[0:02] That's the sound of AI coming in and
[0:04] non-consensually motorboating us all.
[0:07] Whether you want to admit it or not,
[0:09] whether you even recognize it or not,
[0:11] your life is already being affected
[0:14] deeply by AI. And I know I've yapped
[0:17] about it a lot, but I did just see The
[0:20] Atlantic's most recent report where they
[0:22] actually pulled down the pants on a lot
[0:24] of what AI models are being trained on
[0:26] here, aka what they're stealing to train
[0:29] their AI models on because they don't
[0:31] have permission for it. The word, the
[0:33] nomenclature is stealing. That's the
[0:35] scientific term. But because these
[0:37] multi-billion corporations are in the
[0:40] the driver's seat now at the helm, they
[0:42] can go ahead and rebrand that whole
[0:44] stealing thing into something entirely
[0:46] different. They're trying to argue that
[0:48] it's all by the book in fact. So, uh,
[0:51] The Atlantic now has this AI watchdog,
[0:53] which is just a open tool. Anyone can go
[0:55] in and they can just directly search
[0:57] what music AI is being trained on as
[1:00] well as like videos AI is being trained
[1:02] on. Now, a ton of huge artists have been
[1:04] made aware of this and have called out
[1:06] AI companies for just directly ripping
[1:09] their music. Some of it unreleased, by
[1:11] the way, in order to train their AI
[1:13] models on. Notably, Sunno has been
[1:16] getting called out with spitballs fired
[1:18] at it and Rotten Tomatoes thrown at it
[1:20] because they're pretty shameless and
[1:22] unapologetic about it. Anyone remember
[1:25] back in the day the whole you wouldn't
[1:26] download a car, one of the most iconic
[1:28] [ __ ] antipiriracy advertisements
[1:31] ever? arguably just one of the most
[1:33] well-known uh campaigns ever as well.
[1:36] Well, now the same people that used to
[1:39] run that [ __ ] are the ones that are
[1:41] literally just downloading everything,
[1:43] stealing everything they can. It It's
[1:46] It's pretty incredible the uh 180
[1:48] they've done there on that whole
[1:50] messaging. And I know what you're
[1:52] thinking. Yes, they have stolen my
[1:56] videos to train their AI models on. I
[1:59] searched my name in their data set. 67
[2:02] of my videos have made it into nine data
[2:05] sets used by AI corporations to train
[2:08] their models.
[2:10] Heaven help us. Lord have mercy on the
[2:14] absent souls of these AI models that are
[2:17] being [ __ ] clockworked orangeed
[2:19] having the eyelids pulled wide open to
[2:22] be trained off my videos here. Those
[2:25] have to be the dumbest AI models you can
[2:27] find. Just look at this poor bastard
[2:29] here from YT Temporal 180M who's got 221
[2:33] of my videos shoved down its throat
[2:36] being trained off of things like
[2:37] horrible chair where I'm just making fun
[2:40] of a dog [ __ ] chair. Some old game plan
[2:42] commentary like Doom four feathers where
[2:44] I'm playing a Doom mod where I'm a
[2:46] chicken shooting at warthogs from Halo.
[2:50] It's [ __ ] uh washing machine repair
[2:53] championship where I'm just commentating
[2:56] a washing machine repair competition
[2:59] between some of the highest quality
[3:00] athletes you can find in the washing
[3:02] machine circuit. Not to bismerch the
[3:04] good name there. A foot fetish wedding
[3:06] dildo circus where you know I did a lot
[3:08] of cool trick shots with dildos. Like
[3:11] this this [ __ ] AI model. It must be
[3:14] just sitting there drooling. Actually,
[3:16] let me learn a little bit about my son
[3:18] here, seeing as I taught him everything
[3:20] he knows. YT Temporal 180M. It's a
[3:23] collection of 5.4 million YouTube videos
[3:26] compiled by a team of researchers at the
[3:29] University of Washington and the Allen
[3:31] Institute for AI to train a multi a
[3:33] multimodal model called Merllo.
