Alexander Wettig
Researcher at Cursor; contributor to Ai2's OLMoE and co-author of SWE-bench
- Researcher working on RL for the Composer models, Cursor (Anysphere)1
Alexander Wettig is a researcher at Cursor working on reinforcement learning for the Composer models. Wettig was previously a PhD student at Princeton University advised by Danqi Chen, with research on building better language models through training data, including pre-training data selection, data mixtures, data repetition, and long-context recipes. The OLMoE paper credits Wettig with advising on pretraining and analyzing load balancing, routing, and domain specialization in Ai2's open mixture-of-experts model, and Wettig is a co-author of the SWE-bench benchmark paper.123
Work
OLMoE (OLMo family)Open weights
Co-author of the OLMoE paper; its author-contributions section says Wettig advised on pretraining, analyzed load balancing, routing, and domain specialization, and helped edit the paper. The paper releases the model weights, training data, code, and logs.2
SWE-bench (external site: arxiv.org)Published research
Third author of "SWE-bench: Can Language Models Resolve Real-World GitHub Issues?" (2023); Wettig's site says Wettig was part of the team that built SWE-bench and SWE-agent.31
Other documented roles
Roles the sources document but that are not confirmed as current.
- PhD student advised by Danqi Chen, Princeton University1
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.