As a novel artificial intelligence (AI) model, Meta has introduced the “Self-Taught Evaluator.”
The company’s research division unveiled the novel concept in an August report with the goal of reducing the amount of human intervention in AI development.
Like OpenAI’s o1 models, the Self-Taught Evaluator employs a “chain of thought” methodology.
The technique increases response accuracy on difficult problems in domains like science, coding, and arithmetic by breaking down complicated problems into smaller, logical steps.
AI Model Trained Without Human Input
At this point, human input was no longer required because the Self-Taught Evaluator was trained only on AI-generated data.
Building autonomous AI agents that can learn from their mistakes may be made possible by this ability to utilise AI to accurately evaluate other AIs.
The project’s researchers think these self-improving models could take the place of the existing Reinforcement Learning from Human Feedback (RLHF) approach, which is costly and ineffective because it depends on human annotators with specific knowledge.
Meta’s AI Model Aims for Super-human Level of Intelligence
One of the project’s researchers, Jason Weston, had high hopes for artificial intelligence.
“We hope, as AI becomes more and more super-human, that it will get better and better at checking its work,” he explained.
Weston explained that the secret to achieving superhuman AI is self-taughtness and self-evaluation.
The development of Meta coincides with ongoing research on Reinforcement Learning from AI Feedback (RLHF) by other tech behemoths like Google and Anthropic.
Meta Also Updates Other AI Tools
Other AI tools have also been improved by Meta in addition to the Self-Taught Evaluator.
These include a tool that accelerates the time it takes to generate Language Model (LM) responses and an enhancement to the company’s image-identification Segment Anything model.
To further demonstrate its dedication to developing AI technology, the tech giant has also made information available to aid in the discovery of new inorganic materials.
