Holy moly EVERYONE needs to read this thread. A really precise and simple analogy of generative AI vs. human learning, and why we're being gaslit when we are told generative AI isn't simply plagiarizing people's work. - ThreadSky

davecobb.com • 13 hours ago

Holy moly EVERYONE needs to read this thread. A really precise and simple analogy of generative AI vs. human learning, and why we're being gaslit when we are told generative AI isn't simply plagiarizing people's work.

Reposted from Oisín McGann

One of the most common arguments you hear from fans of generative ‘AI’ is that it’s not plagiarizing people’s work, it’s just learning like a human learns. So I’m going to break down why that’s just not true, and why it can never be true, with the existing systems. 1/

Comments

davecobb.com•12 hours ago

And, to be transparent, I *have* used AI for creative purposes. But I have two VERY SPECIFIC cases where I find it acceptable.

Short thread... 1/3

davecobb.com•12 hours ago

First, I have used generative AI to create reference images for mood boards, in the same way I would use found images on Google, or pulled from books & magazines. Professionally, these are for inspiration only, are never public-facing, and are never the basis for any final artwork.

2/3

davecobb.com•12 hours ago

Second, I sometimes use ChatGPT as an "idea engine" when writing. I will put in early story prompts & ask things like "give me character conflict examples" or "what would be interesting environments for this scene" -- but final output used for clients or public-facing comes entirely from me.

3/3

mmadventures.bsky.social•11 hours ago

Was a fascinating read, but this isn’t a great way to read it.

rollercoasterphilosophy.com•11 hours ago

This brings me back to learning some philosophy of language and specifically Wittgenstein's idea of language games.

Yes, we use language to model our experience of the world, but it's also based on how we expect social interactions to play out, and we adjust our language based on others' reactions.

rollercoasterphilosophy.com•11 hours ago

If you say "cup" and I'm confused until I say "oh, you mean a mug" then you're probably disappointed that this language didn't have the social action you expected and you're going to use "mug" from then on, at least around me.

rollercoasterphilosophy.com•11 hours ago

It also easily solves the "is a hot dog a sandwich" debate, since it completely sidesteps issues of construction or physical properties between the two, and simply asks "would you call a hot dog a sandwich in mixed company and expect a normal reaction?"

rollercoasterphilosophy.com•11 hours ago

In that way AI does seem a little bit closer to the language game model, but even there it's far more limited because the only social actions it can process are approval or disapproval. It doesn't exist socially in a world where it uses language to influence many possible actions of those around it.

rollercoasterphilosophy.com•11 hours ago

Like obviously large language models don't work anything like human learning if only for the simple fact that the human brain works nothing like a computer chip.

People used to compare brains to steam engines (i.e. let off steam), it's just whatever the most advanced tech of the day happened to be.

davecobb.com•10 hours ago

Thanks for all that. The social-cue aspect of learning and communicating is something that AI *simulates* (or at least they want it to in chat-bots), but there’s always going to be an uncanny valley of sorts

Comments

Posting Rules

Reply