William J. Marshall

NICE.. Ive taken a crack at the builder/researcher environment to keep my chats from stumbling over each-other.. Excited to work in it.. Does it support Local Quants?

Interesting, I found that models of the 9B size do well with a simple Problem declaration and solution approach in the Thinking.
Of coarse It was doing intense mathematics but that seemed to be the sweet spot.

This was written as soon as the Reasoning models were released so it was pretty reactive. I was working with heavily post-trained models.

are the experiments documented? I'd like to see them, I'm still wrapping my head around the effects of reasoning on long context models. It seems to work well with new information or on Single objective and it's come quite a ways on the industry harnesses. In dynamic long context back and forth I usually rely on a well trained model and an external State system over reasoning.