General Body Meetings
We host weekly GBMs discussing an interesting concept in AI safety! Anyone can come and snacks are provided!
Meetings are usually held in GITC 3600 on Wednesdays at 4 PM, though meeting times can vary based on the week. Check our Instagram or Discord for updates.
10/14 – LLM training pipelines
Discusses components of LLM training pipelines like pretraining, finetuning, and reinforcement learning, and how they intersect with AI safety.
Slides10/21 – Mechanistic Interpretability w/ IHEARTCOMPUTER
An interactive presentation demonstrating mechanistic interpretability, a popular field in technical AI safety that seeks to examine and understand model internals. This presentation also shows how to create a linear probe for refusal behavior, with a companion Colab notebook.
Slides11/4 – Emergent Misalignment
Coming soon! Come to GITC 3600 at 4 PM on Wednesday 11/4!
11/11 – Alignment Faking
Coming soon! Come to GITC 3600 at 4 PM on Wednesday 11/11!