Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Generative AI. "Slop Generators, are unsuitable for use [ ]"

asahilinux.org

Post reply on HN
Print view View original

41–50 of 92 posts

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#41

Post by mpalmer »

vrighter wrote: post #36
simianwords wrote: post #16

Earlier quoted context omitted.

>First of all, it isn't wrong. The power consumption of AI training and inference is massive. This is massively overstated. We ought to be more careful in performing such calculations.

No it is not, honestly. The argument that inference is cheaper than training does not hold up when you keep in mind that as the current model is being used for inference, another one at least as big is being trained simultaneously. Current models can't adapt to new information. So training is an ongoing thing and inference comes in addition to the training costs, not instead of.

This doesn't refute the (unmade) argument that inference is cheaper than training. Inference is cheaper than training, regardless of how much of each is taking place.

Not only that, both keep getting less expensive in various ways. small models, though still not cheap to train, are making inference vanishingly cheap.

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#42

Post by RUnconcerned »

sandworm101 wrote: post #18

> "These resources are better used on quite literally anything else." What shocks me most is that we have found something less useful than bitcoin mining. Remember all the articles about the environmental impact of bitcoin? That is peanuts compared to what the worlds largest companies are biulding to power the next LLM.

While I agree that the value of LLMs is wildly overstated, I disagree that it is less useful than bitcoin mining, which is entirely useless. At least LLMs can produce usable output.

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#43

Post by mpalmer »

roxolotl wrote: post #40
harringdev wrote: post #25

This is very hard to take seriously, it feels ideologically driven. The introduction where they claim LLMs are useless for software engineering is just incorrect. They are useful for many software engineering tasks. I do think that vibe coding is rubbish however, and more junior SWEs very regularly misuse LLMs to produce nonsense code. The only substantive point is that the LLM may regurgitate pieces of proprietary t…

Why is something ideologically driven inherently not worth taking seriously? I appreciate maybe their understanding will be more challenging to change but people don’t necessarily arrive at ideologies for unserious reasons nor do ideologies have unserious results.

I think the better word is extremist, because choosing to brand all LLMs as "Slop Generators" is just that, an extremist position with a rather self-satisfied air about it.

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#44

Post by JimDabell »

mschuster91 wrote: post #24
simianwords wrote: post #16

Earlier quoted context omitted.

>First of all, it isn't wrong. The power consumption of AI training and inference is massive. This is massively overstated. We ought to be more careful in performing such calculations.

To quote [1]: > Alex de Vries is a Ph.D. candidate at VU Amsterdam and founder of the digital-sustainability blog Digiconomist. In a report published earlier this month in Joule, de Vries has analyzed trends in AI energy use. He predicted that current AI technology could be on track to annually consume as much electricity as the entire country of Ireland (29.3 terawatt-hours per year). That's a lot of power, and the…

> He predicted that current AI technology could be on track to annually consume as much electricity as the entire country of Ireland (29.3 terawatt-hours per year).

He didn’t though, did he? If you click through and read the report, it is quite clear in saying that this is a worst case scenario if you ignore things like it being unrealistic for Nvidia to even produce the requisite number of chips. It’s also based on the efficiency of 2023 models; he assumes 3 Wh per prompt when today it is 0.34 Wh.

It also completely ignores the fact that AI use can displace greater energy use:

The carbon emissions of writing and illustrating are lower for AI than for humans

> Our findings reveal that AI systems emit between 130 and 1500 times less CO2e per page of text generated compared to human writers, while AI illustration systems emit between 310 and 2900 times less CO2e per image than their human counterparts. […] the use of AI holds the potential to carry out several major activities at much lower emission levels than can humans.

— https://www.nature.com/articles/s41598-024-54271-x

It isn’t a bad thing if AI uses x Wh if not using AI uses 1500x Wh. Looking at the absolute usage without considering displacement only gives you half the picture.

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#46

Post by harringdev »

roxolotl wrote: post #40
harringdev wrote: post #25

This is very hard to take seriously, it feels ideologically driven. The introduction where they claim LLMs are useless for software engineering is just incorrect. They are useful for many software engineering tasks. I do think that vibe coding is rubbish however, and more junior SWEs very regularly misuse LLMs to produce nonsense code. The only substantive point is that the LLM may regurgitate pieces of proprietary t…

Why is something ideologically driven inherently not worth taking seriously? I appreciate maybe their understanding will be more challenging to change but people don’t necessarily arrive at ideologies for unserious reasons nor do ideologies have unserious results.

You make a good point, and I will clarify what I mean.

A small disclaimer, I am not a AI-booster. I think LLMs do have issues, and one should be careful with them.

I've found that there's a large group of people who dislike LLMs strongly and claim they are totally useless or even pernicious. I think this is grounded in truth -- they are trained on copyright work, they are used for spreading misinformation, they can produce sh^t code. Although some folks take a radical/extremist approach and totally dismiss them as useless -- often without actually using LLM-powered tools in any meaningful way.

They are useless for many applications, but programming is not one of them. I think a blanket ban on LLMs has to be somewhat unfounded/radical because they do have practical applications in writing code. The tab-completion models are extremely useful for example.

For this more niche project I would think LLMs might not be as useful as they are for other projects. This said, I still think there would be a variety of use-cases where they could be helpful.

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#47

Post by krapp »

lucumo wrote: post #37
terminalbraid wrote: post #27

Earlier quoted context omitted.

Luddites in what sense? Because there's the lazy, gross stereotype that's been perpetuated with negative connotation and the movement that was not at face anti-technology, but against how that technology was used to suppress labor. https://en.wikipedia.org/wiki/Luddite

The historical luddites were not heroes of the working class. They were a special interest that used violence to keep less skilled people out of their industry, so they could keep making more money than the hoi pollo. Dumb or evil, either way they are not people to be celebrated.

>They were a special interest that used violence to keep less skilled people out of their industry, so they could keep making more money than the hoi pollo.

So apart from the use of violence, the Luddites were essentially equivalent to modern tech culture. It's weird that they don't get more sympathy on HN.

Regardless, the complaints of the Luddites resonate because they would eventually apply to everyone, including the working class. It just happens that they came for the textile workers first.

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#49

Post by washadjeffmad »

Fade_Dance wrote: post #2

>Slop Generator Using that term through the article makes it hard to take seriously. I know nothing of this project but right off the bat it seems like a project that has little credibility just because of the tone used throughout. There's no need to turn it into a full-on tirade against this set of technologies either. Is this an appropriate place to complain about Reddit comments? Ironically, the author could well…

That mindset is itself part of the problem. A healthy sign of free speech is poor taste.

Re: Generative AI. "Slop Generators, are unsuitable for use [ ]"

#50

Post by cadamsdotcom »

Agreed on the IP point. Strongly agreed on not wanting to go to court in the current US political climate.

On the rest, especially the confidently incorrect argument.. Not. so. much.

Firstly, models are stochastic parrots, but that truth is irrelevant because they're useful stochastic parrots.

Second, hallucinations and confidently incorrect outputs may yet be a thing of the past, so we should keep an open mind. It's possible that mechanistic interpretability research (a fancy term for "understanding what the model is thinking as it produces your response") will allow the UI to warn the user that uncertainty was detected in the model's response.

Unfortunately none of that matters because the IP point is a blocker. Bummer.

Post reply on HN