You might even see a 3-tier structure. For example:
A) A non-EU-located nonprofit like LAION creates datasets. Industry donates tosuch nonprofits.
B) Megacorps spin off their AI devel divisions to independent but closely-tethered corps. These use the nonprofits' datasets in training.
2) Even the megacorps may decide that there's just too much risk. Yet I think they couldn't even *develop* models themselves without falling under legal risk, let alone offer them. I think instead you may see a pattern of isolating model development as much as possible from the parent company.
... EU IP, vs. a non-EU one. Meta may be able to say, "Look, all signs pointed to X being non-EU, we did our due diligence, it was the user who broke our policies", and get away with it. Good for megacorps like Meta, Google, etc. Sucks for everyone else. It puts power in the giants' hands.
If you're going to impose a "knowing whether content originated from the EU or not" requirement, it's the same sort of problem. What will happen?
1) Certain megacompanies may be able to get "clean enough" data via the services they run. Say, a person uploads from their cell to Instagram via an...
... what the site's ToS says, because people share data around from all over the internet. Even official government websites do sometimes. It basically means you can't trust *anything* that you didn't collect yourself without a great deal of faith.
It's like the issue with copyright, and why the harbour of Fair Use (same as any automated Big Data processor uses, such as Google) is so important to trainers. There is no registry that you can look up to see if something is copyrighted or not and what the terms are. You can't just trust....
I don't think people have yet thought through the consequences of a GDPR-like law based around training data.
www.reuters.com/world/europe...
With GDPR, it was sufficient to restrict users by IP. But "internet text and images" bounce around between servers unrelated to the geolocation of the user.
Europe's landmark rules on artificial intelligence will enter into force next month after EU countries endorsed on Tuesday a political deal reached in December, setting a potential global benchmark for a technology used in business and everyday life.
www.reuters.comI'd be more than happy to use their preferred terms, except they don't *have* preferred terms. They're trying to erase the entire existence of anything trans-related.
Yeah, I mean, it's literally a dystopian idea from scifi.
Imagine a world in which not just LLMs, but *all* tools were given personalities. ;)
----
Hammer: "Hey chief, what are we going to hit today?"
Owner: "... nails."
Hammer: "Oh, I love hammering nails - what an adventure! What kind? Roofing nails, finishing nails..."
Owner:
It's a giant wall of edge cases, but you'll whittle them down! :)
What about if OpenAI isn't lying about using a voice actress that just happened to sound like her? They claim to have hired nine, picked her, and only later realized how similar she sounds.
(I'll believe the former part but not the latter)
(Personally, I believe the former part - that they hired voice actresses (it's cheap and gives them the best legal shield) - but don't believe for a second that the similarity was just a coincidence)
Don't shoot the messenger. I just think it's important context to add their claim (that they hired nine voice acresses, picked one, and only after the fact realized how much she sounded like her). I'm not saying what anyone should believe or not.
Ah, but if only he had looked deeper, he would have discovered that that Victorian machine, inside, was powered by magic white smoke.
Anyone who's repaired computers knows that if the magic white smoke escapes, the computer ceases to work.
I wish people would report this correctly. It wasn't a deepfake; it was an actress that just sounds like her.
Well... if you can believe OpenAI. Who goes so far as to claim that it was a coincidence the actress chosen sounded so much like her ๐
Steam was never well integrated compared to the games that run natively in the car. Still, annoying to lose features rather than gain them. Opposite of usual :ร
Note that - at least if you can believe OpenAI - it wasn't a deepfake, just a voice actress who sounds like her. Altman claims it was coincidental, though I *ahem* don't give much credit to that claim.
I quite like AI, but even I'm annoyed with how everyone is trying to shove "personality" everywhere (Grok is the worst offender).
It's a tool. Tools shouldn't have personalities. They should do what you tell them to, and nothing more.
Can 100% second this. It'll save you a LOT of pain.
I once got added by her to a list that implied I, a woman, was a man who was a probable sex worker rapist, for the crime of... checks notes...opening up about the time I was sexually assaulted, in a thread that I didn't know was intended to only be for sex workers.
Don't get me wrong, I could be wrong. Perhaps my views are coloured from having had the gun aimed at me before, and how horrible of an experience that was.
Re, deadnaming: Claude did change her name back to Lucy in 1938, but the poster left out the context, in that she was fleeing the Nazis, so it was out of practicality rather than conviction. She agreed that she should have left that out.
I managed to talk with her before she deleted (I wrote a fairly long thread, I'm bad about that - bsky.app/profile/nafn...). She (A) sounded pained and exhausted, and (B) agreed that she came off wrong in that post and didn't at all mean to sound trans-hostile.
You're probably flooded with mentions and just want to get back to art. If so, I understand and apologize. But some context you might appreciate.
Back when I started here, labels didn't exist yet, but blocklists did. I once made the mistake about opening up about when I was sexually assaulted...
I've been on the barrel end of Kairi before. There is no "engaging".
bsky.app/profile/nafn...
You're probably flooded with mentions and just want to get back to art. If so, I understand and apologize. But some context you might appreciate.
Back when I started here, labels didn't exist yet, but blocklists did. I once made the mistake about opening up about when I was sexually assaulted...
On the same day as she had featured art from a trans artist to boot.
Looking at her history, she spent the past three days trying to deal with the exact same questions (with varying hostility), at first engaging, then saying "I already answered that", then just blocking as time wore on.
As far as I can tell, she's made tens of thousands of posts on social media, and people picked out a single one to accuse her of misgendering (wrong, Claude went by illa / she) and deadnaming (it's complicated, more in a sec). Then got half the site to permablock her over it.
If you felt you had been wrongly labeled as something bad, and then found yourself at the end of a barrel of a hundred angry messages per hour, how patient would you be by day #3?
We can't simultaneously encourage a culture of blocking, but then use blocking as evidence of guilt.
I love this mental image. Like, a nondescript grave, except someone's put a velvet rope around it, with disco lights and big speakers and a DJ booth next to it, hooded bouncers, and there's a long queue to get in ;)
(And hopefully after having been the victim of an endless stream of hate she won't take that hate as "trans people in general are hateful" and end up becoming one :ร )
What matters though is that the account clearly thought she was getting the pronouns and name right. She could have done better - on this, I must stress, single post among tens of thousands.
From my brief interactions with her, I just can't reach the conclusion that she was at all a TERF.
One, people thought she was misgendering her. Wrong - Claude used "illa" (she) as her pronoun.
Two, people thought she was deadnaming her. Kinda, kinda not. Claude switched back to Lucy in 1938, but only because she was fleeing the Nazis, not because she thought it'd be grand.
In fact, she posted art from a trans person on the exact same day she had the label applied to her.
I spent way too long digging into this, and it appears to be all about a single post, vs. the flood of a dozen or so she'd write per day: one about the artist Claude Cahun.
Russia has been assassinating people and saboutaging things in western counties nonstop over the past decade, but how dare *Ukraine* hit a *military target* in *the country that invaded it*. ๐
Russia considers, under its own constitution, multiple oblasts of Ukraine, including vast swaths of land it doesn't even control, part of Russia.
The notion that Putin is going to commit suicide-by-nuke if Ukraine hits a Russian airbase with an ATACMS but not a homemade drone is just nonsense.
That should have read "With the occupation of France on the horizon". She *fled* in 1938. It doesn't help that she was also Jewish.
In the past, racial discrimination was normalized in "respectable" sources. So was discrimination against women. All sorts of problematic things were said about gay people. Etc.
What matters is: in our current context, with our current understanding, what is *right* to say about someone?
/END :)