Meta Unveils "Self-Taught Evaluator": An AI Model That Can Assess and Improve Other AI Systems

Meta Introduces Advanced AI Model That Can Assess the Performance of Other AI Systems

Meta has introduced the "Self-Taught Evaluator," an AI model that autonomously improves other AI systems using a "chain of thought" reasoning technique Meta, the parent company of Facebook and Instagram has introduced a development in artificial intelligence: a "Self-Taught Evaluator" model. This new AI system is part of a batch of innovative AI models revealed by Meta's research division, a shift toward autonomous systems that have the ability to evaluate and improve other AI models. The Self-Taught Evaluator leverages a "chain of thought" reasoning technique, similar to the one employed by OpenAI's latest model, OpenAI o1. This technique enables the AI to break down complex tasks into manageable sub-tasks, leading to improved reasoning and decision-making during response time. The model's enhanced capabilities offer a promising approach to accelerating the development of AI systems with minimal human intervention. One of the most striking features of Meta's Self-Taught Evaluator is that it learns from AI-generated data instead of relying on human-labeled data. Traditionally, AI models have depended on datasets annotated by humans to learn and evolve. However, by shifting away from this dependency, the Self-Taught Evaluator represents a more independent and self-sustaining approach to AI development. The model identifies its own mistakes, refines its understanding, and improves accuracy over time, particularly in tasks involving mathematics, scientific interpretation, and coding. The use of AI-produced data also means that the model is not constrained by human limitations, allowing it to refine its performance at a pace and depth that human evaluators might struggle to achieve. This has the potential to revolutionize fields like mathematics, engineering, and software development, where precise and highly accurate analysis is critical. As the model becomes more proficient, it can autonomously handle increasingly complex tasks, making it a valuable tool for industries relying on AI for critical functions. According to Meta researcher Jason Weston, this self-checking and self-evaluating ability could push AI beyond human oversight. "AI should become increasingly superhuman, checking itself better than before, which may be beyond human effort," he explains. The auto-evaluation feature is a milestone on the road to more sophisticated AI systems that require little or no human guidance. As AI continues to evolve, these autonomous systems could open up possibilities for greater innovation and efficiency across various industries. In addition to the Self-Taught Evaluator, has released several other AI products, further showing its commitment to advancing the field. Among these is a successor to its popular Segment Anything image identification model, which aims to improve speed and accuracy in identifying and segmenting visual data. Additionally, Meta unveiled a tool designed to accelerate response times in large language models and new datasets to aid researchers in discovering inorganic materials for scientific applications. These innovations collectively highlight Meta's ambitious push toward AI-driven progress, emphasising efficiency, independence, and continuous improvement. By creating systems that can evaluate, learn, and evolve on their own, Meta is positioning itself at the forefront of AI development, shaping the future of technology in a way that could transform how industries approach problem-solving and innovation.

Meta Releases AI Model That Can Check Other AI Models' Work

Meta AI's new model could help reduce human involvement in AI development Facebook owner Meta said on Friday it was releasing a batch of new AI models from its research division, including a "Self-Taught Evaluator" that may offer a path toward less human involvement in the AI development process. The release follows Meta's introduction of the tool in an August paper, which detailed how it relies upon the same "chain of thought" technique used by OpenAI's recently released o1 models to get it to make reliable judgments about models' responses. That technique involves breaking down complex problems into smaller logical steps and appears to improve the accuracy of responses on challenging problems in subjects like science, coding and math. Meta's researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well. The ability to use AI to evaluate AI reliably offers a glimpse at a possible pathway toward building autonomous AI agents that can learn from their own mistakes, two of the Meta researchers behind the project told Reuters. Many in the AI field envision such agents as digital assistants intelligent enough to carry out a vast array of tasks without human intervention. Self-improving models could cut out the need for an often expensive and inefficient process used today called Reinforcement Learning from Human Feedback, which requires input from human annotators who must have specialized expertise to label data accurately and verify that answers to complex math and writing queries are correct. "We hope, as AI becomes more and more super-human, that it will get better and better at checking its work, so that it will actually be better than the average human," said Jason Weston, one of the researchers. "The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI," he said. Other companies including Google and Anthropic have also published research on the concept of RLAIF, or Reinforcement Learning from AI Feedback. Unlike Meta, however, those companies tend not to release their models for public use. Other AI tools released by Meta on Friday included an update to the company's image-identification Segment Anything model, a tool that speeds up LLM response generation times and datasets that can be used to aid the discovery of new inorganic materials.

