RE: https://mastodon.social/@Gargron/116206874904164848

I always bring up auto-captioning as a (non-LLM) technology that is probably a net-win for accessibility, but that still led to Youtube ending up with *WORSE* closed captioning, because as soon as they got machine transcription, they dropped community transcription options.

Machine transcription on Youtube led to a local maximum where community options that often resulted in better transcriptions are now unavailable and everything is kind of meh.

I have also mildly criticized Mastodon over this with its AI alt text features. They're not an LLM, and they're probably good for accessibility, but their existence has hobbled alt text on the platform.

We could have better community alt text controls on Mastodon, and I think part of the reason we don't is because of a mentality that accessibility is just like a chore we have to solve.

And I think the AI stuff helps reinforce that, a bit.

I don't want the OCR stuff removed! But I do think that like.. there's a warning here, and it comes up every time we try to automate accessibility.

Accessibility is a community effort, and tools that can alleviate the burden are helpful, but are not a substitute for human interaction and effort.

And many accessibility spaces are under-developed *because of these tools*. We could have a "boost with alt text" option on Mastodon.

We could have official ways (not just replies/comments) to suggest and accept alt text from community members with the press of a button.

But where LLMs are concerned, I see the same thing playing out in that space. I have seen LLM evangelists argue that this is blind accessibility, that this is helping non-English community members, and like.. on the edges, sure! I guess?

But...

I also see a lot of members of those communities saying that things are getting worse. That blind accessibility has not gotten better with the rise of LLMs. That the net effect of these tools is a *less* accessible Internet.

And that is very plausible to me, I believe that, because I've seen it play out before with technology that isn't even as harmful as LLMs are.

Like, who could object to AI voice transcription? And yet, the result isn't simply good.

@foxyoreos Mastodon does not have an AI alt text feature. You may be thinking of some third party apps.

@Gargron is the OCR stuff not merged into the main platform?

Apologies if I'm misunderstanding the feature. And apologies if I'm making it sound like it's the same thing as like "auto describe this image", which is a *much* heavier AI use.

@Gargron also fuck just realized the "AI Bullshit " label on this that I use as a standard CW for AI stuff probably makes this seem like I'm disagreeing with you.

You are 1000% right on the money with this, it's a good take, I agree with it.

@Gargron @foxyoreos if neural network-based OCR is not example of AI, then what does this word even mean?
@mkljczk @foxyoreos AI is a marketing term and functionally useless. It means whatever big tech wants it to mean. I studied neural networks at university back in 2013, before there was even a whiff of a large language model. OCR was described in books from 2001.

@Gargron @mkljczk I do think there's value in distinguishing between OCR, a technology almost as old as I am, and full-on image transcription.

I'm used to using the term AI for both, but I don't have any objection to anypony else using a different term and I understand why Gargon would want to distinguish them.

@foxyoreos @Gargron I'm fundamentally opposed to defining 'AI' as 'whatever big tech declares to be AI'

@mkljczk @Gargron I mean, I don't have a problem with either of these positions. I'm used to using "AI" for both as well. I also feel the frustration of trying to distinguish technologies.

I do think there's a difference between LLMs and OCR regardless of the umbrella term, and I understand wanting to make that difference clear. People/critters can use the word "AI" or not, I don't have strong opinions about a specific word.

@Gargron @mkljczk bringing up OCR in the context of Mastodon is less about "oh even Mastodon uses AI" and more about "if I have mild concerns about the accessibility aspects of a technology as simple as OCR, that's nothing compared to the potential downsides of a technology being hyped by LLM evangelists as THE answer to accessibility."