@shoni.eth

"Existing LLMs have been aligned by prompts (which produces very fragile alignment)" Even more deeply aligned models like gpt and claude are broken through prompt injection.

0 reply

0 recast

0 reaction