AI safety, interpretability and harnessing. Building KAOS and dynolab.dev
Highlights
- Pro
Pinned Loading
-
technical-ai-safety
technical-ai-safety PublicBusiness-document fine-tuning silently rewrites LLM safety behavior: rank-4 LoRA amplifies refusals to 87%, rank-32 collapses them to 10%.” Topics: interpretability, ai-safety, lora, probes
Python 4
Something went wrong, please refresh the page to try again.
If the problem persists, check the GitHub status page or contact support.
If the problem persists, check the GitHub status page or contact support.





