AI can explain how it ‘thinks’ – but can we still trust it? OpenAI finds a problem

GPT-6 Astra can reason more effectively, but OpenAI found that it can also control and reduce what appears in its chain of thought when researchers monitor it.

MediaNews présente le titre et le résumé publiés par le média. L'article complet se lit sur le site de l'éditeur. (www.financialexpress.com)