well, at least I THINK that's how it works, anyway I'm not learned enough to fully grasp it, but that seems to be the gist based on what it says in the git https://github.com/AtomicBot-ai/atomic-agent
S C
I really do think a lot of progress in what models can accomplish from here on out is going to be harness-side the actual models themselves are already fucking cracked, and largely constrained by how and what information is fed into the context, it seems
obviously progress is gonna made in the models themselves, too, but what's already out there is almost certainly very constrained by how uh, crude the harnesses are like sure the pay-per-token models have ctx in the millions, but you're paying for each and every one of those, and that does constrain what you can really accomplish without going fucking broke, even if the model can technically do it
S C
not as fast as the ascii version, this GSQ gguf but it is smaller, so I can give it more context, which arguably is worth it
Samu ／人◕ ‿‿ ◕人＼
mmhmm i just use codex and claude w fairly default setting i figure they probably perform pretty well for their respective model sets and then idk what harness to use if im using my video card anymore, usually maki
