[ home / rules / faq / search ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM / ufo / 420 ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]

/leftypol/ - Leftist Politically Incorrect

"The anons of the past have only shitposted on the Internet about the world, in various ways. The point, however, is to change it."
Name
Options
Subject
Comment
Flag
File
Embed
Password(For file deletion.)

Check out our new store at shop.leftypol.org!


File: 1795321690334-7.png (48.15 KB, 656x531, Untiwadawdawtled.png)

 

what is it about AI that causes people to act like this?

anytime ive seen it come up you'll see a certain subset of people act like being anti-ai is a means to itself, no compromise no surrender, no matter complete seperated from any reason.

Ai generation makes me feel like nothing happens and that I should just give up in life and why should anyone try, idk it’s like extreme alienation

>>2959682
do NOT ask ChatGPT "what hardware do I need to train an open source 7B parameter LLM, and how much electricity would this use, and how long would it take?"

>>2959686
why would anyone train one from scratch instead of just running deepseek locally through ollama or some shit

>>2959690
to prove the model hasn't been tampered with or censored.

>>2959682
>what is it about AI that causes people to act like this?
People are afraid of change so its like kneejerk reaction. Give these people couple of years and they will use AI too. It just takes time for them to adjust.

>>2959682
I think it started of as peyit bourgeois schizophrenia from artists getting competition and spread off from there

>>2959846
European copium about having no AI of their own seeping through into American internets

>>2959850
EU has Mistral and many other smaller companies though

>>2959682
AI is the automation of cognitive labor: it removes the economic incentive to create intelligent human laborers, because intelligence is now a commodity that can be bought.
Therefore the result of the existence of AI is the lowering of the cognitive ability of humanity through the removal of the economic incentive of teaching humans to be intelligent.

>>2959682
Luddites that are afraid of quality changes

File: 1795356897384-1.png (659.89 KB, 1042x789, 1718224442071.png)

>>2959682
You remember the the ways of citing stuff before LLMs? Yeah, it's actually what we did. But it was very useful by scrollin through books to find infomation at things. Now, it's like viewed as a ”thing in the past”. Now, we got this bull-shit-fuck thing called ”AI Chatbots”. Now you put the word "Chatbot" at the end of anything, and it's sure to suck.

>>2959849
This. The astroturfed “backlash” to AI began pretty much exclusively as a movement of the first world labor aristocracy, especially among those that sell disgusting fetish material for a living. Now their “art” can be made for free and if higher quality and they actually have to get a real job

File: 1795357909322-1.png (2.03 MB, 1448x1086, snailcat ai toboggan.png)

To be fair, AI proselytism is just as annoying, if not worse.
Look at the Riemann Hypothesis breakthrough: nobody so far can understand the 170,000 lines of Lean code used to prove it.
What's the point? Where are we heading?

I've been experimenting with vibe coding, and it's genuinely impressive how it can turn a scientific paper with complex linear algebra into Rust code in half-an-hour.
It's also pretty cool to see all the firmware hacking of hardware it has enabled suddenly.
But now, the job of a software engineer is to babysit an AI model. All the parts of the job that was somewhat of an art-form, like crafting elegant code, are now gone.

AI also resulted in massive lay-offs of workers, getting hired is an absolute bitch, and the only answer we get is "learn a trade". Not even the communist left is talking about reducing labor time as a way to mitigate the societal problems, we still act like we will have to work harder than ever to electrify 1917 Russia. Actually, it's even worse than that, it's "hahaha you deserve it petit-bourgeois labor aristocrat", schadenfreude that just serve to boost your own ego and don't think about any implication at large.

And speaking of electricity, the training of AI is consuming a ton of resources, and there is no halting, we don't even take time to explore the possibilities of the current models, we get one model after another, and these models have flaws like talking to you like a McKinsey consultant who make sure you think you are never done with anything, so you keep paying and using more and more tokens.

Everybody wants to play with the cool toys, nobody want to think about the future implications and reform anything. And most of what is produced is slop.
I can understand being reflexively all against AI, even if it's stupid. I will use it anyway, but I prefer people who are skeptical than naive optimists.

