this post was submitted on 04 Apr 2024
89 points (83.5% liked)
Technology
59710 readers
5629 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
If we're talking only about LLMs, then probably the biggest issues caused are threats to support line jobs, the enshittification of said help lines, blatant misinformation spread via those chat bots, and a variety of niche problems.
If we're spreading out to mean AI mor generally, we could talk about how facial recognition has now gotten good enough that it's being used to identify and catalogue pretty much anyone that passes a FR-equipped security system. Israel has actually been picking civilian targets via AI. We could also talk about "self driving" cars and the compeletely avoidable deaths they've caused. We could talk about how most convolution network AIs that identify graphic imagery and other horrific visuals use massive sweat shops to sort said graphic images for pennies. We could also talk about how mimicry AI has now been used to create both endless revenge porn of unwilling victims, and also faked the voice of others to try to scam them or make them not vote. There's plenty of damage AI as a whole has done, even if LLMs are the most minimal of all of them.
I don’t think that this is “AI more generally” as the public (and the current article) understands it. You’re lumping together any slightly self corrective algorithm under the AI umbrella. This might be technically correct, but it’s just operations, it’s not indicative of the current hype.
The limiting factor for self driving cars is hardware, not software. There is no commercially viable video technology available to allow taking the self driving technology out of the lab and into the consumer space. Unless you’re talking about Tesla-like systems which, of course, are neither a “self-driving” system nor consumer ready.
This is not AI. The technology behind the voice or image manipulation has existed for some time and has been used for fake porn and for fake voice calls for a long while. We’re only discussing about it now because they can generate traffic if they’re tied to a hype like AI. Very few people would read a story about a student sticking faces of his colleagues over naked bodies, but say the student used AI and suddenly everyone wants to find out what happened. It’s even worse: headlines are discussing the reaction of X celebrity to porn fakes in the context of AI even though porn sites have been having a fake porn section ever since the late 90s and they’re available to anyone with the mental capacity to click “I’m over 18”. Maybe you’re too young to remember, but google wasn’t always censoring search results. Before 2010-ish, fakes like these would routinely appear in google searches of a celebrity’s name. I’m not really sure why AI makes this any different
You are pulling a no true Scotsman fallacy here. AI has always been a somewhat vague term, and it's explicitly a buzzword in today's systems.
This AI front has also been taking the current form for more than a decade, but it wasn't a public topic until now, because it was terrible up until now.
The relevant things is that AI is automating a normally human-centric practice via extensive training on a data model. All systems I've mentioned utilize that machine learning practice at some point in their process.
The statement about the deepfakes is just patently incorrect on your part. It is a trained model which takes an input, and outputs a manipulated output based on its training. That's enough to meet the criteria. Before it was fairly difficult and almost immediately identifiable as AI manipulated. It's now popular because it's gotten good enough to not be immediately noticeable, done fairly easily, and is at the point where it can be mostly automated.
I never claimed that the current software didn’t use machine learning. I simply said that faking video/images has been happening long before machine learning was involved in it and I completely disagree that it is harder to identify fakes now than it used to be. Maybe it wasn’t a technology that was easily available, but image manipulation is something we have been seeing for a long time. If anything, the fact that it is know public knowledge that image, voice and movie clips can be faked will help people to stop trusting them when they shouldn’t
This is your straight statement, and your only argument was saying it was done before AI was used in it. That's a poor argument. That's like arguing that self driving isn't AI because remote control car piloting existed.
Automated image manipulation vs having 100s of hours in Photoshop. That's AI vs what came before. Inputting a source file and getting a manipulated file after some amount of time, vs hours of meticulous work trying to get minor details right.
If we want to compare oldschool manipulation vs AI Manipulation, then yes, fakes now are on par with the insane skill of some image doctoring artists - you're just looking for different things - but it's at an exponentially lower cost than hiring a professional. Compare AI to itself, though? It's night and day. Early AI manipulation was atrocious. And modern AI manipulation is only going to get better. That is all due to breakthroughs in AI. imagine what the hell will happen when Sora becomes usable by anyone.
Machine learning has taken an originally hard thing to do and made it cheap and easy. Now, any schmuck can pump out doctored footage in an afternoon. That's why the AI porn is big- you can pay dirt cheap and give the model photos of any random woman and it'll make porn of them - and that fact has turned it into a much more viable business model than before, that's currently creating massive amounts of non consensual porn fakes- exponentially more than before.
I think I explained in the rest of the paragraph what I meant by that. What’s the point of replying to someone while truncating their message? I know what I said and why.
Sure, but I don’t see how this is a problem except maybe for the workers of the porn industry? And in general, with generated content (text or anything else) the major concern seems to be quantity. Exponential growth of content means nothing. The consumer cannot grow the consumption exponentially so this will just end up using storage space
A lot of what you’ve mentioned has existed for decades in some fashion. It’s just code.
And making these tools mass market instead of being something niche that requires actual talent to do is absolutely something to blame ChatGPT for.