Physical AI long-tail cases are such a major problem in ML that people don't even perceive the pain: "it just is".
You launch 100 agents in parallel to test different setups and, in the end, pick the best results. But doing so, you just rendered 99% of the computation useless.
@haider1@b1rkh0ff Code better than most devs... The top 1% of devs code better than most devs 99.9999% of the population is absolutely trash at any given specialised task. My point is that, the tech is good an useful, but let's not hype it out of proportion.
@zack_overflow Come on, Opus, it's not really smarter than you. It's like saying a hair comb is smarter than you cause its better at tasks than your fingers.