I am afraid that will be as proprietary as a HP printer cartridge.

…Not that it wouldn’t be fun to try the hack.

…finding LLMs that write half-decent prose?

They don’t. In fact, newer LLMs are regressing on that aspect.

But I am surprised the institutional response has been so slow. This is existential; every school should be moving heaven and earth to do stuff exactly like blue book tests.

All true.

This was happening before “AI” though. It’s just more oil for the fire.

Swathes of the world are going full cyberpunk, for sure.

Evangelizing llms is pretty much all you do on here, you have spent an enormous amount of time and effort on them.

This is a fair point. I talk about it a lot here, too much TBH.

Sorting by New Comments, topics like this are mostly what come up and peak my interest… I suppose I need to curate/moderate my Lemmy feed, like I’ve been telling myself I should do for some time.

I dispute all of this

The entire premise of chatbots is to have a simple natural-language interface instead of those scary GUIs and programming languages.

Absolutely not. This is not what text models were made for, it’s what jerks like Altman and Modi twisted them into with marketing.

The idea that maybe they didn’t know enough about how to interact with ai, and that’s why it failed to help them, is a little silly, and the more you explain how we’re all using ai wrong, the sillier you look.

I’m not blaming the user, I’m saying they’re crap services. I think the superlative/cursing is necessary here; they are so bad and enshittified it is astounding.

You’ve obviously sunk a lot of costs into genai, both in time and money, and that is making you defensive.

I have paid a total of a few bucks into APIs, I’ve never had a Claude or OpenAI subscription in my life. I don’t know why you would assume this.

Somebody that isn’t impressed by Claude premium isn’t going to do any of what you suggest, let alone get any more value out of it, and that’s valid for them. You’re not seriously interested in helping them professionally, you are just insecure.

I’m not insecure about this. I don’t need LLMs or generative ML, though I am interested in it. I have many dire personal issues, but people are free to program however they wish and I’m not going to gaslight anyone about that.

I do think text models have been bastardized into “a simple natural-language interface instead of those scary GUIs and programming languages.” That is NOT what they were intended to do, no matter how much Claude tries to sell people on it. I think this misunderstanding is extremely important because it’s exactly what’s propping up abominations/scams like OpenAI/Anthropic.

But they can still be extremely useful tools in narrower scopes, like all machine learning has for over a decade, and I’m not going to shy away advocating for that. I do resent all of LLMs being pigeonholed so tightly, and if people are going to try them, I at least want to point them to a more reasonable/frugal source than Claude premium and Claude Code.

Some models like Nemotron use open datasets, and utilize hardware more efficiently than OpenAI and such.

Its (IMO) also quite different when the weights are Apache licensed. If absolutely zero profit is involved, that gets close to the “fair use” umbrella, IMO, like other noncommercial derivative works of restrictively licensed content (fanfics/fanart).

You are not wrong about the closed datasets though, for the vast majority of LLMs. Lord knows what the Chinese trainers or OpenAI use to train…

But anything truly “open weights” is still a different order of magnitude in the ethics scale, IMO. I view the primary issue to be profiting off the scraping and training; take that away, and its not nearly as egregious.

I don’t. I’ve never had an OpenAI or Claude subscription. Basically the only thing I’ve used Claude API for is science/testing when comparing it to an open weights model, and I haven’t done that in some time.

I’ve never felt a need to, either.

Not that Im perfectly innocent or anything. Or that you’re wrong. People DO exactly this, and it makes me scratch my head.

I’d wager people like linux devs will learn exactly how unreliable they are, quickly, once they start using them. Its kind of impossible to miss.

There’s a chance some devs succumb to AI psychosis, but that’s why these large open source projects are institutionalized. No one person can ruin them, by design.

As for ethics, I think the open weights LLMs are far more ethical, albeit far from perfect. They’re trained modestly, presented as tools, apache licensed, self host able, more efficient to run, and good enough; people just need to know they exist as an alternative.

Well that’s because copilot and cursor are pure shit, and claude code is overpriced and over rated.

TBH I have been hopping between agent frameworks, but mostly I use locally run Qwen or Deepseek to draft individual function blocks or brainstorm in a notepad interface:

https://github.com/lmg-anon/mikupad

And manually insert them as I review the code… I like my stuff uncomplicated and unobstructed.

If I were to pick an API provider today, I’d buy one year of Xiaomi MiMo, because it is dirt cheap and good enough. GLM and Deepseek are great, but more expensive since they are popular now.

I think the people most extreme about it haven’t done much development.

Or they have, professionally, and associate it with very real AI psychosis at work.

Or they’re in school and see LLMs ruining the student body, as we’ve seen some in other Lemmy posts.

It all comes down to ethical tool sourcing, and using those tools in moderation, responsibly. Like all sorts of other automation/technology that’s popped up in programming…

But I guess I’m saying I can sympathize with absolute “no-LLM-code” positions, too. The way true vibe-coding has been peddled to the public really has done tremendous damage, and honestly the space deserves the negative attention its getting, and then some.

In practice, I found it breaks sites more, lets some things through, triggers more adblock detectors, and doesn’t work as well with the annoyance list.

It doesn’t work on YouTube. And the maximum number of entries to block is limited.

Its not catastrophic though.

So UBO Lite is not wholly ineffective, no. It works great, mostly. Buts its definitely more of a nuisance to use, which is exactly what Google wants.

It won’t for long. Its only because Ungoogled Chromium is on a stable channel, but this removal will catch up to them soon.

This goes for all Chromium browsers, except those that hack in their own adblock engine like Cromite or Brave.

Well, it anonymizes you by “faking” the tracking. I would hope it takes steps to sandbox the ads.

1 million users…

You didn’t answer my question.

What does rosetta have to do with any of this?

Intel must be kicking themselves for not using their own fabs for Arc, AND divesting from Micron, Optane, and HMC.

Apple and others must be too, for snubbing Intel and Samsung.

They would be making money hand over fist, and relieving the market at the same time.

No, @GamingChairModel@lemmy.world is right.

What does Rosetta have anything to do with what OpenAI needs? Machine learning on Macs is done in MLX or something else with Metal, running OSX. Most agenic frameworks or other ML-related infrastructure are just Python, or open code that can compile on OSX/Macs just fine. In fact, OSX is often the first target platform.

This even includes inter-node frameworks for clustering Macs together.

No one doing “AI” on Macs is using Rosetta. There’s no point, and its a bad idea. They’re doing everything in OSX, which works fine, and I can’t imagine what OpenAI would be doing with them that wouldn’t work in OSX.

But maybe I’m missing what you’re implying here. And I’m sorry to come off as thorny, but condescending comments like “I’m not here to teach a computer science class” rub me the wrong way.

Well I was thinking the opposite.

Take the existing community fragments, and unify them into a structure so they don’t wither.

The problem I see on Lemmy is that niche communities aren’t very discoverable, hence never cultivate much interaction, but “percolating” posts up would address this.

Lemmy and Piefed need a “subcommunity” structure to get that ball rolling IMO.

EG you may have an overarching “sports” community, and in layers:

sports -> basketball -> NBA -> team

• So if your team sub was kinda dead, you could browse /c/NBA or /c/basketball and get a wider feed of other teams.

• Similarly, you could mark non-spammy posts to “filter up” to the higher communities that accept it. So you make a post in your team community, and folks browsing basketball might see it, depending on their settings.

Permissions would be granular to set this up, consensual for both the “upper” and “lower” communities.

I’ve written this up as a more general proposal ti help niche communities, but have yet to put it on the development gits as a feature request.