A New Trick Could Block the Misuse of Open Source AI

https://media.wired.com/photos/66ac18df881e05e1fe4ec09f/master/pass/business_ai_training_safeguards.jpg

Researchers have developed a way to tamperproof open source large language models to prevent them from being coaxed into, say, explaining how to make a bomb.

https://www.wired.com/story/center-for-ai-safety-open-source-llm-safeguards/