[3:37] It was released in 2021.
[3:40] Bro, is Merllo an idiot? Be honest with
[3:42] me. Is this the dumbest AI you can find?
[3:44] If it's being trained even partly on
[3:46] some of my videos, there's a chance you
[3:47] ask it who wrote the Declaration of
[3:49] Independence and it says dildo titty
[3:51] fart or something. Very fascinating.
[3:54] Very interesting. So, YT Temporal got a
[3:58] huge dollop of some of my incredible
[4:02] work like hintai survive. What the hell
[4:05] is this? I don't even remember this. As
[4:07] the 21st century continues to evolve,
[4:10] human sexual fetishes are evolving right
[4:12] there alongside it. Okay, you know what?
[4:14] That one might actually be somewhat
[4:15] educational. That that actually might
[4:18] help them out a little bit because it
[4:20] it's not wrong. Now, obviously, my
[4:22] videos weren't handpicked for these AI
[4:24] models or anything. They made it into
[4:26] those giant compilations a lot of these
[4:29] groups put together solely to train
[4:32] models off of to give like a huge sample
[4:35] size. I I understand that, but it
[4:38] doesn't make it any less garbage. It's
[4:40] so [ __ ] ridiculous because it is just
[4:43] stealing it. Same thing when they do it
[4:45] with music. It is just stealing all of
[4:47] that to train their models off of. And
[4:50] it's been a huge contentious topic for a
[4:53] while now when it comes to AI. And no
[4:55] amount of like trying to put makeup on
[4:58] it changes the truth that they are just
[5:01] stealing to train their models. Now, if
[5:04] you as just a normal person try and
[5:06] follow their footsteps and do the same
[5:08] thing of just taking a ton of artists
[5:11] music and videos in order to make your
[5:14] own product off of those works, you're
[5:17] going to get arrested and charged with a
[5:19] little something called theft. But that
[5:22] word doesn't exist once you reach a
[5:24] certain level of wealth. These
[5:26] billion-dollar corporations get to kind
[5:29] of skirt around that a little bit.
[5:31] They're able to tiptoe around that. and
[5:33] we don't need to worry about a little
[5:34] pesky thing called theft for them. It's
[5:36] very different rules they play by there.
[5:39] And I know there's not a soul on this
[5:41] planet surprised by this, but it's still
[5:44] something I think worth yapping about,
[5:46] especially now that it's so easily
[5:47] accessible to see how many things just
[5:50] get up by these AI companies to train
[5:55] their models off of. all of these huge
[5:57] compilations of work and IP that they
[6:00] steal to train their models on in order
[6:02] to sell it to the people that are now
[6:04] hooked on AI as everyone is in this
[6:06] giant [ __ ] gold rush, this whirlwind
[6:09] of the in the industry. It's just so
[6:12] bizarre how things have just accelerated
[6:16] to this point. Now, one thing I got very
[6:18] curious about is trying to think of like
[6:20] the most latigious company I could think
[6:22] of to see if their work had also just
[6:24] been stolen by a ton of these AI models.
[6:27] So, obviously, the first thing that pops
[6:30] in my noodle is Nintendo. And yeah,
[6:33] Nintendo is not exempt from this. 4,926.
[6:37] That is a pretty big chunk of their
[6:40] trailers being used to train AI models
[6:43] that I'm sure Nintendo didn't sign off
[6:45] on because Nintendo wouldn't just be
[6:47] giving away this for free. 1,000%.
[6:51] You would need to pay oodles of clams to
[6:55] have access to their material in order
[6:57] to train your own stuff off of to sell
[6:59] that product. They are extremely strict
[7:03] when it comes to their copyright. They
[7:05] rule that [ __ ] with an iron fist. They
[7:07] are judge dread when it comes to their
[7:10] copyright. And yet here we have Runaway
[7:12] Gen 3 that just shamelessly takes 4,772
[7:17] of their trailers. Again, Nintendo is
[7:19] extremely strict with their trailers.
