Starcraft is APM-dependent. Unless the latency will improve greatly in frontier reasoning LLMs (which is unlikely), it will remain a bit like knitting with an excavator.
replyI predict latency will improve greatly in the next 12 months to more than 4x speed on current frontier tasks
replyYeah but Starcraft needs, like, 10-20x the APM these agents are doing.
replyI’m not convinced a lot of it can’t be solved with code mode.
replyMarine staggering for example seems like an ideal code mode task.
Do the humans get to use this auto-stagger too?
replyYeah. Some of it may just be "thinking" less rather than faster token generation.
replyShould be possible to play this with Jev.
reply