Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

←

Публикация

Dr Abeba Birhane

despite popular beliefs, incorporating "AI" in workflows does NOT reduce time. "when developers use AI tools, they take 19% longer than without—AI makes them slower" metr.org/blog/2025-07... the stark difference between forecasts and actual results in the plot is damming

Обсуждение

6 прямых ответов · 17 сообщений

Показаны не все ответы

Публичный счётчик показывает 7, а источник передал 6. Остальные ответы могли быть удалены, скрыты или не выданы публично.

Jon P (moved)

To me the most interesting result here was how much developers overestimated the value of AI even in retrospect -- thinking it had saved them time when it hadn't. That said @grimalkina.bsky.social has some very good critiques of the study at www.fightforthehuman.com/are-develope...

Are developers slowed down by AI? Evaluating an RCT (?) and what it tells us about developer productivitySeven different people texted or otherwise messaged me about this study which claims to measure “the impact of early-2025 AI on experienced open-source developer productivity.” You know, when I decid...www.fightforthehuman.com
Ответ для Jon P (moved)
Jon P (moved)

As somebody who used to work on static analysis tools that had high false positive rates but were still very useful in practice, I tend to think that good experienced devs will be able to leverage AI tools effectively despite their limitation, and over time so will less experienced devs. BUT!!!!!

Ответ для Jon P (moved)
Jon P (moved)

My fundamental issues with AI-based tools aren't whether or not they can be used effectively. Even if stealing people's stuff, burning down a rainforest, and causing cancer to a bunch of people living near data centers makes me more "effective" ... it's still very problematic!

Ответ для Jon P (moved)
Jon P (moved)

Is it even more problematic if I've fooled myself and it doesn't actually make me more productive? Yes, but that also potentially leads to thinking that if and when the productivity issues are addressed we should rethink our attitudes even if the deeper issues remain.

Ответ для Jon P (moved)
Cat Hicks

I find it difficult to separate that critique from all of computing though

Ответ для Cat Hicks
Jon P (moved)

Agreed that there's a general critique, I see "AI" as it's practiced today distills and intensifies the problems of the field as a whole. But yeah, when I read @abeba.bsky.social's paper on the values encoded in ML research, I remember thinking that it applies to software engineering in general

Ответ для Jon P (moved)
Cat Hicks

I totally agree. Just like the pandemic, the commercial enterprise of AI has both pulled back the curtain on *and* intensified a lot of the conditions that were already there and now finding new form.

Ответ для Jon P (moved)
GG Anderson

💯

Ответ для Jon P (moved)
GG Anderson

💯

Rarian Rakista

Some of these experts probably have written books on what the AI is trying to slop together. AI is like clip art, and real artists can just create, you know, art.

Ответ для Rarian Rakista
BardRaelyn

Clip art—bad though it often is—was created by real artists who got paid for the work and created the images specifically for others to use. And nobody is mistaking a clip art collage for real art. AI steals from artists in order to mimic their styles and techniques. AI is the opposite of clip art.

Jaroslav B.

Study says "AI-allowed and AI-disallowed groups is not statistically significant at common p-value thresholds." Only 16 devs and about 500h of work in total. Also tasks were done on top-tier high-quality open-source repositories issues/todos. Writing "just another react front-end" is much easier

Ответ для Jaroslav B.
Jaroslav B.

But it's a cool study anyway... it would be interesting to see how it compares with routine tasks like React front-end development, given the large number of developers and hours, which would make the p-value statistically significant.

Teresa M

📌

Mary Branscombe

there is some really good analysis by @grimalkina.bsky.social on how what the study shows is really what the summary says, let alone what the headlines say

Ответ для Mary Branscombe
Mary Branscombe

tssk; really *not*

Annie B Kay

I THOUGHT so!