Meta expands AI horizons with "Self-Taught Evaluator": Less work for humans, more for machines

Meta announced a new AI model "Self-Taught Evaluator" that can autonomously evaluate and train other AI models.

Here's What We Know

This innovation aims to reduce costs and speed up AI development by eliminating the need for human input required in the "Reinforcement Learning with Human Feedback" (RLAIF) method.

Self-Taught Evaluator uses a 'chain of reasoning' technique to break down complex tasks into simpler steps, improving the accuracy of answers in areas such as science, programming and maths. The model was trained solely on data generated by the AI, and according to researcher Jason Weston, in the future the AI will be able to check its performance better than the average human.

This development could significantly reduce human involvement in various fields, bringing us closer to a future where AI is fully autonomous.

Source: The Times of India