Skip to content
AI News HubLIVE
More
Summary only

Teaching AI models to say “I’m not sure”

A new training method improves the reliability of AI confidence estimates without sacrificing performance, addressing a root cause of hallucination in reasoning models.

SourceMIT News AIAuthor: Rachel Gordon | MIT CSAIL
Teaching AI models to say “I’m not sure”
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions

Summary only

Under the source content permission, this page provides an editorial summary, key points, and a link to the original article.

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Standard reinforcement learning trains AI models to be overconfident by rewarding only correctness.
  • MIT researchers developed RLCR, adding a calibration reward to improve confidence estimates.
  • RLCR reduced calibration error by up to 90% without sacrificing accuracy.
  • The method enhances trustworthiness of AI in critical applications like medicine and finance.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models