258
this post was submitted on 17 Jul 2026
258 points (92.4% liked)
Fuck AI
7560 readers
1150 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Wow, they are at best cherry picking the article they cite on productivity with AI, if not outright trying to lie. They say
But if you, y'know, actually read this February update, you would see that the researchers are lamenting the fact that they cannot draw conclusions effectively because of the proliferation of AI use. To wit, they say:
In fact here is the full passage Arstechnica cherry-picked from (emphasis mine):
Based on conversations, which is to say, not on their data. This after the study a year prior which suggested developers are not reliable reporters of their own productivity.
Yeah, METR is an industry group funded by Google, Microsoft, Anthropic etc. I happened to summarize their work just yesterday actually: https://lemmy.blahaj.zone/comment/21358850 but one thing I think it's important to point out is that there is no second study, not yet. The blog, and the open source dataset, are all we have, because the researchers decided to adjust their method. So the intimation arstechnica gives that they completed a second study is false.
And the selection effects they are talking about are pro-ai software developers being unwilling or unable to complete complex tasks without their precious ai anymore, whereas a year ago they were more willing to do both ways in order to prove how powerful ai is. In my opinion, as I stated, this reflects poorly on their levels of self confidence from using ai, more than saying anything about a speed boost/decline.
Ahaha. The research showed that developers don't know how fast they are, but when we talked to them they said that they're pretty fast, so we'd rather report on that. I see no problems or contradictions with this.
It would make sense that it's better now though?
Claude Code came out in Feb 2025, widely available in May 2025. The tool itself has had many improvements as has the model behind it. It's pretty much THE tool people get into AI with currently.
Shit, I can get bugfixes as oneshot prompts now with Fable 5 using the superpowers plugin (not really superpowers, just skill files and stuff to tell it how to use test-driven development and other nice practices). I describe the bug in some detail, it'll create a failing test, then fix the issue. Some of those have not been trivial to track down and fix at all, but I can give it to an LLM and go work on another issue at the same time.
That's irrelevant. I'm talking about what the researchers wrote, how that relates to their own prior research, and how Arstechnica reported on it. I am not making and have not made any claims about whether or not developers are actually more productive.