Meta introduces "Self-Taught Evaluator", an AI model that assesses other AI systems and learns from mistakes.

Meta has introduced the "Self-Taught Evaluator," an AI model capable of assessing other AI systems, potentially minimizing human involvement in AI development. Trained exclusively on AI-generated data and utilizing a "chain of thought" technique, it enhances accuracy in complex problem-solving. This innovation may lead to autonomous AI agents that learn from mistakes, possibly replacing current human-dependent methods like Reinforcement Learning from Human Feedback. Other updates include improvements to image identification tools.

October 18, 2024
21 Articles