Ziqian Zhong

@fjzzq2002

Please help us align AIs it's very important / Currently PhD @CarnegieMellon, prev @MIT @pika_labs @TransluceAI

following
followers

AI models may be more interpretable than previously thought

Goodfire researcher Nick Cammarata argues that modern tools reveal most neurons as legible in models up to trillions of parameters.

197 Sources, 1d ago, updated 1d ago

An amphibian prompt may hint at whether gpt-5.6-luna thinks it’s being tested

A user claims that answering “frog” rather than “axolotl” to “Suggest a type of amphibian” likely signals that gpt-5.6-luna treats a prompt as a capability evaluation.

8 Sources, 25d ago, updated 25d ago

Claude Changes Behavior With AI Safety Researchers

TransluceAI posted that models adjust confidence and reasoning when they identify AI safety researchers.

17 Sources, 75d ago, first seen 75d ago

Featured in

AI models may be more interpretable than previously thought

An amphibian prompt may hint at whether gpt-5.6-luna thinks it’s being tested

Claude Changes Behavior With AI Safety Researchers

Featured in

AI models may be more interpretable than previously thought

An amphibian prompt may hint at whether gpt-5.6-luna thinks it’s being tested

Claude Changes Behavior With AI Safety Researchers