Tickets, A Necktie & 26 Years of Footage: Long-Form 3 — Transcript
Full transcript
- 0:05Long form episode 3, a vehicles episode
- 0:08of sorts. I am being sucked into the AI
- 0:11rabbit hole. I know that will disappoint
- 0:13a lot of you, but more will be revealed.
- 0:16A build about [music] survival for a
- 0:18Patreon subscriber. Finding the
- 0:20connection between a necktie, traffic
- 0:22tickets, Ariel Schulman,
- 0:24>> hi.
- 0:25>> And Catfish the movie [music] using AI
- 0:27to fight the man. See what this little
- 0:29bad boy businessman is all about.
- 0:31>> We made an archive footage crawler. It
- 0:33works. Hallelujah. Plus a few spices
- 0:36[music] to keep things interesting.
- 0:43>> All cars today are painted
- 0:47in grayscale.
- 0:51Look, they're all grayscale.
- 0:56They're between the colors of black and
- 0:58white.
- 1:00Car in front of me.
- 1:02Car to the left of me. The oncoming car.
- 1:07And here's the break as a Mercedes right
- 1:09here.
- 1:12Just one exception. Look. Grayscale.
- 1:15Grayscale. Gray scale. Gray scale.
- 1:17Grayscale. Grayscale. Gray scale.
- 1:21Gray scale. Grayscale.
- 1:23Blue. Ding ding. Grayscale. Grayscale.
- 1:26Grayscale. All three of our cars are
- 1:28grayscale. Grayscale. Grayscale.
- 1:30Grayscale.
- 1:33Gray scale. Gray scale. Gray scale.
- 1:34Grayscale. Blue. Red.
- 1:41All right. So, it was somewhere around
- 1:43here.
- 1:45I think that Morgan's dad got demolished
- 1:48by a car on a bike ride.
- 2:33The thing that I found most interesting
- 2:35about the story of Morgan's dad's bike
- 2:37crash
- 2:39is that he didn't tell anyone about it
- 2:43until he was asked.
- 2:46So,
- 2:50you really have to kind of look for this
- 2:52date. You have to catch the
- 2:55light just right or else it doesn't show
- 2:57up.
- 3:00And this will mount in their beautiful
- 3:03house in Alpine.
- 3:08And then a lot of people are going to
- 3:10ask them about it.
- 3:15We're coming up on the 20th anniversary
- 3:18of the Respectability Tour. [music] And
- 3:21for some reason, while I was laying out
- 3:24the modules for this episode, I got the
- 3:28urge to include the ironing my tie
- 3:31sequence from the Respectability Tour.
- 3:34[music]
- 3:34And what's special about the ironing my
- 3:38tie sequence [music]
- 3:39is who made it. My brother Casey didn't
- 3:43make it. I didn't [music] make it. Ariel
- 3:47Schulman made it.
- 3:49>> Hi, how are you?
- 3:51>> Who is Ariel Schulman? I'll tell you
- 3:54after we watched Ironing My Tie from the
- 3:58Neistat Brothers Respectability Tour
- 4:002006.
- 4:12[music]
- 4:14I bought this tie on the Salvation Army
- 4:19on Bank Street in New London,
- 4:21Connecticut. [music] And I like to
- 4:22imagine it belonged to some like old guy
- 4:25who is a Yaley [music] or something, but
- 4:26it's probably not that old.
- 4:29Well, I'm not considering. I think my
- 4:31plan [music] is for 2007, I'm going to
- 4:33switch from a casual jeans and
- 4:36t-shirtbased
- 4:38um wardrobe [music] to a suitbased
- 4:41wardrobe.
- 4:43Yeah. Respectability this tour. Like the
- 4:46people that the producers of this
- 4:49project, you know, they said we're going
- 4:50to have people do us a lot of favors,
- 4:52but [music] don't try pull any stunts.
- 4:54And I think people have the wrong idea
- 4:56about us that we're like bad and mean to
- 4:59people. We just screw people over all
- 5:00the time. [music] But really, we're not
- 5:02like that. Um,
- 5:04you know, and what [music] we've tried
- 5:05this trip is just to be really
- 5:08respectable to everybody. and it's been
- 5:09really really beneficial.
