Castigation is your right, but understanding shows nuance. Why do you think some people use it? Beyond the glaringly obvious of 'well it's easy and I don't have to try hard?'
Alt-text. A new hotel room with a touch-screen remote on the wall.
In that same hotel room, 'These bottles all feel the same. Which one is shower gell, shampoo and conditioner?'

Blind people can ask these questions of AI at 3 in the afternoon or 3 in the morning without having to bother others.
It uses LLM behind the scenes to describe that picture, and you seriously want to tell me that's so terrible?
universeodon.com/@FreakyFwoof/…

Andre Louis

2025-07-21 20:30:37

People: 'I hate AI!'
Also people: 'Image with no description'
Know what that means? one of two things in my case.
1. I ignore your post.
Or
2... I use #AI to check out your post. How'bout that?

Can we get a t-shirt that says 'Always add #AltText'

An interesting point.

I would say that I made a request, I didn't castigate any reader who isn't a broligarchy scam artist.

As for alt-text? I say it here several times a week: "Alt-text or it didn't happen."

I don't read or fave or boost posts that have images w/o alt-text, being not blind but severely vision-impaired myself.

If someone's indifferent to whether I can parse their post, I'm indifferent to parsing their post.

(1/x)

Now, having said that. The hotel room case is an interesting one, and I will have to think that over.

So far, I have not experienced sufficient trust in LLM output to distinguish those three bottles.

But I wonder, if I *did* experience that trust, if I believed the LLMs were reliable, would I, even then, pay the monstrous costs to society for that service to me?

I can't say no, truly. But I can wonder.

(2/x)

Note, too, that using LLM output in that hotel case is *not* actually doing what I said I wished people did not do.

(3/3)

It's still using it though, even if it's not posting.
Virtual trees fall in virtual forests all the time with people taking pictures (and I am happily one of them) but we don't post the text it outputs for various reasons.
It's still happening.

I have been in that precise situation, and have trusted the LLM to get the bottles right. I later verified with a human and yes, both agreed.

You have the kneejerk reaction (which I honestly do understand) of 'All LLM = bad, evil, drinks more water than a thursty camel in the desert' but from my point of view, I emphatically think this is just wrong.

I am heavily biased of course, and I make no bounds about that, but not all LLM is bad, and not all starves the environment of resources...
I run local LLM daily on my Mac, so I'm using *my* resources. No more than I would be if gaming under-load. Think about that...

And now we're done. My post, however mistaken it might have been, is not a "kneejerk" reaction.

Bye!

"Kneejerk" implies thoughtlessness. It costs little to treat a stranger on the Internet with respect.

Also, posting anything critical about LLMs generally gets you a deluge of personal insults, implying that the reason you don't like LLMs is some personal failing of your own.

Given that all the people behind AIs are promising to destroy all jobs, and fear that it will destroy all humans, it hardly seems unreasonable to worry...

techradar.com/ai-platforms-ass…

OpenAI's CEO says he's scared of GPT-5

Not that it's slowing down development

^{Eric Hal Schwartz (TechRadar)}

In the real world, you don't say something like "I wish people wouldn't post AI content" and then have all these strangers suddenly appear.

My criteria for blocking are high - generally, the person either has to say something that would start a fight in a bar, or tell gross lies ("COVID is a hoax, the climate crisis is a hoax", etc) in an aggressive fashion.

As a cryptocurrency and AI skeptic, I've gotten an awful lot of rudeness.

1/

(And I have worked professionally as a programmer in both areas, it's not like I'm ignorant.)

In both cases, crypto and AI, I feel that both the technology is flawed, and the consequent costs to almost everyone but 0.1% richest humans are incredibly high and do not outweigh the value.

Given that we needed to decrease our CO2 90% by now, and instead it continues to increase exponentially, both technologies feel like ecological suicide, too.

AI image describers, particularly for helping differently abled people, are not what people are against.

It's the LLMs. Those are the ones stealing our work; those are where nearly all the computational power goes; those are what is supposed to take our jobs; with the hallucinations (theirs), and the psychosis (ours).

@melioristicmarie

Out of curiosity, naive question for knowledgeable people :
Is there any device that can translate a perceived 3d image into a tactile representation, with slimmed down 3d on a panel, applied on a densely perceptive part of the body for some kind of instant perception, and with high enough definition or zoom capability to allow perceiving written language with more finger inspection on the other side. It would spare the need to translate visuals in braille, help perceive shapes and designs from a distance, with an optional audio output, and which one would master with some training ?

If it exists is it somewhat affordable ?
if you have a set of spare batteries to swap when needed it could be environmental cost effective and pretty flexible anywhere ?
I always thought about something like that, i wonder if it exists or if there's something similar providing somewhat of an instantaneous tactile view to blind people ?

Whether it exists or not, would you think it could be a good tool or do you think of obvious caveats right away that would make it useless or unpractical in your experience ?

No. People try this for 2d, it doesn't work very well, and it's "new car" levels of expensive. 3d is sort of easier ish, because many things can be done with 3d printers. Those are mass produced, so companies can recoup their development costs among many more units, which makes them somewhat affordable.

They're still pretty finnicky and nowhere near instantenous. For a quick "what is this thing" question, LLMs are the best.

Yes they hallucinate, but especially if you know that the thing you're looking at is one of a few options, but you just don't know which one, they're extremely useful.