@0xr4re@thsottiaux a banked reset on pro100 is worth half of a reset on pro200. I suppose the point is just convenience. maybe they'll announce new features only available for pro200 at devday ๐ค
@AiDoomScroll@theo nah. apps can expose an mcp server or cli or smt for agents to use if they want, but ai is getting so good at computer use that it's not strictly necessary. lemme use my own damn harness.. i don't want a shitty chatbox inside my cad software
@jbouhier can't say I have favorites, but a great example would be the deepseek r1 paper. there they describe how they replicated openai o1-style reasoning RL for their own model. Im not 100% certain about this, but the kimi K2/k2.5 and glm 5 papers may also discuss the RL work they did
@jbouhier modern llms already get RL'd to all hell. RL on verifiable tasks is why frontier models (or even small models like glm5.3spark) can perform as well as they do.
llms did seem to have fundamental limits up until the labs figured out multimodality and RL on verifiable tasks
@behzadotdev@maria_rcks i use @โpi-plugins/subagent
it's not the greatest or most advanced subagent plugin, but it works without any custom ui elements. i think most extensions could work without adding custom ui / overlays
@behzadotdev@maria_rcks i suppose t3code would just not support custom pi ui, so any extensions that add it wouldn't work. not a huuuge loss, just ask pi to rewrite the extension to not use custom ui
@DaveShapi If all tables were replaced by planets and there is a bomb in space and there is a bug dancing, and I am the average American. Can I walk up the stairs in my house?
@kan_gaming in most places vaccines are open to 12-year-olds and up, so but not vaccinating (unless you have an actual reason), your best reason should be that you have the mental capacity of an 11-year-old
@ScrapManYT there needs to be a box that closes by itself, so that when you open it, you can grab a slice of pizza, and then a small flap can lock it into place (kinda like on a tea box)