to me, it’s not about the number of iterations, but the “fitness function”. in nature, “survival of the fittest” is who can reproduce. machine learning on problems where there are right and wrong answers works well. but these “AGI” LLMs are just trained to bullshit confidently

“watch this 3 minute movie i spent 6 months making”

“oh wow, even if it’s bad, i’ll understand you better”

“watch this 3 minute movie i spent 5 seconds making by writing a prompt”

“why are you wasting my time? just tell me the prompt”

when i watch movies or listen to music, i examine to the details. i ask myself, “why did the artist make that choice?”

with ai-generated art, there’s no intentionality to the details. it’s just filler

it’s a cold chicken mcnugget without the breading

the thing is, when someone’s switching jobs, they’re already changing teams, so the cost is baked in

if you say, “well, people keep leaving for new jobs that are in-person, so we’re gonna RTO”, then you aren’t properly accounting for that cost

in other words, once you’ve formed teams that span the continent, you can’t really “RTO” without destroying those teams. maybe the destruction pays off somewhere else, but you’ve gotta at least acknowledge it

Later on I saw an episode of Voyager where Torres does the same thing. In both cases, the CO accepted the hard estimate.

(I put on 90s scifi tv shows in the background while I’m coding.)

before november there was a lot of good tech discussion on my mastodon feed (i know i know), but since november, it’s also 90% politics

Heh, the last systems I worked on at my previous job were called “zinc” and “copper”.

(I can talk about this because they were (a) open source, and (b) cancelled.)

I wonder where the ones they built for American Gladiators ended up.