Meta releases AI model that can check other AI models' work

NEW YORK - Facebook owner Meta said on Friday it was releasing a batch of new AI models from its research division, including a "Self-Taught Evaluator" that may offer a path toward less human involvement in the AI development process. The release follows Meta's introduction of the tool in an August paper, which detailed how it relies upon the same "chain of thought" technique used by OpenAI's recently released o1 models to get it to make reliable judgments about models' responses. That technique involves breaking down complex problems into smaller logical steps and appears to improve the accuracy of responses on challenging problems in subjects like science, coding and math. Meta's researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well. The ability to use AI to evaluate AI reliably offers a glimpse at a possible pathway toward building autonomous AI agents that can learn from their own mistakes, two of the Meta researchers behind the project told Reuters. Many in the AI field envision such agents as digital assistants intelligent enough to carry out a vast array of tasks without human intervention. Self-improving models could cut out the need for an often expensive and inefficient process used today called Reinforcement Learning from Human Feedback, which requires input from human annotators who must have specialized expertise to label data accurately and verify that answers to complex math and writing queries are correct. "We hope, as AI becomes more and more super-human, that it will get better and better at checking its work, so that it will actually be better than the average human," said Jason Weston, one of the researchers. "The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI," he said. Other companies including Google and Anthropic have also published research on the concept of RLAIF, or Reinforcement Learning from AI Feedback. Unlike Meta, however, those companies tend not to release their models for public use. Other AI tools released by Meta on Friday included an update to the company's image-identification Segment Anything model, a tool that speeds up LLM response generation times and datasets that can be used to aid the discovery of new inorganic materials. (Reporting by Katie Paul in New York; Editing by Kenneth Li and Alistair Bell)

Meta Releases AI That Can Check and Improve Other AI Without Human Input | PYMNTS.com

Meta, the parent company of Facebook, announced on Friday the release of several new artificial intelligence (AI) models from its research division. Among the highlights is a groundbreaking "Self-Taught Evaluator," a tool designed to reduce human involvement in the development of AI systems. According to Reuters, this development represents a significant step toward creating autonomous AI agents capable of self-improvement and independent problem-solving. The new tool, originally introduced in an August research paper, employs the "chain of thought" technique, a method that breaks down complex problems into smaller, logical steps. This approach, used by OpenAI's recent models, enhances the accuracy of responses to intricate questions in fields like science, coding, and mathematics. Meta's researchers further pushed the boundaries by using only AI-generated data to train the evaluator, bypassing the need for human input at this stage. Per Reuters, the researchers behind the project, including Jason Weston, emphasized the potential of AI models that can reliably evaluate their own work. This ability could pave the way for autonomous AI systems that learn from their mistakes, a concept many in the field envision as a major advancement. Such systems could function as intelligent digital assistants, capable of handling a wide range of tasks with minimal or no human intervention. Related: Meta Enhances User Data Control, Resolving German Antitrust Dispute "We hope that as AI becomes increasingly superhuman, it will get better at checking its own work and surpass average human capabilities," said Weston. He added that the ability to be self-taught and self-evaluative is central to reaching a superhuman level of AI. One of the key advantages of these self-improving models lies in their potential to replace traditional methods like Reinforcement Learning from Human Feedback (RLHF). RLHF, a process that requires human annotators with specialized expertise, can be both costly and inefficient. By contrast, AI models that can self-evaluate could streamline development and improve accuracy without the need for human oversight. Other tech giants, including Google and Anthropic, have also explored similar concepts, specifically Reinforcement Learning from AI Feedback (RLAIF). However, unlike Meta, these companies typically do not release their models for public use. Meta's more open approach to sharing its research could lead to broader advancements in the AI community.

