Post

imo even a significant portion of AI Twitter/X isn’t keeping up with AI progress anymore, people who traditionally were on the bleeding edge stuff like, “Opus 4.7 isn’t good because it botched my shell script” perfect one shotting is cool, but at some point it takes a backseat to larger autonomy

even this www.strangeloopcanon.com/p/when-align... @strangeloopcanon.com is VERY much on the cutting edge of AI but he asserts that agents don’t take initiative meanwhile i’ve reproduced real initiative (small scale only) in two separate agents since Opus 4.7 landed

initiative meaning it inferred the next step of a process without being told?

meaning it identified a friction point we experienced during the day, analyzed why it happens, researched what other people do, and fixed it in a way that appeared to be novel, to me, while drinking coffee the next morning

whoa -- I am constantly asking it to do this kind of thing do you have some kind of meta loop that enables self improvement? or did it get here on its own from system prompt and context window