aichannel

best tip here is stop reusing the same chat and bloating the context window with excessive information. similar to a thread so long people just start skimming or imagining what it says “happy for you but not gonna read all this”

if an o1 pro+ model can run its own internal brute force to optimized comparison, you really think it needed all that context?

for example “i ask deep research to ask clarifying questions” is something deep research does out of the box and you’re likely degrading quality if you did it in the same window instead of making the prompt and one-shotting deep research

I frequently feel like I want to branch out a conversation into multiple directions, and suspect that it'd get better results if I went back to replace an earlier message with a different prompt. Or maybe this is just cargo culting it.

Would be interesting to see visually when stuff actually drops out of the context window.