- 5:17Why did I want to put this in this
- 5:19episode? What's the connection? This
- 5:21episode is about using AI to beat
- 5:23traffic tickets. But without knowing the
- 5:26connection, I decided to put ironing my
- 5:30tie in there and figure out the
- 5:32connection later. And I think I know the
- 5:37ironing my tie connection to this
- 5:40episode.
- 5:42And the connection is three-fold. First
- 5:47and loosest connection, traffic
- 5:49violations and the DMV. Casey and Mickey
- 5:53go to battle with the DMV 20 years ago
- 5:56in the respectability tour. I go to
- 5:58battle with the DMV in this episode. I
- 6:01got a ticket. I deserved in the
- 6:03respectability tour 20 years ago. I get
- 6:06a ticket I don't deserve in this
- 6:09episode. The second connection between
- 6:13this episode and the ironing my tie
- 6:16sequence is decency.
- 6:21That's what I think the ironing my tie
- 6:24sequence is about. I think it's about
- 6:26decency. And I think it was also
- 6:30indecent of the park rangers back in
- 6:34March of this year to drive around the
- 6:38OV area parking lot to ticket the dads
- 6:43there with their kids just trying to
- 6:45keep them off of screens. Ticketing the
- 6:47dads with hundreds of dollars of
- 6:50violations for not having green stickers
- 6:54on their bikes and trucks. Those rangers
- 6:57could have given us warnings and
- 6:59achieved the same results.
- 7:02And the third connection is catfish.
- 7:06The movie catfish and why catfish was
- 7:10titled catfish. What's the source of the
- 7:14newest meaning of the word catfish? And
- 7:18the newest meaning is in the dang
- 7:20dictionary. Miriam Webster's catfish. A
- 7:25person who sets up a false personal file
- 7:29on a social networking site for
- 7:32fraudulent or deceptive purposes.
- 7:37And it was Henry Just, Nev Scholman, and
- 7:41his brother Ariel Schulman.
- 7:43>> Oh, hi. How are you? who made Catfish
- 7:47the movie and coined the new definition
- 7:50of catfish by titling the movie Catfish.
- 7:55But it was Vince
- 7:58in the movie Catfish who gave us the
- 8:01beautiful explanation of where the
- 8:04catfish archetype comes from.
- 8:09They used to tank uh
- 8:12cod from Alaska
- 8:14all the way to China. They'd keep them
- 8:16in vats in the ship.
- 8:20By the time the codfish reached China,
- 8:23the the flesh was was mush and
- 8:27tasteless.
- 8:29So this guy came up with the idea that
- 8:33if you put these cods in these big vats,
- 8:36put some catfish in with them and the
- 8:39catfish will keep the cod agile.
- 8:44And there are those people who are
- 8:46catfish in life and they keep you on
- 8:49your toes. They keep you guessing. They
- 8:52keep you thinking. They keep you fresh.
- 8:54And uh I thank God for the catfish
- 8:56because we'd be droll, boring, and dull
- 9:00if we didn't have somebody nipping at
- 9:01our fan.
- 9:05And in that respect, me using AI to
- 9:08fight these tickets and me using AI to
- 9:12make trouble at the DMV, maybe that
- 9:16makes me a catfish.
- 9:33>> [music]
- 9:35>> This is the current way I fight a
- 9:38traffic ticket. [music] Google traffic
- 9:40lawyer. Hire traffic lawyer. Traffic
- 9:43lawyer beats ticket. I pay traffic
- 9:45lawyer hundreds of dollars. But for
- 9:47these, I'm going to try a new way. New
- 9:50way. What's this about? A new way.
- 9:52>> You know what?
- 9:56>> Let's use Claude for this.
- 9:59See what this little bad boy AI is all
- 10:02[music] about.
- 10:08>> [music]
- 10:10>> This next bit is about using an AI model
- 10:13as a sort of lawyer, which involves a
- 10:15lot of back and forth that this episode
- 10:17sponsor, Whisper Flow, drastically
- 10:20accelerated. Whisper Flow is a
- 10:22voicetoext tool. Works on my computer,
- 10:24works on my phone, works inside all the
- 10:26apps I actually use. You talk, it types.
