Open michaelneale opened 2 weeks ago
I have been using this all week to great success.
@michaelneale curious what specific sort of thing do you feel was more effective or possible with this vs without?
@codefromthecrypt mostly harder problems - for example the "interactive" fix for goose came from it initially (goose wasn't able to solve things that deep before). It also seems to help avoid goose prematurely jumping to a solution by misunderstanding the nuances of a problem and doing a "too obvious" thing. Well worth it IMO (and as a toolkit - so it doesn't kick in all the time).
This starts to introduce the concept of a "slower" reasoner model (alongside accelerator and processor) - and uses it via a toolkit for enhanced debugging/understanding and code authoring.
It will use the o1 models in some cases as needed, to compliment gpt4-o and gpt-4o-mini in the openai case. The "reasoner" model doesn't do tool calling or planning directly, but is consulting.