Physicists Say Formula Predicts When AI Chatbots Turn Bad

The significance here is less about any single chatbot and more about whether AI reliability can be treated as a measurable property rather than something discovered after deployment. If a formula can flag degradation before it reaches outputs, that has obvious implications for any RWA or crypto operation leaning on AI for compliance, monitoring or customer-facing tools, where a silent quality drop carries real cost. The open question is whether the approach holds on the large, production-scale models that matter, since the reported tests were on small ones.