AI & Tech Unverified First seen 2026-10-10 15:01 UTC

Physicists propose formula to predict AI chatbot failures

Physicists at George Washington University say a formula can estimate when an AI chatbot will flip from good answers to bad ones, with early tests on small models backing it up.

Why “Unverified”

Reported by Decrypt alone, with no second source or primary confirmation yet

1 independent source group across 1 post · evidence score 56/100. How labels work

What the sources claim

Physicists at George Washington University say a formula can estimate when an AI chatbot will flip from good answers to bad ones.

“Physicists at George Washington University say a formula can estimate when an AI chatbot will flip from good answers to bad ones”
Decrypt
Early tests on small models back up the formula.

“early tests on small models back it up”
Decrypt

Sources

Decrypt newswire 2026-10-10 15:01 UTC
Here’s a Way to Predict When AI Chatbots Will Turn Bad

More in AI & Tech

All AI & Tech stories →

Summary and claims extracted by AI with prompt analyze-story@3 on 2026-10-10 15:10 UTC. Labels are computed from the sources, not by the model.