Spaces:

---

cyber-chris

/

dead-mans-switch

like 0

Runtime error

Hey, I'm on the https://aisafetyfundamentals.com/ course and I'm doing a small project to identify and use an LLM's own understanding of deception (via SAE extracted features) to see if it's a more robust alternative to RLHF safety training.

It'll be very useful to get a GPU since my demo is focused on llama3.

(This is purely for alignment research, no commercial interest. Plus, I'm a student, so a free GPU upgrade would be vastly preferable.)