• last month
Transcript
00:00All right, so I have a theory, and I don't know what to call this or if I should name
00:13it or anything, but here's my theory.
00:15So the really impressive part about artificial intelligence is that as it gets better and
00:19better it feels more and more like human intelligence, right, which is why we use that word.
00:25To the point where it can actually pass for human intelligence in some cases, like it
00:28passes tests, it solves problems, it finds patterns, it mimics things, and so it can
00:34actually kind of fool you into thinking it's intelligent.
00:37And so now we have this generative AI that can be trained on massive data sets and spit
00:42out unique outputs that are increasingly impressive.
00:46Now to be clear, AI, like artificial intelligences that have been better at humans at some things,
00:52that's been around for a while, like that was kind of the point actually.
00:54They could comb through large amounts of data and find patterns that humans never would,
00:59like literally detect disease in its earliest stage of development better than an actual
01:03doctor.
01:04So AI has already been useful.
01:06But where it's generative AI that we're talking about, where we're asking AI to be creative
01:11and come up with new things, new text, new images, new sounds, that's what I'm talking
01:18about.
01:19That's the scary one.
01:20So my theory is there are two levels of success of efficiently advanced AI.
01:29The first level is when AI-generated content fools you because you weren't paying attention
01:34to AI or looking for AI.
01:37So maybe you, like me, saw the Pope photo on your timeline a couple days ago and just
01:41kept scrolling.
01:42You just thought, oh, that's a weird picture of the Pope, kind of a cool jacket, I guess.
01:46But only when you get told that it's AI and not a real picture does it trigger your skeptical
01:51eye and suddenly you're like, oh, that's obviously not a real picture.
01:54There's so many things wrong.
01:55The hands look weird.
01:56The details are garbled.
01:57Of course it's a fake picture.
02:00But the second level of AI-generated content fooling you is when you know you're supposed
02:06to be looking at AI and it still fools you.
02:10That's when it gets scary.
02:13So we've had a couple of examples of the level one instances, like the Pope photo,
02:17like I mentioned.
02:18There have been others floating around, like there was one of Trump getting arrested even
02:21though he didn't.
02:22But what's level two?
02:24I got an email a couple weeks ago from an artist you may be familiar with, Mr. Jay Medeiros,
02:29one half of Alta and Vaincil, and they had a new track that they were working on, which
02:33was a collab where he raps alongside Jay-Z, except the Jay-Z voice is not real.
02:42It's AI-generated.
02:44And they were going back and forth on how to release this, whether or not they should
02:48release it at all.
02:50I want you to listen to this.
02:52Not the whole thing.
02:53I'll play a little bit of it for you.
02:54You should still go to the source and absolutely watch the original video.
02:56I'll link it below.
02:57But I just needed to hear this and feel what I felt when I first heard this song.
03:12So, for those of you who know what Jay-Z sounds like, that was absurd.
03:31Like that sounded a lot like Jay-Z to the point where even knowing that I'm listening
03:35to AI-generated content, I can still hear it and sort of enjoy it like I'm actually
03:41listening to Jay-Z.
03:44It's a lot.
03:45Now, to be clear, Mr. Jamie Jaros told me they wrote those lyrics before the AI tools
03:48became available, and the tools they used weren't even plug-and-play perfect.
03:53There were still lots of layers to it, tweaking things, experimenting with different methods.
03:58Apparently, words like feeling and sealing and appealing were hard to get it to rhyme
04:04because they'd pronounce it slightly differently, so they ran it through a bunch of times.
04:07But the fact that it's still so good in the final result, that's what opens up the can
04:12of worms.
04:13It's kind of borderline scary, knowing how good it actually is.
04:17Just knowing that I'm listening to AI and still enjoying it like it's bars by Jay-Z.
04:23And the cherry on top is that this is the worst this technology will ever be.
04:27So, there's examples of level one everywhere, like low-stakes content where it passes because
04:33the audience isn't really looking for AI and it doesn't really matter.
04:37It's true of a lot of the chatbot stuff, too, like ChatGPT and BARD and all the others,
04:42like Bing, they might write an email.
04:44You might read an email that seems pretty sincere, but turns out most of it was written
04:48by one of those chatbots, and it's kind of fine because it's just an email.
04:53But just keep in mind that the goal of all of these technologies is to advance to get
05:00to level two.
05:03The goal of the chatbot, eventually when it gets better, is to be able to pass as a human
05:08in conversation.
05:09What does that mean?
05:10And the goal of the image generator is to be able to produce usable paintings or drawings
05:16or whatever, like a human.
05:18What does that mean?
05:19The goal of self-driving cars is to drive alongside other humans on the road.
05:23And that's wild, but that's what they're all working towards.
05:28So is there a solution to this?
05:30I mean, the answer really is not yet.
05:33It's so new that there is no answer.
05:35Some people think regulation.
05:37Some people think just outright banning it is an option.
05:41I don't really believe in that.
05:42But I think what's going to have to happen is there probably has to be some sort of a
05:46parallel development of tools that are specifically designed to detect AI content, so that you
05:52need to learn to deploy those tools when there's a reason to detect AI.
05:57For now, let's enjoy level one, because we won't have just that for much longer.
06:05Thanks for watching.
06:07Catch you in the next one.
06:08Peace.