>>2959683
this is a capitalism problem not an AI problem IMO. AI and machines could liberate humanity from all of the dreadful labor while leaving time for the fun/enjoyable stuff. Every human being could live the life of the leisure class today if control over AI was equally distributed.

do you stop playing sports because a machine can move faster than you? no
do you stop playing video games because a computer can aim better than you? no

We need to destroy these steam machines or else they will take our jobs.

>>2959921
>>2959921
I think there will be serious efforts to condense the RH down to a human-understandable format.

(Or we will have to take some steps to make humanity smarter so we can follow along with the machines)

>>2959921
>we don't even take time to explore the possibilities of the current models
I don't disagree with your position that skepticism is warranted a lot of the time but I think this is wrong. There is an entire micro industry/discipline (evals) based around exploring the possibilities of current models.

>>2959921
I agree with most of your post but want to push back on this:
>All the parts of the job that was somewhat of an art-form, like crafting elegant code, are now gone.
Employers were never hiring software engineers to write elegant code. They were hiring them to produce a working product. Imo, elegant code is simple, readable, and maintainable code. Clever one liners and abstractions nobody else can follow were always a problem. Good code is code the next person can pick up six months later without having to reverse engineer what you were thinking.

If you let an AI agent go wild on a complex feature without any constraints, rules, or guidelines, you sometimes get the exact opposite: duplicated logic, weird abstractions, five different patterns for the same thing. It does work, though. Imo that's more the fault of the dev who gave it the task that way. Give an inexperienced junior dev the same task with no design doc, no conventions, and no code review, you'd get pretty much the same result, because they don't have the experience yet to know what could trip em up.

The agent mostly mirrors what you give it. If the codebase is consistent, the rules files are clear, and the feature is broken into well-defined pieces with defined interfaces, the output tends to follow those patterns. If the codebase is a spaghetti and the prompt is "build X, Y, Z" it'll pile more shit onto the mess and produce more spaghetti.

I don't think the craft is gone but moved. The skill now is in deciding how a feature should be structured before any code is written, defining guardrails, knowing what not to build, and reviewing diffs to catch when something works, but might be a maintenance problem later. Imo typing code was never the difficult part, it was maintaining and avoiding tech debt, fixing old shit, etc.

I'm a dev that wears a lot of hats and it's been invaluable for me. Reviewing old codebases, finding security issues, debugging, writing one off scripts for maintenance, improving sql queries, etc. In my off time it's allowed me to move a lot faster in personal projects. I'm pretty much in agreement with how Linus views AI and LLMs

>>2959945
NTA. Please point me to some directions on how to define guidelines, architecture, and how to code review AI output efficiently. I feel like I'm just slopping my way through my job.

>>2959921
>AI also resulted in massive lay-offs of workers, getting hired is an absolute bitch, and the only answer we get is "learn a trade".

That's normal capitalism, though. You people are angry at AI for what is capitalism's fault. Like, what, do you seriously expect normal ass capitalist behaviour to not happen under capitalism? Do you actually believe in "Europe is a rich socdem lad where everyone is happy despite living under capitalsim" nonsense?

>>2959950
Architecture is still on you imo so it's worth it to learn different patterns and approaches. Figure out the structure and interfaces 1st, then hand the agent small chunks, not giant whole features. Domain driven design for example is worth learning for this, since agents follow those patterns pretty well once in the codebase. For process look into Spec Driven Design (something like github.com/github/spec-kit). Basically you write the specs, review the plan, let it build one task at a time.
Obviously first step put an agents.md or claude.md or whatever in your project root and they're read automatically. Keep it short and simple. build+test commands, naming conventions, "new services follow X pattern".
Every time you correct the agent on the same thing twice, add a line, put in memory. github/awesome-copilot has decent examples to steal from.
Test driven design works well with agents too. Get the tests written first, then have it code until they pass so you're not just trusting whatever when it says it's done.
Also check the plan before it writes anything. Keep diffs small commits small. Let existing tooling like linters, analyzers, and tests catch the dumb stuff so you're only looking at logic and design. Watch for duplicated code, random abstractions, and changes it wasn't asked to make. And if you don't understand something, make it explain or bin it. Don't merge anything you can't defend or explain yourself or in a PR review but learning to review stuff basically comes from experience.



[Return][Go to top] [Catalog] | [Home][Post a Reply]
Delete Post [ ]
[ home / rules / faq / search ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM / ufo / 420 ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]