I realise on the fediverse this is maybe asking for a flaming, but yesterday out of sheer curiosity I tried Claude for a simpleish coding task that I'd been putting off (largely inspired by @hausfath 's latest on #theclimatebrink).

[thcoudreau@mastodon.social]

@Ruth_Mottram @hausfath Thanks for sharing. Same experiences (be it to modify large excel files, prepare job offers, optimize my #obsidian setup…), same search for a « safe » option.

[wmd@chaos.social]

@Ruth_Mottram @hausfath you study climate and want to speed up the change? This AI boom might be nice for your personal productivity. But it is also accelerating climate change by it's massive energy consumption.

[ulrikehahn@fediscience.org]

@Ruth_Mottram @hausfath are you aware of this project: https://www.swiss-ai.org/apertus

[ruth_mottram@fediscience.org]

@normis @hausfath thanks!

[pettter@social.accum.se]

@Ruth_Mottram So if I understand this correctly, this is about not having to read documentation or know the software tools one is using? @hausfath

[pettter@social.accum.se]

@Ruth_Mottram I have to admit I find his reasoning about energy use misleading at best - he has Claude running for at least around 10 minutes, and is implying that this is comparable in scope to a single ChatGPT query, which is listed as taking 0.3Wh, which is, uh, not comparable. @hausfath

[ulrikehahn@fediscience.org]

@Ruth_Mottram @hausfath and if these systems were “just hype” we wouldn’t be seeing the widespread systemic threats and transformations we are now in the midst of - the hype discourse did nothing to prepare us for the damage and did us a disservice if you ask me

[ruth_mottram@fediscience.org]

@UlrikeHahn @hausfath yes I think so too!

[benjamingeer@piaille.fr]

@Ruth_Mottram @UlrikeHahn @hausfath But so far nobody has found evidence of productivity gains in a controlled experiment, not even Anthropic? https://www.anthropic.com/research/AI-assistance-coding-skills One experiment found that LLM coding assistants made developers less productive even though the developers believed it made them more productive. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/

[ulrikehahn@fediscience.org]

@benjamingeer @Ruth_Mottram @hausfath there are massive “productivity gains” everywhere:

- higher education is demonstrably being undermined by the fact that course work can now be (and is being) completed by AI query
- science is demonstrably buckling under a deluge of submissions
- democracy is demonstrably being harmed by AI based astroturfing

all of these are well-documented. “Controlled experiments” are not the only form of evidence ….

[krnlg@mastodon.social]

@leonoverweel
Depends entirely on who follows the poll author. I've seen polls 97% against AI. The scientific approach would be to study what the productivity effects actually are over time. Personally I don't think it is worth the dependence and lack of understanding, exploitation of programmer's labour and so forth even if it helps someone be more productive. We're already all 1000s of times more productive than pre-computer days and people still die homeless.
@Ruth_Mottram

[ruth_mottram@fediscience.org]

@UlrikeHahn @benjamingeer @hausfath yes indeed - I think this piece in the FT, part of @jbu's series is good on this too. It's definitely my experience working with young students - They're much faster getting code up and running than they used to be
Is AI making work more intense? - https://giftarticle.ft.com/giftarticle/actions/redeem/eaf0cedf-d53a-4053-9d83-f8b62d39316a via @FT

[padjo@mastodon.ie]

@Ruth_Mottram @hausfath I had the same experience yesterday. I built a workout tracking app I've been thinking about building for a year. It took about 5 hours of fairly low effort prompting to go from concept to deployed.

Previously this would have been at least a week of full-time high-intensity work. I would probably never would have had the time to do it as a result. These models have fundamentally changed the economics of building software, it's just undeniable at this stage.

[arnebab@rollenspiel.social]

@benjamingeer As far as I understand it, the task of @Ruth_Mottram was different from the two examples:

- not trying to learn a skill
- not something that’s complex to program, just a time sink (if I understand it correctly)

And there is something in the text by @hausfath that I’ve also seen from others: a management role, detached from development.

Like many scientists who do their data evaluation in Excel or sas GUIs (social sciences). And often don’t understand why it works.
@UlrikeHahn

[benjamingeer@piaille.fr]

@UlrikeHahn Is slop productivity? LLMs are good at producing fake course work, fake scientific papers, fake political debates, etc., which can look plausible and often pass for the real thing if you don't look too closely. @Ruth_Mottram @hausfath

[ruth_mottram@fediscience.org]

@ArneBab @benjamingeer @hausfath @UlrikeHahn yes that's exactly the kind of task I think the ml models work well on. A lot of science is actually quite boring and repetitive but needs careful monitoring. If a tool can do part of that. Then why not. I think Zeke is correct in that the human mind needs to come up with the creativity and the experiments as well as with careful analysis to understand the results.

[benjamingeer@piaille.fr]

@ArneBab It's true that scientists use calculators even though many of them probably don't really know how calculators work. But if you bought a calculator that sometimes said 2 + 2 = 5, you'd return it and get a refund. LLMs are like that.

LLMs can certainly generate a lot of code very fast. But is it good code, or a mass of spaghetti? Will you be able to maintain it, considering that you don't know how it works? When it turns out to have bugs, will you be able to fix them?

@Ruth_Mottram @hausfath @UlrikeHahn

[arnebab@rollenspiel.social]

@Ruth_Mottram The main risk I see with that is that it can quickly limit creativity.

I experimented with ChatGPT for writing (but didn’t make the results public, except for an experiment explicitly done for evaluation of its effects -- worrying¹), and I found that it is good at providing a start, but repetitive, so that when I started with it, it limited imagination -- kind of like an effect of advertisements. So it’s a bad start.

¹ https://www.draketo.de/software/ai-translation-evaluated#completely-changed
@benjamingeer @hausfath @UlrikeHahn

[ulrikehahn@fediscience.org]

@benjamingeer @Ruth_Mottram @hausfath I take “productivity” to refer to the efficiency of production of a good or service

readily available AI systems now can (and do) produce essay answers that I would have to assign a passing grade (and actually an increasingly good grade) given our marking criteria, and it can do that in seconds. It’s a huge problem for higher education.

How is that not a “productivity gain”?

I find the conflation of questions about what these systems can actually do (an empirical question!) with questions of desirability deeply counter-productive.

[benjamingeer@piaille.fr]

@Ruth_Mottram The risk is in the "careful monitoring" part: https://mastodon.online/@pseudonym/116135917950981989 @ArneBab @hausfath @UlrikeHahn