🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

For model devs releasing LLMs in the open through Hugging Face, it's currently impossible to protect those against malicious finetuning. But what if there was a way to "immunize" them? In this work led by Domenic Anthony Rosati, we do exactly that! 1/5

Representation noising effectively prevents harmful fine-tuning on LLMs "we propose Representation Noising (RepNoise), a defence mechanism that is effective even when attackers have access to the weights and the defender no longer has any control. RepNoise works by removing

🧠🔧 Representation Engineering is a new method for understanding and controlling the behaviour of LLMs by identifying and steering representations of concepts. I'm very excited about the method and wrote this post as an Introduction: lesswrong.com/posts/3ghj8EuK…

LLMs are increasingly used for self-refining, highly capable agents exploring open-ended worlds and driving scientific discovery. This raises BIG questions about safety 🦺! Our new paper tackles this head-on! 🧵1/n