[7:21] You can use their trailers in like
[7:23] YouTube videos if you follow a very very
[7:26] particular set of guidelines around it
[7:28] that is extremely transformative.
[7:31] Like there are in intense rules
[7:33] Nintendo. so brutal with this. I
[7:35] remember there was a couple YouTube
[7:37] channels that got taken down because
[7:38] they used Nintendo music. Like they they
[7:41] took Nintendo music from games they
[7:43] owned, put it in their videos, and they
[7:44] lost their whole channels for it. There
[7:46] was also that time where some streamers
[7:48] got banned because they watched Nintendo
[7:50] trailers during the direct on stream.
[7:52] Like the point is they take that very
[7:54] seriously. And now Runaway Gen 3 just
[7:57] comes in here from the top turnbuckle.
[7:58] Takes 4,772
[8:00] of their trailers, most likely
[8:03] completely for free without Nintendo's
[8:05] permission, didn't pay a dime for it, to
[8:07] train their model on. So, uh, let's see
[8:09] what this is. Runaway AI collected
[8:11] YouTube videos to train a
[8:12] videogenerating AI model released as Gen
[8:14] 3 in 2024. An internal company document
[8:17] that was obtained by 404 media lists
[8:19] 3,970.
[8:21] What? Wait, that can't be right. Oh, oh,
[8:23] channels. I thought I was talking about
[8:24] videos in general. I was like, "No, they
[8:26] have more than that from just Nintendo."
[8:27] That Runaway identified as sources of
[8:29] highquality video for training.
[8:32] Nintendo video game trailers. Okay. The
[8:35] spreadsheet contains comments describing
[8:37] what is desirable about some of the
[8:38] channels. For example, beautiful
[8:40] cinematic landscapes, high quality
[8:42] scenes from movies, only four videos,
[8:44] but they are really well done. Super
[8:45] high quality sci-fi short films, and the
[8:48] holy grail of car cinematics so far.
[8:52] That must be talking about like Mario
[8:53] Kart or something. It's not clear which
[8:56] if any videos Runway actually used for
[8:58] training its AI system. Our search our
[9:00] search tool includes all YouTube videos
[9:02] from the named channels that were
[9:03] published before May 17th, 2024, which
[9:06] is 1 month before Runway introduced
[9:08] their model. So, this is something that
[9:11] The Atlantic also made note of. It's not
[9:14] 100% confirmed that they used all of
[9:17] these videos, but the fact that they
[9:18] were compiled by these companies. I
[9:20] think you can make a pretty educated
[9:22] guess that it was likely used to train
[9:24] their models. And also another thing
[9:26] they mention is that there are likely a
[9:29] lot of other ones that even though
[9:30] they're not here, doesn't mean they
[9:31] weren't used to train their model. It's
[9:33] a very tricky and messy, sloppy thing to
[9:36] nail down exactly what and what isn't
[9:38] being used to train models. But I would
[9:41] wager a guess that since it's here in
[9:44] the data set, they probably used it.
[9:47] Much like all of these also most likely
[9:49] used it, like this company, which also
[9:51] used a lot of my YouTube videos, which
[9:54] that actually kind of gave me a giggle
[9:56] when I saw that Nintendo of America was
[9:58] part of the same data set as this one
[10:00] where my videos are in, such as
[10:03] Seaweed's [ __ ] cool or whatever that
[10:06] one was. I already forgot. So Nintendo
[10:09] and I, we're basically in the in the
[10:12] same ballpark now when it comes to
[10:14] quality. That's pretty cool. I bet
[10:16] Nintendo's thrilled about that. Uh but
[10:19] anyway, point is thanks to the Atlantic
[10:22] data sets here that you can freely
[10:23] explore. You can see so so so many
[10:28] things have just been taken by these
[10:29] companies and put in these data sets
[10:31] that are presumably being used to train
[10:33] their AI models on. It's just it's
[10:36] pretty egregious. I I wanted to yap
[10:38] about it a little bit. That's it. See
[10:40] you.