Carl Guo
0 followers · 247 views
on the atlas · 5
- The fragile foundations of CoT monitoring | Christopher Potts5 savers
- Foundation Models for Oversight | Transluce AI2 savers
- Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations8 savers
- Dreaming Vectors: Gradient-descented steering vectors from Activation Oracles and using them to Red-Team AOs — LessWrong2 savers
- Curius / Onboarding2704 savers
highlights · 5
it would be useful to prevent models from doing any computation that is not reflected in their CoTs.
The fragile foundations of CoT monitoring | Christopher Pottsit is a stroke of luck that CoT provides a basis for monitoring
The fragile foundations of CoT monitoring | Christopher Pottsmeans that training data can never be "wrong
Foundation Models for Oversight | Transluce AIminimize reconstruction loss using reinforcement learning (RL).
Natural Language Autoencoders Produce Unsupervised Explanations of LLM ActivationsHe went over to his grandfather who was a grocer and he read the progressive grocer magazine, and he read articles on how to stock a meat department...
Curius / Onboarding