- 10:30I bought Whisper Flow before they
- 10:32sponsored me and I've been pushing
- 10:34Whisper Flow word of mouth since I
- 10:36started using it. Okay, watch this. Can
- 10:38you investigate whether or not the DMV
- 10:40will issue OV green [music] stickers for
- 10:42electric dirt bikes like mine? Typing
- 10:44that out took me 26 seconds. Speaking it
- 10:47took about 7 seconds. My phone's builtin
- 10:50talk to [music] text makes me sound
- 10:52janky. Whisper flow makes me sound
- 10:54eloquent. [music] Watch. Um, so three
- 10:59filmmakers I have a tendency to [music]
- 11:01steal from are uh one Verer Herdzog,
- 11:07um, two, uh, Adam Curtis and, uh, three,
- 11:13um, Agnes Varda. Whisper Flow strips out
- 11:16the ums and the H, builds the list
- 11:18right, and spells the hard names
- 11:20correctly. Herzog, Agnes Varda. Somehow
- 11:23this state-of-the-art [music]
- 11:24technology lets me live more analog. Use
- 11:27the code spirited man for 1 month of
- 11:29Whisper Flow Pro free. Links in the
- 11:32description. Thank you, Whisper Flow.
- 11:43>> Take a picture of the tickets with the
- 11:44phone. Drag it into Claude. Claude has
- 11:48the plan.
- 11:48>> Okay, this is what you're going to do.
- 11:50immediately pay for the permits
- 11:54>> then I'll write a letter to the judge
- 11:56>> and that's where it up because
- 11:58you can't send a letter directly to the
- 12:00judge but I don't know that so I just do
- 12:03what Claude says tells me to get the
- 12:05permits right now I get the one for the
- 12:08Tacoma no problem [music] but then I
- 12:11tried to get [singing] the one for the
- 12:13bikes I tried to get the stickers for
- 12:14the bikes
- 12:16and that's what started my war with the
- 12:19DMV and That's what started my war with
- 12:21the DN.
- 12:38I've got dozens of hours putting
- 12:40together paperwork in preparation for my
- 12:42DMV visit to get green passes so that I
- 12:45can use the OV. Despite hours and hours
- 12:48of research and care and consulting with
- 12:51multiple AI frontier models,
- 12:57there is no way to do it. There is no
- 13:00way to get offhighway vehicle green
- 13:03stickers for this style of bike.
- 13:07And I'm learning that at the Department
- 13:09of Motor Vehicles right now.
- 13:12I put dozens and dozens of hours of
- 13:16effort and work based on your assessment
- 13:20that these electric bikes could be
- 13:22authorized with green stickers to use on
- 13:24OV property when that is not true. Why
- 13:28wouldn't you advise me of this from the
- 13:30beginning and save me all the trouble
- 13:33and hours and hours of work that could
- 13:40have gone into me making money to make
- 13:42it irrelevant as to whether or not I was
- 13:44in compliance.
- 13:47[music]
- 13:48Don't you just love DMV storylines?
- 13:54Here's
- 13:55[music]
- 14:00the paradox. You need OV sticker to ride
- 14:04in the OV area. You get the OV sticker
- 14:07at DMV. The DMV won't give OV stickers
- 14:12to electric dirt bikes. However,
- 14:16thanks to AI, I found out that something
- 14:20called SB586,
- 14:23new law, went into effect January 2026.
- 14:26The DMV has to issue OV stickers to
- 14:31electric dirt bikes. So, the DMV is not
- 14:35in compliance. That sort of helps me
- 14:38because I can go to the judge and say,
- 14:41"Look, I tried to remediate this
- 14:43problem, but [music] the DMV is not
- 14:45issuing the the stickers even though
- 14:49this law SBA 586 [music] says that they
- 14:52they must that they're compelled to."
- 14:54So, I had AI draft up the letter to the
- 14:58judge, put the whole case together. It
- 15:00told me how to put together all of the
- 15:02all of the it's called a mitigation
- 15:04packet and it's so it's like exhibit A,
- 15:07exhibit B. Here's me buying the pass for
- 15:10the Tacoma. Here is my testimony of
- 15:13being denied the stickers at the DMV.
