Facebook owner Meta released a series of new AI models on Friday, including one that can evaluate independently and with less human intervention.

The new model uses a “thought chain” technique to break complex problems down into smaller steps, Reuters reports. This increases accuracy in science, programming and mathematics.

“Autodidactic model”

This model is trained entirely on AI-generated data without any human involvement. In this way, Meta gives a taste of autonomous AI agents that can learn from their mistakes. These agents could soon be intelligent assistants capable of autonomously completing a variety of tasks without manual input.

This would mean that it is a cumbersome and expensive process Reinforcing learning from human feedback would be superfluous. “We hope that AI will become more and more in control of its work, so that it will eventually become better than the average human,” said one of the researchers.

Meta also released other AI tools on Friday, including an update for Segment everything Image recognition model, a tool that accelerates the response time of large language models, and datasets that can help discover new inorganic materials.