Brad Smith urges steps to curtail deepfakes "with an intent to deceive or defraud."
See full article...
See full article...
Microsoft president declares deepfakes biggest AI concern
- Thread starter JournalBot
- Start date
Upvote
21
(45
/
-24)
I'm a little bit more worried about the whole universe being turned into paperclips, but ok, I guess someone being embarrassed is pretty bad, too.
Joking/not joking.
Joking/not joking.
Last edited:
Upvote
-19
(7
/
-26)
Licenses! We obviously get our license automatically of course, since we wrote the regs, but everyone else, LICENSES!
In other words, secure oligopoly position through the vein of security.
ETA: It's also rich coming from Microsoft given how many thousands of ignored GPL licenses they blew through collecting training data.
In other words, secure oligopoly position through the vein of security.
ETA: It's also rich coming from Microsoft given how many thousands of ignored GPL licenses they blew through collecting training data.
Last edited:
Upvote
50
(56
/
-6)
What! Microsoft is worried about deepfakes that lead to deceit or fraud?! Aren't they getting a little ahead of themselves and shouldn't we be waiting for, ya know, actual harm to occur. Let's not jump the gun, we should wait until a gun goes off and prove that actual harmed has occured, by oh, I dunno, $1000 or more. I mean... isn't that their line of thinking? When did it change?
/s
/s
Upvote
-17
(6
/
-23)
Clicked through to read the speech. This mf thinks someone is going to enforce content labeling in order to police disinformation and that there will be "accountability". Delusional pablum.
Upvote
46
(47
/
-1)
In other words, what they actually mean to say is: now that we have a jump ahead on the technology in front of others, we must secure our dominance.
Upvote
29
(34
/
-5)
"What does it mean when companies like Microsoft are selling the very product that they are warning us about?" -- great question. i'd suggest that, in the final analysis, they don't give a fu*k. all the rest is mr smith trying to create a (semi) plausible statement-trail to disguise that.
Upvote
27
(28
/
-1)
Every single medium used on electronic devices for communicating will be hijacked by an ever changing series of LLM, and other AI systems, and there will be no way to keep up with the details in the moment. We may months or years later be able to confidently say something is generative, or not, but the world will likely have made their assumptions about it before verification can be had.
For example. LLMs will be able to create convincing texts, and emails that scam, and defraud people, but do not contain enough verbiage to run through systems with a 100% confidence rate on if it was generative.
For example. LLMs will be able to create convincing texts, and emails that scam, and defraud people, but do not contain enough verbiage to run through systems with a 100% confidence rate on if it was generative.
Upvote
15
(16
/
-1)
says the company who's research developed a ai capable of imitating someone's voice with only a few seconds of audio
Upvote
22
(25
/
-3)
Celery Man
Ars Legatus Legionis
"I'm so sorry you feel that way. It was not our intent to cause harm, so if you feel harmed, that is not our fault."
Upvote
6
(8
/
-2)
I see a real potential here for an Elizabeth Holmes type person to come out soon with technology to make all cameras record real pictures on the blockchain using ntfs and digital ai bullshit buzzword buzzword bullshit, and fool people into investing billions into some bullshit mumbo jumbo that absolutely doesn't work.
Upvote
22
(22
/
0)
like most tools .. it can cut both ways. I find the Belenciaga Ai generation entertaining... the Marvel Gucci Ai is in the same boat...
Upvote
-2
(1
/
-3)
Hmm. So Altman wants licensing, which would conveniently cement incumbent control over the market, and now Smith also wants "know your customer", which would be a reason to collect all kinds of data that couldn't possibly be used for anything else... an area in which Microsoft products are already pushing the frontier.
Strange coincidences...
Strange coincidences...
Upvote
22
(22
/
0)
How do you program something to determine the intent of its output? This isn't U238 enrichment, there isn't a quantifiable threshold.
The same technology that lets fans "ship" two celebrities or characters for amusement value can be used to create fake sexual material for defamation or blackmail.
How do you determine the difference between someone who wants to see Maryanne and Ginger leave the boys and "explore the other side of the island", and someone who wants to create a fake photo of an ex-lover or coworker in a compromising situation?
How do you distinguish between someone who wants to see a politician they dislike get eaten by a Sarlacc, and someone who creates a fake photo that depicts a political assassination?
Human moderators can decide in-context which uses they want to host, but I'm not sure how you would prevent these things from a technical standpoint.
The same technology that lets fans "ship" two celebrities or characters for amusement value can be used to create fake sexual material for defamation or blackmail.
How do you determine the difference between someone who wants to see Maryanne and Ginger leave the boys and "explore the other side of the island", and someone who wants to create a fake photo of an ex-lover or coworker in a compromising situation?
How do you distinguish between someone who wants to see a politician they dislike get eaten by a Sarlacc, and someone who creates a fake photo that depicts a political assassination?
Human moderators can decide in-context which uses they want to host, but I'm not sure how you would prevent these things from a technical standpoint.
Upvote
13
(13
/
0)
The aspects of ai that would put thousands or even millions of Americans out of work aren't the issue, the issue is rich people like himself might be embarrassed...
I wish I were shocked at the lack of introspection.
I wish I were shocked at the lack of introspection.
Upvote
11
(11
/
0)
Your "good" examples aren't exactly slam dunks.
"Mary Anne" and "Ginger" were actual living actresses, one still living, who might not want their images used that way.
