alexechoi

Matches what others are saying here: on frontier models the extra structure mostly costs tokens unless the tool returns something grep can't easily get you. The piece I'd actually want is the session memory side, like "we tried approach X in this module last week and reverted it, here's why." Have you measured the code graph and the memory separately in your evals? I suspect almost all of the win is from the memory half.

Thanks! Let me know if you need any help getting set up!

Back in August I took a day trip to Lille in France. I found walking routes from blogs online but often they started far away from where I was or didn't see everything I wanted. I thought since LLMs offer so much information on this domain custom walking tours should be a thing!

Simply prompt where and what you wanna see and feel you can even add quirks like a "gangstar themed tour of New York". Would love any feedback!