Work
At Trillion Labs, pushing on foundation model training and evaluation. Day-to-day this looks like throughput work on the training stack, designing eval suites we can actually trust, and arguing about data.
Side threads
- Rebuilding this site as a proper research notebook. Blog posts, paper landing pages, Korean translations — slowly.
- Reading the long tail of post-training papers that didn't get the attention they deserved.
Not doing
- New collaborations through end of quarter — head down on the current release.
- Conference travel until the next checkpoint lands.
This page is updated when reality changes, not on a schedule. Last edit above.