Questions I'm working on

Research

Where facts live inside language models, and small models post-trained for messy real-world data.

  1. 01ExpenseLMFine-tuning a 4B model to turn messy expense text into policy-checked JSON. SFT more than doubled policy accuracy.Researcher2026
  2. 02Knowledge editingCan you change one fact inside a language model without breaking the rest of it?Researcher2026 · In progress