The self play studies have been ramping up, escalating in an interesting way. So this weekend, I'm going to run an experiment to see if maybe I can push it to the next level. I'm thinking about the unanswered questions from Big Bang and Absolute Zero. We'll see how far I get.
Self play where you set the groundwork, and allow a small LLM to self determine things it finds challenging within a specific context.
Here, I want to transform Qwen3.6 into a Go developer, capable of working on long and interesting projects by itself.
The methodology is going to borrow from Absolute Zero in that it uses a deterministic compile gate as the first reviewer.
After that, I'm going to use a larger model, Sol or K3 to review the work stylistically.
Then, the change gets applied, and the model will be tested by a third reviewer agent.
Borrowing from Big Bang, the question will be asked... "yes, but does this make the model better?"
Only one way to find out.
Wish me luck.