- 15:15The mitigation packet says, "Judge,
- 15:17look, I know I broke the rules, but I
- 15:21tried to fix them. The DMV won't let me
- 15:23fix them. Can you please cut me a break
- 15:25here?" And I did fix one of them. I got
- 15:27the sticker for the truck. And this is
- 15:29called curing the tickets. [music] AI
- 15:32came up with this strategy. I broke the
- 15:34law, but subsequently I've cured the
- 15:37infractions. Well, one of them [music]
- 15:44[music]
- 15:45Zoom court preparation. My entire case I
- 15:49put together with Claude
- 15:52knowing that the law and the court hates
- 15:55that. Let's see if I win.
- 15:58I'm not allowed to record the
- 15:59proceedings. It's illegal.
- 16:02>> Thank you, Court Deputy Jerry, for
- 16:04forwarding my mitigation packet to the
- 16:06prosecutor, as I found out I wasn't
- 16:09allowed to send it to the judge, which I
- 16:11had done under the advice of my counsel,
- 16:14Claude. [music]
- 16:19So, here's what happened. They offered
- 16:22me a deal.
- 16:25don't have to pay for the Tacoma not
- 16:27having the sticker cuz I got I got one
- 16:31subsequent to the violation. Then I get
- 16:34to pick one of the two bikes to pay the
- 16:37violation for. The prosecutor offered me
- 16:40this or take my chances with a judge at
- 16:44a later date. So, I took the deal. Even
- 16:48though the DMV makes it impossible
- 16:53to register these bikes, even though the
- 16:56California state law says that the DMV
- 16:59has to make it possible for us to
- 17:01register these bikes.
- 17:04My next battle is with the DMV.
- 17:08It's not going to work, but maybe I can
- 17:10light a fire under their ass.
- 17:23I dreamed of a footage crawler that
- 17:25lived on a Raspberry Pi inside my
- 17:27Odyssey book. Plug it in, tell it what
- 17:29shots I'm looking for. Motorcycle night
- 17:31city. Crawler searches through my
- 17:3326-year archive, finds me the shots, and
- 17:35gives me the choices. [music] 9 terab of
- 17:38footage. That's 9,000 gigabytes.
- 17:42Now, how the do you make something
- 17:44like that?
- 17:46Where do you even start? Where do you
- 17:49even start? If you're lucky like me,
- 17:52Isabelle will connect you with Jake Van
- 17:54Clee, who is the absolute best person on
- 17:58planet Earth to teach you how to build
- 18:01something like this using AI. So, having
- 18:04gone through this, I am as dumb as you
- 18:07can be and still be able to maybe work
- 18:10with this stuff. I want to try to
- 18:12explain [music] to a non-technical
- 18:14person like me. I don't really know
- 18:16computer stuff beyond editing. I want to
- 18:18explain [music] how Jake envisioned this
- 18:21process such that I without knowing any
- 18:25of the computer words like data cleaning
- 18:28or vector database. Jake was able to
- 18:31explain to me what I would in turn have
- 18:35to explain to an AI model in order for
- 18:37the AI model to make this footage
- 18:39crawler. I want to try to explain that
- 18:41process to you
- 18:43without boring you to death. God help
- 18:46me.
- 18:47>> But realistically, the three things that
- 18:48you need to know,
- 18:50>> I need a bunch of frames and I need
- 18:53descriptions of those frames attached to
- 18:58a video in a text file. Overall, the
- 19:01problem we're trying to solve is to
- 19:03reduce the nine terabytes of video data
- 19:06into data small enough for a computer to
- 19:09search through quickly. Video is
- 19:12[snorts]
- 19:14huge. One minute of video shot on shot
- 19:18on this Canon 1DX is five gigabytes. One
- 19:23minute, five gigabytes.
- 19:27I have 451
- 19:29hours of footage. Granted, it's not all
- 19:32in 4K. So Jake understands the strategy
- 19:36of how to reduce this amount of data and
- 19:40it's absolutely ingenious.
- 19:43>> So if you say I want this video, well it
- 19:47needs instead of it going through
- 19:49searching through every video figuring
- 19:50out these images, you're doing that hard
- 19:52work up front. You're freezing what
- 19:56those description are in time, taking it
- 19:58away from visual data and turning it
- 20:00into text data. And now text data much
- 20:04easier to search through, much simpler,
- 20:06much more efficient.
