this post was submitted on 27 Aug 2026
37 points (100.0% liked)
technology
24468 readers
251 users here now
On the road to fully automated luxury gay space communism.
Spreading Linux propaganda since 2020
- Ways to run Microsoft/Adobe and more on Linux
- The Ultimate FOSS Guide For Android
- Great libre software on Windows
- Hey you, the lib still using Chrome. Read this post!
Rules:
- 1. Obviously abide by the sitewide code of conduct. Bigotry will be met with an immediate ban
- 2. This community is about technology. Offtopic is permitted as long as it is kept in the comment sections
- 3. Although this is not /c/libre, FOSS related posting is tolerated, and even welcome in the case of effort posts
- 4. We believe technology should be liberating. As such, avoid promoting proprietary and/or bourgeois technology
- 5. Explanatory posts to correct the potential mistakes a comrade made in a post of their own are allowed, as long as they remain respectful
- 6. No crypto (Bitcoin, NFT, etc.) speculation, unless it is purely informative and not too cringe
- 7. Absolutely no tech bro shit. If you have a good opinion of Silicon Valley billionaires please manifest yourself so we can ban you.
founded 6 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I'm very AI-skeptical in terms of the incredibly negative societal impacts of it, but as least in the field that I most associate myself with (software engineering, but I've been unemployed for more than 2 years), they're somewhere between very useful and wildly useful. Most of my friends who are employed in this field haven't written a single like of code by hand in months, and spend at least $1000/month on behalf of their company on tokens (per engineer). It's the single biggest change to the field that has happened in at least 20 years.
I think that a big part of the crazy valuations come from the idea that this will be able to replace lots of labor, as least in the tech sector, if not more broadly in other blue collar sectors. In terms of software engineering, at least, I don't think it works to replace engineers wholesale, but I do think that it can work as a productivity multiplier, where now one engineer can do the work of what used to take five to ten.
(I think the valuations of the AI companies are still completely ridiculous, because they all rely on the idea that one of these companies, i.e. Anthropic or OpenAI, will achieve a monopoly on having the useful model, which, outside of AGI
, will never happen. Firstly, I think that currently they are both top tier models, so one would have to extinguish the other, and secondly, the Chinese AI labs are pretty close behind, or in some cases even better, and they release their weights so there's no lock-in.)
Outside of tech, I'm a lot less sure. Right now, writing software is the main thing they're being trained on, and they're getting better and better with each release. Once they hit a wall with that, these companies might focus their training efforts on other tasks, and it would likely get better at those too. I'm not sure what areas would come next (I'm sure it would be based off of a combination of what they think will make them the most money and what is easiest to get large amounts of training data for), but it has the potential to do a lot of white collar computer jobs in the near-ish future imo.
I agree with a lot of other points in the thread, but I think they do correspond to a pretty substantial shift in the field of tech, which is a big part of the US's economy.
Yes, I hate them for all other purposes but for software it is unreal. As long as you're the centaur, at least.
They can't stick to a plan, none of them can, they're terrible for that. But for short feedback loop work with frequent verification and good test coverage, there's nothing else like it. That's the key difference between software and other domains; you can rigorously test it, and you'll put hands on it before it goes to a customer. If you care even a little bit, the experience doesn't have to be sloppy for the consumer.
I can show opus (4.8 high is the sweet spot) a screenshot of an error message and a problem description and it'll trace back from that error message across four different repos and have a fix ready in under a half an hour, basically every time. Logic that would have taken me hours to trace through by hand on codebases I know really well.
I had a big, nasty, multi-table join database query that badly needed optimization. I told claude to design and build a test database that reproduced the issue and then iterate on query shapes that would get the query planner to do the right thing. Ultimately I unlocked it by suggesting a small schema change, which it implemented and tested almost as fast as I could type it out. 175x improvement in maybe an hour, after dozens of hours of real engineer effort over the years. I just needed a 3D display to wave my hands around and stare pensively at for the full Tony Stark / Jarvis experience.
It's hard to argue with the results; with claude in my hands I have been able to fix everything about the product that has ever bugged me going back a decade. Of course, since we have this tool now, my company let two thirds of the engineers go... given the choice I'd rather just have them back.
Anthropic and OpenAI are both screwed in the medium term though. Fable and Opus 5 are not only too expensive, they're functional regressions in my team's experience. And Kimi is already as good as Opus in real world use. Qwen3.8-27B will run on my laptop and is as good as Sonnet.
I very much doubt that it can be used for things other than software and perhaps physical simulation applications. In software, it either works or it doesn't, and you can build it around working and optimizing certain metrics. Literally just optimization of automation. I can see it coming for alot of ME and data entry jobs.
Once you get out into the more vague aspects of engineering and social science, or even something as basic as factory management, the level and ability to do experimentation and replication studies plummets, which means the advice and solutions that it comes up with are going to be based entirely on what the user thinks work, or based on the 'scientific consensus' is. It is a much more difficult issue to tackle epistemologically, like, what is the best spice to add to broast chicken?
Not super complex, but i think it works well at topics pertaining to language and linguistics (which are closely related to software, computer science, and formal logic anyway)