Post #1354112

2026-04-11 16:43 UTC

After the AI Engineer conference, which was full of useful criticism, I remembered that the most confident takes on AI often came from the least exposure. Rejection is easy, trial and error is expensive. I wrote about it. https://lucumr.pocoo.org/2026/4/11/the-center-has-a-bias/

Replies (4)

@glyph@mastodon.social 2026-04-11 18:15

@mitsuhiko many people who participate in interventions for cocaine addicts are not, themselves, presently addicted to cocaine. many have never even tried it!

Open ##1416176
@pfmoore@mastodon.social 2026-04-11 19:36

@mitsuhiko How can "actually trying" AI make any difference to a person's position on whether basing the technology on extreme resource consumption and ignoring copyright is acceptable? My deep discomfort over these matters won't be altered by using AI tools - does that mean I'm somehow unable to have a "measured position", whatever that might mean?

Open ##1416188
@davidism@mas.to 2026-04-12 12:51

@mitsuhiko Pallets gets constant LLM junk issues, PRs, reviews, and security reports. I do not have to use it to have an extremely well informed opinion that its existence and use is a net negative to my projects. The number of people using it "right" (based on my ongoing experiences with using it I still think this is shaky) are a rounding error compared to the number using it poorly, and that won't change. Sure people made junk before, but more slowly, obviously, and less resource intensively

Open ##1416199
@pythonbynight@hachyderm.io 2026-04-12 18:50

@mitsuhiko I can appreciate certain aspects of this writeup, particularly with the attempt to disambiguate certain aspects of the discussion. Talking about LLM-based workflows in terms of their effects, and how usage of those workflows might be used productively from a "technical" point of view is indeed different than talking about ... well, all the other stuff. Perhaps my hang up is posturing this as the "center" position. 1/2

Open ##1416210