⏳ Just-in-time fine-tuning

[hejny]

When you use language models, you have a choice:

1. use a strong model (such as gpt-4) and pay nothing up front, but more as you go.
2. You can fine tune the model, which requires a lot of data and you also have to pay for training, but then you can get better results and pay less.

ℹ There is an analogy between this and compilation vs. interpretation.
💡 But there is a third option, just-in-time compilation.

We can do something similar with model training and fine tuning:

3. Just use the strong model for the first x runs and then start fine tuning with the data from those x runs.

---

See also #37

[hejny]

What we tried

(Describe our experience here)

[hejny]

🔎 Existing solutions

(Do some research / discussion)

• Has anyone else had the same problem?
• Has a project come up with the solution
• Is there a research paper about it?
• Is there an article, video, podcast about it?
• Is this even a good idea?