Post #2556714
2026-05-14 19:07 UTC
Replies (16)
-
@jornfranke@social.anoxinon.de 2026-05-14 19:20
@shadow53@floss.social I would ask to measure that it indeed brings improvement and have additional validation (unit testing, static code analysis, fuzzing etc.) that it does not decrease quality. Static code analyzers give you also statistics on duplicated code and code complexity. They can be indicator that things go wrong. Additionally, you can collect number of incidents (if LLM would work they should go down).
-
@alcinnz@floss.social 2026-05-14 19:36
@shadow53@floss.social cc: @baldur@toot.cafe He has https://needtoknow.fyi/ which might be handy.
-
@fundamental@social.treehouse.systems 2026-05-14 19:43
@shadow53@floss.social https://www.anthropic.com/research/AI-assistance-coding-skills
-
@mastodonmigration@mastodon.online 2026-05-14 19:43
@shadow53@floss.social https://ai-project-website.github.io/AI-assistance-reduces-persistence/
-
@rhempel@cosocial.ca 2026-05-14 19:53
@shadow53@floss.social Unfortunately, you are unlikely to convince your manager that using AI for coding is a bad idea - and it's not because your argument is wrong, or that they are too dumb to understand. Your manager is probably also getting pressure from THEIR manager to use more AI to improve ... something. Consider saving yourself some stress and just keep doing your work without AI - it will probably be fine. Also consider putting out feelers for your next job ...
-
@tuban_muzuru@beige.party 2026-05-14 19:56
@shadow53@floss.social Here's a good compromise I have found works well with me coding with LLM code assist and my client who wants a Rust subsystem: I talk to the LLM about my spec. I bounce it back and forth several times. Just English. Spec is just words. Tokens are cheap at volume. You can iterate a spec ten times for the cost of one bad code generation attempt that you then spend three days debugging. The bottleneck is always the clarity of the ask.
-
@houba@spore.social 2026-05-14 20:16
@shadow53@floss.social If they appreciate academic research, there is always this https://link.springer.com/article/10.1007/s10676-024-09775-5 "We argue that these falsehoods, and the overall activity of large language models, is better understood as bullshit in the sense explored by Frankfurt (On Bullshit, Princeton, 2005): the models are in an important way indifferent to the truth of their outputs."
-
@hjhornbeck@freethought.online 2026-05-14 20:49
@shadow53@floss.social Are news reports allowed? If so: > All the developers I talked to were excited to try using LLMs at work at first, or were at least curious about them. Their feelings about the tools, based on their personal experience, are now overwhelmingly negative. ... > The problem all the developers I talked to agreed on is that the more they relied on AI to code, the more the skills they’ve honed for years deteriorated. This is by now a well studied phenomenon sometimes referred to as "cognitive debt” or "cognitive atrophy.” The idea is that people who use AI to automate certain parts of their job lose the ability to do those tasks well, therefore de-skilling themselves. https://www.404media.co/software-developers-say-ai-is-rotting-their-brains/
-
@bayo@me.dm 2026-05-14 22:33
@shadow53@floss.social Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity https://arxiv.org/pdf/2507.09089
-
@alienghic@timeloop.cafe 2026-05-14 22:47
@shadow53@floss.social Press summary of an ACM study that found that vibe coding introduces a host of maintenance and security issues. https://thenewstack.io/acm-vibe-coding-ai-agent/
-
@Spoofer3@infosec.exchange 2026-05-14 22:52
@shadow53@floss.social https://mastodon.online/@jamesshore/116552564152018598 talks about maintenance costs
-
@julienw@pouet.chapril.org 2026-05-15 07:25
@SmartmanApps@dotnet.social I've seen one had been retracted though @shadow53@floss.social
-
@educoder@mastodon.online 2026-05-15 04:34
@shadow53@floss.social https://www.forbes.com/councils/forbestechcouncil/2026/05/14/the-ai-layoff-bill-is-coming-due-and-ctos-are-going-to-pay-it-twice/ This might be of use to the corporate types, maybe they'll listen to Forbes.
-
@madcollector@retro.pizza 2026-05-15 07:19
@shadow53@floss.social security risks of publishing internal data to a third party, rising costs of tokens
-
@WhombeX@bne.social 2026-05-15 08:40
@shadow53@floss.social about two months back someone tooted on Mastodon about asking your work’s HR department what plan they had to treat staff who fell into AI dependancy/mental health harm. (It isn’t one of the open 30ish Masto tabs in my browser sorry.)
-
@Kichae@tenforward.social 2026-05-15 12:42
@shadow53@floss.social Unfortunately, I don't think most businesses are going to care about environmental harm or skills atrophe. One is an externality, the other is covered by the technology (and is a reason to start paying people less). Businesses speak in dollars, so you need to show that LLM usage is more expensive in the long term. That LLM-generated code bases become more expensive to maintain over time, and that token usage will become unaffordable.