Meta, the owner of Facebook, announced on Friday the launch of several new AI models from its research division. One of these models, the “Self-Taught Evaluator,” could reduce the need for human intervention in the AI development process.
The tool’s release comes after Meta first described it in a paper published in August. In that work, Meta explained how the tool uses the same “chain of thought” technique as OpenAI’s recently released o1 models to produce accurate assessments of the models’ replies.
This method divides difficult problems into more manageable logical steps. It seems to increase the precision of answers in disciplines like math, science, and coding.
The evaluator model was trained by Meta’s researchers using just AI-generated data, removing human involvement at that point as well.
Two Meta researchers on the project told Reuters that AI’s ability to reliably evaluate other AI offers a promising glimpse into the future. This could lead to autonomous AI entities capable of learning from their own mistakes.
Such agents are envisioned by many in the AI industry as digital assistants that possess the intelligence to do a wide range of tasks without the need for human participation.
Self-improving models could eliminate the need for Reinforcement Learning from Human Feedback. This costly and often ineffective method relies on human annotators with specialized knowledge to label data and verify responses to complex writing and math problems.
“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 stated.
Research on the idea of Reinforcement Learning from AI Feedback, or RLAIF, has also been published by other businesses, such as Google and Anthropic. However, those businesses typically don’t make their models available to the general public, unlike Meta.
Meta unveiled several AI capabilities on Friday, including an update to its image-identification Segment Anything model and a tool that speeds up LLM response times. The company also introduced datasets to aid in developing novel inorganic materials.
Click here for more news on Technology.