Meta Unveils Self-Learning AI Model

Meta has launched new AI models, including a “self-taught evaluator,” marking a step towards reducing human involvement in the development of neural networks, according to Reuters. 

The artificial intelligence is based on a tool that relies on “chain of thought” technology.

Previously, the startup OpenAI introduced a large language model, o1, trained using reinforcement learning to perform complex reasoning. The neural network can “think”—it is capable of creating a long internal chain of thought while analyzing a question.

Researchers from Meta’s research division trained the evaluator model using data generated by another artificial intelligence, eliminating human involvement at this stage. 

In a conversation with Reuters, they noted that this approach paves the way for creating autonomous AI agents capable of learning from their mistakes.

Self-improving models eliminate the need for the costly and inefficient process of “reinforcement learning from human feedback,” the publication notes.

“We hope that as AI becomes increasingly superhuman, it will be able to check its work more effectively. Therefore, it is indeed better than the average person,” commented researcher Jason Weston.

Back in September, Meta introduced a series of AI innovations, including new versions of the large language model Llama 3.2, the voice AI assistant Meta AI, updates to Ray-Ban smart glasses, and other developments.

In October, the corporation launched Movie Gen—an AI generator for creating new videos, editing existing ones, and adding sound to them.