It would be nice if this extended to all text, images, audio and video on news websites. That’s where the real damage is happening.
Actually seems easier (probably not at the state level) to mandate cameras and such digitally sign any media they create. No signature or verification, no trust.
No signature or verification, no trust
And the people that are going to check for a digital signature in the first place, THEN check that the signature emanates from a trusted key, then, eventually, check who’s deciding the list of trusted keys… those people, where are they?
Because the lack of trust, validation, verification, and more generally the lack of any credibility hasn’t stopped anything from spreading like a dumpster fire in a field full of dumpsters doused in gasoline. Part of my job is providing digital signature tools and creating “trusted” data (I’m not in sales, obviously), and the main issue is that nobody checks anything, even when faced with liability, even when they actually pay for an off the shelve solution to do so. And I’m talking about people that should care, not even the general public.
There are a lot of steps before “digitally signing everything” even get on people’s radar. For now, a green checkmark anywhere is enough to convince anyone, sadly.
I think there’s enough people who care about this that you can just provide the data and wait for someone to do the rest.
I’d like to think like that too, but it’s actually experience with large business users that led me to say otherwise.
It could be a feature of web browsers. Images would get some icon indicating the valid signature, just like browsers already show the padlock icon indicating a valid certificate. So everybody would be seeing the verification.
But I don’t think it’s a good idea, for other reasons.
An individual wouldn’t verify this but enough independent agencies or news orgs would probably care enough to verify a photo. For the vast majority we’re already too far gone to properly separate fiction an reality. If we can’t get into a courtroom and prove that a picture or video is fact or fiction then we’re REALLY fucked.
Seems reasonable to me. If you’re using AI then you should be required to own up to it. If you’re too embarrassed to own up to it, then maybe you shouldn’t be using it.
What about my if else AI algorithm?
It’s not really an llm
IMO if your “A*” style algorithm is used for chatbot or any kind of user interaction or content generation, it should still be explicitly declared.
That being said, there is some nuance here about A) use of Copyrighted material and B) Non-deterministic behaviour. Neither of which is (usually) a concern in more classical non-DL approaches to AI solutions.
Its insane how a predictive chat bot model is called AI
Marketing
So does the EU AI act
My LinkedIn feed is 80% tech bros complaining about the EU AI Act, not a single one of whom is willing to be drawn on which exact clause it is they don’t like.
Oh, so just like with the GDPR, cool.
Ok, my main complaint about GDPR is that I had to implement that policy on a legacy codebase and Im pretty sure I have trauma from that.
Sounds like that codebase was truly awful for user privacy then.
Incredibly so, yes.
Skill issue.
My point is higher than yours, get on my level
My LinkedIn feed
Yes… it’s so bad that I just never log in until I receive a DM, and even then I login, check it, if it’s useful I warn people I don’t use LinkedIn anymore then log out.
I get it though, if you’re an upstart. Having to basically hire an extra guy just to do ai compliance is a huge hit to the barrier of entry
That’s not actually the case for most companies though. The only time you’d need a full time lawyer on it is if the thing you want to do with AI is horrifically unethical, in which case fuck your little startup.
It’s easy to comply with regulations if you’re already behaving responsibly.
That’s true with many regulations. The quiet part that they’re trying to avoid saying out loud is that behaving ethically and responsibly doesn’t earn them money.
I am of the firm opinion that if a machine is “speaking” to me then it must sound a cartoon robot. No exceptions!
I want my AI to sound like a Speak & Spell.
i would like my GPS to sound like Brian Blessed otherwise i want all computers to sound like Niki Yang
Be sure to tell this to “AI”. It would be a shame if this was a technical nonsense law to be.
Same old corporations will ignore the law, pay a petty fine once a year, and call it the cost of doing business.
Has anyone been able to find the text of the law, the article didn’t mention the penalties, I want to know if this actually means anything.
Edit: I found a website that says the penalty follows 5000*sum(n+k) where n is number of days since first infraction, this has a closed form of n^2+n= (7500^-1)y where y is the total compounded fee. This makes it cost 1mil in 11 days and 1bil in a year.
That might end like the cookie popups in the eu…
What if it’s foreign AI ?
How do you enforce this
If you ask ChatGPT, it says it’s guidelines include not giving the impression it’s a human. But if you ask it be less human because it is confusing you, it says that would break the guidelines.
ChatGPT doesn’t know its own guidelines because those aren’t even included in its training corpus. Never trust an LLM about how it works or how it “thinks” because fundamentally these answers are fake.
Any word on the 3 laws of robotics?
When I read that shit as a kid, I thought Asimov’s laws of robotics were like natural laws, so that it was just naturally impossible for robots to behave otherwise. That never made any sense to me so I thought Asimov was just full of shit.
Weird how California keeps being the most progressive state in the US.
It’s like being the best smelling turd in a toilet, but at least it’s something.
