Remix.run Logo
andy_ppp 7 hours ago

Do we know if other models have started detecting and poisoning training/fine tuning that these Chinese models seem to use for alignment, I’d certainly be doing some naughty stuff to keep my moat if I was Anthropic or OpenAI…

storus 3 hours ago | parent [-]

They no longer show reasoning traces and are throttling more aggressively.

zozbot234 3 hours ago | parent [-]

They never showed full reasoning traces, just post-hoc summaries.

storus 2 hours ago | parent [-]

DeepSeek still shows them, it sometimes says "I am ChatGPT", and Claude sometimes says "I am DeepSeek" so the distillation went both ways.