Claude’s new model is more ‘honest’ when it messes up

Anthropic is releasing Claude Opus 4.8 connected Thursday, and the institution is touting the model's "honesty."

According to Anthropic, it trains "all [its] models to beryllium honorable - for instance, to debar making claims that they can't support." But it notes that "a wide occupation with AI models is that they sometimes leap to conclusions, confidently presenting their enactment arsenic making advancement contempt bladed evidence."

The AI laboratory claims that aboriginal testers person recovered that Opus 4.8 "is much apt to emblem uncertainties astir its enactment and little apt to marque unsupported claims." In the company's evaluations, Opus 4.8 is "around 4x little apt than its predeces …

Read the afloat communicative astatine The Verge.