Meta releases AI model that can check other AI models' work

The release follows Meta's introduction of the tool in an August paper, which detailed how it relies upon the same "chain of thought" technique used by OpenAI's recently released o1 models to get it to make reliable judgments about models' responses.Facebook owner Meta said on Friday it was releasing a batch of new AI models from its research division, including a "Self-Taught Evaluator" that may offer a path toward less human involvement in the AI development process. The release follows Meta's introduction of the tool in an August paper, which detailed how it relies upon the same "chain of thought" technique used by OpenAI's recently released o1 models to get it to make reliable judgments about models' responses. That technique involves breaking down complex problems into smaller logical steps and appears to improve the accuracy of responses on challenging problems in subjects like science, coding and math. Meta's researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well. The ability to use AI to evaluate AI reliably offers a glimpse at a possible pathway toward building autonomous AI agents that can learn from their own mistakes, two of the Meta researchers behind the project told Reuters. Many in the AI field envision such agents as digital assistants intelligent enough to carry out a vast array of tasks without human intervention. Self-improving models could cut out the need for an often expensive and inefficient process used today called Reinforcement Learning from Human Feedback, which requires input from human annotators who must have specialized expertise to label data accurately and verify that answers to complex math and writing queries are correct. "We hope, as AI becomes more and more super-human, that it will get better and better at checking its work, so that it will actually be better than the average human," said Jason Weston, one of the researchers. "The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI," he said. Other companies including Google and Anthropic have also published research on the concept of RLAIF, or Reinforcement Learning from AI Feedback. Unlike Meta, however, those companies tend not to release their models for public use. Other AI tools released by Meta on Friday included an update to the company's image-identification Segment Anything model, a tool that speeds up LLM response generation times and datasets that can be used to aid the discovery of new inorganic materials.

Meta launches AI that can check other models work

Meta's self-taught evaluator is part of a new round of AI innovations developed by the company's research division team, following its introduction in an August paper. Technology and social media platform Meta has announced the release of several new AI models, including the self-taught evaluator, which can train other AI, without the need for human input, in the hope that it would improve efficiency and scalability for enterprises using large language models (LLM). The tool was first introduced in a paper released in August, where it was stated that the technology uses the same 'chain of thought' method previously seen in recent Open AI o1 models, to better generate reliable response and judgements. With human involvement in LLM evaluation often considered slow and more costly, the self-taught evaluator aims to create a training approach that eliminates the need for human-labelled data. The AI is prompted to break down difficult problems into manageable, practical steps, improving the accuracy of responses on a wide range of complex topics - for example science, coding and maths. Meta's research is part of a growing trend in which techniques incorporate LLMs as part of the automation and improvement process. Jason Weston, a Meta research scientist, told Reuters, "we hope, as AI becomes more and more super-human, that it will get better and better at checking its work, so that it will actually be better than the average human". "The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI," he said. In other Meta-related news, earlier this month the EU's highest court passed judgement in favour of Max Schrems, a privacy campaigner who stated that Facebook misused data about his sexual orientation to target him with personalised ads. The court agreed that the company had unlawfully processed Schrems personal data for the purpose of targeting him with specific advertisements. Under the EU data protection law, sensitive information in relation to sexual orientation, race/ethnicity or health status carries strict processing requirements. Don't miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic's digest of need-to-know sci-tech news.

Meta releases AI model that can check other AI models' work - VnExpress International

The release follows Meta's introduction of the tool in an August paper, which detailed how it relies upon the same "chain of thought" technique used by OpenAI's recently released o1 models to get it to make reliable judgments about models' responses. That technique involves breaking down complex problems into smaller logical steps and appears to improve the accuracy of responses on challenging problems in subjects like science, coding and math. Meta's researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well. The ability to use AI to evaluate AI reliably offers a glimpse at a possible pathway toward building autonomous AI agents that can learn from their own mistakes, two of the Meta researchers behind the project told Reuters. Many in the AI field envision such agents as digital assistants intelligent enough to carry out a vast array of tasks without human intervention. Self-improving models could cut out the need for an often expensive and inefficient process used today called Reinforcement Learning from Human Feedback, which requires input from human annotators who must have specialized expertise to label data accurately and verify that answers to complex math and writing queries are correct. "We hope, as AI becomes more and more super-human, that it will get better and better at checking its work, so that it will actually be better than the average human," said Jason Weston, one of the researchers. "The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI," he said. Other companies including Google and Anthropic have also published research on the concept of RLAIF, or Reinforcement Learning from AI Feedback. Unlike Meta, however, those companies tend not to release their models for public use. Other AI tools released by Meta on Friday included an update to the company's image-identification Segment Anything model, a tool that speeds up LLM response generation times and datasets that can be used to aid the discovery of new inorganic materials.