Well, that's not a very credible assassination, but anyway maybe the people providing the models aren't that interested in supporting that anyway?
On edit: Given the downvotes without associated comments, I'm assuming everybody on here thinks that agreeing to be on a silly TV show in 1964 makes you fair game for deep fake porn, and that there's some obligation to provide people with a way of living out their bloody fantasies about any public figure.
Last edited:
Upvote
-9
(5
/
-14)
Upvote
8
(9
/
-1)
Upvote
3
(3
/
0)
"every organisation behaves as if it is run by secret agents of its opponents" - Robert Conquest
Upvote
-2
(1
/
-3)
Strategically deluded pablum. The risks he talks about are unrelated to the revenue producing parts of AI that MS can foresee for their business.
The obvious, real risks are in text LLMs being used for misinformation: "flooding the zone with shit" as well as text LLMs giving people bad input, that is actually in the interest of the corporations running the AI (e.g., What toaster should I buy? Gives you advice based on what toaster makes the most money for MS to recommend). He's not going to mention those problems b/c those problems are actually good for Microsoft.
Upvote
20
(20
/
0)
Satire is valid, and it would not surprise me if "I was presenting satire" should become the goto defense, should there be legislation about this abuse.
Upvote
1
(1
/
0)
Upvote
-3
(1
/
-4)
For humans it’s usually easier to critique than it is to create. Technology is obviously very different from people but I’d think that if computers can create images then it should be easier than that to create automated ways to check images for signs they may be fake.
Alternatively, maybe what we really need is a better way to combat misinformation.
Upvote
1
(1
/
0)
Upvote
-2
(0
/
-2)
I think the regulations China implemented were basically right, for the deepfake issue. If you produce a work that could reasonably be perceived as depicting real events and it's edited by AI, you have to note prominently that the image is AI edited or generated. Similar to the French laws around Photoshop, but across society rather than just in specific classifications of publications.
Upvote
2
(2
/
0)
Is there evidence that works? I can easily see the unintended result being to teach people that anything not labeled as AI is automatically real, which makes passing off a fake that much easier if you already don't care much about the ethics or legality of what you're doing.
Upvote
8
(8
/
0)
Well, I think we are all screwed. Also, why do we think that bad actors like geopolitical opponents will agree to participate in such regulation?
Upvote
3
(3
/
0)
Not that I have access to. It's only a couple of years old, and I don't speak Chinese well enough to understand primary source material. Anecdotally, I follow a few people in mainland China and there doesn't seem to be an epidemic of fake images. Most normal people seem to be complying with the regulations (more than I thought would be required, so you'll see it sometimes on all AI generated imagery) so if you see an AI generated image, everyone I know says it is labelled clearly. I don't think the Pope puffy jacket went viral there but I can ask sometime.
In general, it's not really about preventing all instances of deceptive images from existing - you would need to ban Photoshop and go remove it from millions of computers for that. It's just about trying to institute new norms backed by law, so that if someone does engage in the practice almost everyone agrees is bad (intentionally creating deceptive imagery using AI to try and convince people a fake event happened or a real one didn't), then that person can be charged with a crime.
Upvote
0
(0
/
0)
If your going to light something then twirl,the birds and show the badge - no problem. This spy verses spy stuff is rediculous. Mean nobodies keeping anybody from joining the loop. Except for perhaps clueless politicians,and dateless housewifes.
The server monopoly,and the media monopoly arent going to leave anybody clueless unless someone actually notices. That guy too.
The server monopoly,and the media monopoly arent going to leave anybody clueless unless someone actually notices. That guy too.
Upvote
-7
(0
/
-7)
train_wreck
Ars Scholae Palatinae
Upvote
-1
(1
/
-2)
"So the question remains: What does it mean when companies like Microsoft are selling the very product that they are warning us about?"
Well, in an outburst of perfectly fine logic, for many it clearly means that we can therefore completely disregard the worries and warnings.
Well, in an outburst of perfectly fine logic, for many it clearly means that we can therefore completely disregard the worries and warnings.
Upvote
4
(4
/
0)
Upvote
0
(2
/
-2)
No, that was a decade ago, and even then you'd have missed the boat to some degree.
Say what you want about the current state, but the cat's out of the bag. Even if the US banned it, the tech's out there, it would continue to be developed elsewhere and you'd still have to deal with the results even if it were the first truly effective ban in history and nobody on US soil ever touched it again.
Upvote
4
(4
/
0)
Also, the problem is human beings anyway. There were some hilariously obvious bad Photoshops doing the rounds during the last US election, but the target audience didn't seem to care. So long as it was in the correct meme format, they believed it anyway because they wanted it to be true, and that overrode the fact that it was obvious fraud.
Upvote
10
(10
/
0)
Pretty much this right here.
—-
The collective dissolution of reality isn’t going to happen by dis/misinformation - though that will be a compounding factor.
As in this article, the ‘art’ is generated by stable diffusion. We are not far at all from it being easier to generate “Microsoft CEO looking scared” than Google and browse for images of the same person. Which in short order will be filled with generated images…
even well-meaning, respected reporting that DOESNT have an agenda, will contribute to the dissolution of reality (what is real), as these tools become part of workflows for huge numbers of people.
Deepfake fraud - yes will be a problem - but it’s not going to scale in anywhere near the same level as generated-reality-breakdown will, because there far more good actors than bad actors in the world just doing their jobs.
Upvote
2
(2
/
0)