New Keichobot
Once ChatGPT 3.5 was able to achieve some of the Keichobot-like things, we decided to wait and see.
New Keichobot
- Temporarily placed as the name of the new one for now on 2024-09-02.
2025-08-09
- Keichobot in 2025
2025-09-11
- Last time I showed Keichobot's code to Devin for his review.
- 2025-08-14
- After setting up Keicho's code, I asked Devin, "What would you do to improve it using LLM?" He suggested that we start by replacing the keyword extraction part and expand it step by step, which is roughly the same flow I had in mind.
- Uh, the current KeichoBot has deterministic behavior with no randomness, and tests are written assuming that, but is it ok to lose that behavior? I was told.
- This is a tricky problem...
- KeichoBot's testing problem is exactly the same as the common problem of how to test a product that uses LLM.
- and left it alone.
- I was doing pConceptMap and I thought if this can be made from zero at this speed, Keichobot should be made from zero again.
You don't do a very good keyword extraction.
- Debugging required
- Maybe just the thing for Langfuse to experiment with.
This page is auto-translated from /nishio/新Keichobot using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.