20/01/2026 3:20 PM - One-on-one prep

⬅️ [20/01/2026 10:13 AM](<./20_01_2026 10_13 AM.md>) | ⬆️ [2026 - January](<./README.md>) | [28/01/2026 10:48 AM](<./28_01_2026 10_48 AM.md>) ➡️

[x] @TODO Talk about the embedding update thing I was thinking about.
[x] Talk about Andy's research and the tool invocation novelty. Does he think that being able to choose a new tool that was not seen during training is enough?
[x] Where put references?

Set up fine-tuning script for cold start. (Q and A models)
Start writing evaluation scripts so that we can compare the fine-tuned model to the original. (Depth versus accuracy)

Emphasize ties to hazy oracles. Why are we doing this?

⬅️ [20/01/2026 10:13 AM](<./20_01_2026 10_13 AM.md>) | ⬆️ [2026 - January](<./README.md>) | [28/01/2026 10:48 AM](<./28_01_2026 10_48 AM.md>) ➡️