Facebook owner Meta said Friday it is releasing a batch of new AI models from its research division, including a “self-taught evaluator” that offers a path toward less human involvement in the AI development process. Can do.
This release follows Meta’s introduction of the tool in an August paper, which explained that it uses the same “thought chain” used by OpenAI’s recently released O1 model to make reliable decisions about the model’s responses. “How depends on the technology.
That technique involves breaking complex problems into smaller logical steps and appears to improve the accuracy of responses on challenging problems in subjects such as science, coding and mathematics.
Meta’s researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well.
The two meta researchers behind the project told Reuters that the ability to use AI to evaluate AI offers a glimpse at a potential path toward building autonomous AI agents that can learn from their own mistakes.
Many in the AI field envision such agents as digital assistants that are intelligent enough to complete a wide range of tasks without human intervention.
Self-improving models could reduce the need for an often expensive and ineffective process used today called reinforcement learning from human feedback, which requires input from human annotators who have the ability to accurately label the data. Must have special expertise to perform and verify answers to complex mathematics and writing questions. Are correct.
Jason Weston, one of the researchers, said, “We hope, as AI becomes more and more super-human, it will get better and better at scrutinizing its work, so that it really is better than the average human. “
“The idea of being able to be self-taught and self-evaluate is fundamentally important to the idea of AI reaching this kind of super-human level,” he said.
Other companies, including Google and Anthropic, have also published research on the concept of RLAIF, or reinforcement learning from AI feedback. However, unlike Meta, those companies do not release their models for public use.
Other AI tools released by Meta on Friday include an update to the company’s image-recognition Segment Anything model, a tool that speeds up LLM reaction generation times and a dataset used to aid in the discovery of new inorganic materials. May go.
(Except for the headline, this story has not been edited by NDTV staff and is published from a syndicated feed.)
