@gregpr07 chatgpt is underrated once you realize it can dig really deep on github and can also clone your repo etc.
keep things clean and siloed to focused projects and it's the perfect free high-grade mega-prompt compiler
@gregpr07 > 5.6 sol high in chatgpt to research/plan
> 5.6 terra xhigh herding luna xhigh subagents
ideal 24/7 hill climbing setup as long as the loop is high quality
for sufficiently long-horizon agent rl, env engineering is probably as important a scaling axis as model flops; how many high-quality, causally rich environment-years can you generate per dollar? who is going to build the first planet-scale ci cluster?
@nickcammarata with minimalist malleable harnesses like pi and ds4 flash 0731 the bar to entry to iterate on the swarm control plane has never been lower
@nickcammarata feels within reach; need the right coordination primitives/substrate for efficient curated knowledge propagation. weak models can do narrow tasks well, just need the right distributed systems layer that coevolves all actors towards a broader higher dimensional mission
Given some of the results I'm seeing recently, it's pretty clear Codex is a good harness.
But it will seem primitive in 2-3 months and we're about to go through another major evolution in how we use AI at the frontier. The next generation of models need more than your laptop.
how can we build the matrix that will self-adapt and reorganize towards a high-dimensional, partly unquantifiable goal - not an obvious scalar target (achievable today) but more a matter of being able to touch the elusive higher construct we call "taste"
My job? I'm a rare token hunter. I track down dead languages in Tibetan monasteries, decrypt Tesla's private journals, chase whispers of pre-contact Amazonian dialects. The AIs pay top credit for tokens they've never tasted, you know. Work is work, even if it's for the machines.