Alexander Wettig (original) (raw)

My research interests revolve around building and understanding large language models, with a particular focus on their training data (QuRating, WebOrganizer, ProLong, Masking Rates). I have also worked on understanding why LMs are easy to adapt (via Kernel Behavior) and how we can interpret their internal workings (Transformer Programs, Edge Pruning).

I am also part of the team that built SWE-bench and SWE-agent.I am currently working on training coding agents at Cursor.

Publications

(* indicates equal contribution)