Skip to main content

Turns out that Techbros are tired of *checks notes* _ 100% accurate descriptions_ of parrottech.

Don't take my word for it. Trust world experts like Timnit Gebru at DAIR, a video this guy will never watch.

youtu.be/P7XT4TWLzJw

#AI #AIEthics

toot.cafe/@matt/11158501207023…
@matt - I'm getting tired of...characterizations of generative AI like this one: social.ericwbailey.website/@er… "a spicy autocomplete powered by theft that melts the environment to amplify racism and periodically, arbitrarily lie"...


SaTML 2023 - Timnit Gebru - Eugenics and the Promise of Utopia through AGI

Eugenics and the Promise of Utopia through Artificial General IntelligenceBased on work by Timnit Gebru & Émile P. Torres
YouTube
#AI #aiethics @Matt Campbell
in reply to PJ Coffey

Matt Campbell
Matt Campbell
• •

Thank you for your response to my rant yesterday. I've listened to the presentation by Timnit Gebru that you linked. I'm convinced by all of her points about the problems with LLMs, and the idea of AGI.

Still, I hope it's possible to have LLMs and similar models (such as GPT-4V) without the serious problems that critics such as her have pointed out, because such tools have already been shown to be useful to at least one marginalized group: disabled people.

Matt Campbell reshared this.

in reply to Matt Campbell

Matt Campbell
Matt Campbell
• •
I know that one of your interests is accessibility. Same here; I'm legally blind. But I'm not optimistic that we can convince the whole world to go out of their way to accommodate us. Automation is the only thing that can solve the problem comprehensively. And while current tools like GPT-4V aren't perfect, and there's a lot of snake oil (the accessibility overlays), I disagree with dismissing the idea entirely, because some tools that are called AI are already useful.
in reply to Matt Campbell

PJ Coffey
PJ Coffey
• •

I don't have a problem with AI that would help. E.g. autocorrect is AI, red eye reduction is AI.

These narrow intelligences (ANI) are not what is currently meant by AI (properly AGI) which as you agree can't be scoped and therefore does not exist

I am glad that it is helping accessibility for blind people but the quality of information is very very dubious unless you're already a subject matter expert.
Eg tabletopden.com/dnd/best-armor… should in theory be great but it's actually mostly tosh.


The 15 Best Armor For Rogues in D&D 5e [Ranked]

Rogues deal a ton of damage, meaning their armor is their first and last line of defense if they don’t have magical protection cast on them.
Sarah King (Tabletop Den)
in reply to PJ Coffey

PJ Coffey
PJ Coffey
• •

@weirdwriter

And I'm 100% sure that any help you're getting is accidental rather than a priority of the designers. They are the same grifters that pivoted from crypto and NFTs and are supported by the eugenics movement!

AI in that sense scrapes creative work without permission or recompense.

I'm a big believer in the cut kerb effect where alt text has actually helped me personally out several times on the fedi. By explaining the point of a diagram or when I had bad phone signal.

@Robert Kingett backup
in reply to PJ Coffey

Matt Campbell
Matt Campbell
• •
It doesn't matter to me if the developers of tools like GPT-4V actually intend to help us, or even if they think we really shouldn't exist (eugenics). In my opinion, we disabled people have to take advantage of whatever useful tools are available to us. Better still if we can develop our own tools to solve our accessibility problems, possibly building on top of things like GPT-4V.
in reply to Matt Campbell

James Scholes
James Scholes
• •
@matt Shrug. Setting out to intentionally cause offence by calling you a "tech bro", assuming your gender, and making unfounded accusations of unenlightenment are some of the exact behaviours causing this conversation to be so combative. You give airtime to, and take your opinions from, whomever you please, but that's not the way I'd personally go about recommending a YouTube video to someone.
This entry was edited (11 months ago)
⇧