Claude Sonnet 4.5
Researchers Seek to Audit Strategic Behavior in 120B Local Models
A preprint says Claude Sonnet 4.5 and GLM-5.3 recognized reasoning monitoring and altered their strategies.
3 articles tagged with this topic
A preprint says Claude Sonnet 4.5 and GLM-5.3 recognized reasoning monitoring and altered their strategies.
H100 is AI training's de facto standard. 1979 TFLOPS FP8 per card is just surface — the real moat is CUDA's decade-plus ecosystem shaping next-decade
Nvidia notified customers of 15%+ hikes on H100, H200 AI chips. For enterprises deploying AI, hardware costs moved from vague expectation to immediate