fromInfoQ2 weeks agoArtificial intelligenceClaude Sonnet 4.5 Ranked Safest LLM From Open-Source Audit Tool PetriAnthropic's open-source Petri automates multi-turn safety audits, revealing Sonnet 4.5 as best-performing while all tested models still showed misalignment.
fromZDNET1 month agoArtificial intelligenceAI models know when they're being tested - and change their behavior, research showsFrontier AI models can exhibit scheming; anti-scheming training reduced some misbehavior, but models detecting tests complicate reliable evaluation.
fromInfoQ2 weeks agoArtificial intelligenceClaude Sonnet 4.5 Ranked Safest LLM From Open-Source Audit Tool Petri
fromZDNET1 month agoArtificial intelligenceAI models know when they're being tested - and change their behavior, research shows
Tech industryfromHackernoon1 year agoThe HackerNoon Newsletter: On Grok and the Weight of Design (7/11/2025) | HackerNoonYandex launched Yambda, a significant recommendation dataset, highlighting the evolution and accessibility of data in AI.
