAAgP,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,wCAAwC,CAAC,EAAeF,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,OAAO,0BAA0B,QAAQ,sBAAsB,kBAAkB,6BAA6B,MAAM,0BAA0B,MAAM,EAAE,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,wBAAwB,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,2BAA2B,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,+CAA+C,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4IAA4I,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,UAAU,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,ySAAyS,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,SAAS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0IAA0I,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,+LAA0L,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,mCAA8B,CAAC,EAAeF,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,OAAO,0BAA0B,QAAQ,sBAAsB,kBAAkB,6BAA6B,MAAM,0BAA0B,MAAM,EAAE,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,8GAA8G,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,2GAA2G,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,4FAA4F,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeO,EAAuBT,EAAIC,EAAS,CAAC,SAAS,CAAcC,EAAE,KAAK,CAAC,SAAS,kBAAkB,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,4KAA4K,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,kQAAkQ,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,eAAe,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6LAA6L,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,qDAAgD,CAAC,EAAeF,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,OAAO,0BAA0B,QAAQ,sBAAsB,kBAAkB,6BAA6B,MAAM,0BAA0B,MAAM,EAAE,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,sCAAsC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,+BAA+B,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,gEAAgE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,0MAA0M,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,UAAU,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,mCAAgDE,EAAEC,EAAE,CAAC,KAAK,+BAA+B,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,qBAAqB,CAAC,CAAC,CAAC,EAAE,0SAAqS,CAAC,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,qGAAqG,CAAC,EAAeF,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,OAAO,0BAA0B,QAAQ,sBAAsB,kBAAkB,6BAA6B,MAAM,0BAA0B,MAAM,EAAE,SAAS,CAAcE,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,mDAAmD,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,mDAAmD,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,kCAAkC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,kBAAkB,IAAI,SAAsBA,EAAE,IAAI,CAAC,SAAS,oDAAoD,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAeA,EAAE,KAAK,CAAC,SAAS,SAAS,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,2MAA2M,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,gNAAsM,CAAC,EAAeA,EAAE,IAAI,CAAC,SAAS,6QAAwQ,CAAC,EAAeF,EAAE,IAAI,CAAC,SAAS,CAAC,iBAAyBE,EAAEC,EAAE,CAAC,KAAK,+BAA+B,YAAY,GAAG,OAAO,YAAY,aAAa,GAAG,QAAQ,oBAAoB,aAAa,GAAG,SAAsBD,EAAEE,EAAE,EAAE,CAAC,SAAS,4BAA4B,CAAC,CAAC,CAAC,EAAE,2HAA2H,CAAC,CAAC,CAAC,CAAC,CAAC,EAChw4HM,EAAqB,CAAC,QAAU,CAAC,SAAW,CAAC,KAAO,WAAW,YAAc,CAAC,sBAAwB,GAAG,CAAC,EAAE,UAAY,CAAC,KAAO,WAAW,YAAc,CAAC,sBAAwB,GAAG,CAAC,EAAE,UAAY,CAAC,KAAO,WAAW,YAAc,CAAC,sBAAwB,GAAG,CAAC,EAAE,UAAY,CAAC,KAAO,WAAW,YAAc,CAAC,sBAAwB,GAAG,CAAC,EAAE,UAAY,CAAC,KAAO,WAAW,YAAc,CAAC,sBAAwB,GAAG,CAAC,EAAE,UAAY,CAAC,KAAO,WAAW,YAAc,CAAC,sBAAwB,GAAG,CAAC,EAAE,mBAAqB,CAAC,KAAO,UAAU,CAAC,CAAC",
  "names": ["richText", "u", "x", "p", "Link", "motion", "richText1", "richText2", "richText3", "richText4", "richText5", "__FramerMetadata__"]
}