A new approach to training large language models is drawing attention as a possible "next-gen" method for shaping how AI systems behave.
According to Forbes, reinforcement learning with metacognitive feedback is being offered as a next-generation way to shape AI large language models — the technology behind chatbots and other generative tools.
To unpack the terms in plain language: reinforcement learning is a training method in which a model is rewarded for desirable responses and discouraged from poor ones, gradually steering its behavior. "Metacognition" refers to thinking about one's own thinking — being aware of how you reason, not just what you conclude. Combining the two suggests a training loop that uses feedback about a model's reasoning process itself, rather than only judging its final answers.
The Forbes piece frames this as an emerging or proposed direction rather than a finished, widely deployed product. The details of how such systems are built, tested, or measured are not specified in the source material available here.
Why it matters: how AI models are trained determines how reliable, transparent, and controllable they are — so new methods that claim to shape not just what AI says but how it reasons could influence the trustworthiness of the tools millions of people increasingly rely on.