- 20:07>> Instead of making the computer search
- 20:09through gigantic video files, we're
- 20:13going to create a database of text
- 20:16descriptions of the video files. So the
- 20:20computer will be searching through tiny
- 20:22word data, not enormous video data. A
- 20:26picture is worth a thousand words. A
- 20:29word is worth 1 1,000th of a picture.
- 20:32It's actually less than a thousandth in
- 20:35computer data. My 9 terabytes of video
- 20:39is only 12 megabytes of description.
- 20:43750,000th
- 20:46the amount of data to describe videos
- 20:50rather than watch them. In mass, it's
- 20:53the difference between a paperclip and
- 20:56an F1 car. A paperclip weighs 750,000th
- 21:01the amount of Max Versstappen's Red
- 21:04Bull. We're going to reduce the 9
- 21:06trillion bytes of data the computer has
- 21:08to look through to 12 million bytes of
- 21:13data of information the computer has to
- 21:15look through by creating a database of
- 21:18text that labels each video. Computer
- 21:22don't got to look through the video.
- 21:24Computer just got to look through the
- 21:25words. But how do we get the words? How
- 21:29do we label the 48,000
- 21:32videos? motorcycle,
- 21:35city, night, ladder, hammock, snowfall
- 21:40times 48,000. How do we do that? We use
- 21:44the doomsday machine. AI, baby. An AI
- 21:48model is going to look at the videos and
- 21:50describe them. And in the description is
- 21:53the name of the video file, too. So, now
- 21:55the video is searchable by file name and
- 21:58descriptive words. Okay. So, we use AI
- 22:01to watch the videos and create a giant
- 22:04database of video labels or tags or
- 22:07descriptions, whatever you want to call
- 22:09them. So, are we going to make this AI
- 22:12watch 450 hours of video and write
- 22:16descriptions? How many tokens is that
- 22:18going to take? How much is that going to
- 22:20cost? Tokens is like the electricity the
- 22:24AI uses when you use Open AI or Claude.
- 22:29That's what they're charging you for,
- 22:31the tokens. Probably tens of thousands
- 22:35of dollars in tokens to have one of the
- 22:37Frontier models watch and label 9
- 22:41terabytes of video. Jake had two
- 22:44brilliant ideas for that.
- 22:46>> And you actually just pull like two or
- 22:49three frames from every single video.
- 22:51>> And that's brilliant idea number one.
- 22:53Yeah. Yeah. I ain't got to watch the
- 22:55whole video, just a few frames. So, we
- 22:58told this software called FFmpeg to pull
- 23:02the 50th frame, which is 2 seconds in,
- 23:06pull the frame that's 3 seconds from the
- 23:09end of the video, and then a frame every
- 23:1130 seconds in between them. Most of my
- 23:14clips are less than a minute long. So,
- 23:16FFmpeg
- 23:18only got to pull a few frames per video.
- 23:22So, that's half the solution to the
- 23:25token problem. We're reducing the frames
- 23:29with fffeg.
- 23:31>> Um, and then what you could end up doing
- 23:33is just having like a decent vision
- 23:35model. Go through, look at the image,
- 23:38give a description, throw that
- 23:40description onto um, uh, just a little
- 23:43markdown file or something simple like
- 23:45that.
- 23:46>> But now we need a model to watch and
- 23:49label those frames. And that's a lot of
- 23:52tokens. You're going to get probably
- 23:54like 20,000 minutes of use time, right?
- 23:58Billing minutes is a lot of them build
- 24:00by minutes. I mean, that's going to be
- 24:02four figures.
- 24:03>> Still a lot of data to sort through. Got
- 24:05to look through all of them frames and
- 24:07describe them. About two or three days
- 24:10of compute, letting the computer just
- 24:12run and chew and chew. Two or three
- 24:15days. That's a lot of tokens.
- 24:17>> Um, so the local tool process is
- 24:19definitely what I would recommend.
- 24:21Sovereignty is part of the narrative.
- 24:23>> Yeah. Okay. Then yeah, you're going to
- 24:25want to do it local.
- 24:26>> So the second brilliant idea to save
- 24:29tokens that Jake gave us, a local model.
- 24:33Download a local model. It's just
- 24:35software. It was called Moon Dream. A
- 24:38local model is a model that lives on my
- 24:41computer. Download it for free. My
- 24:44computer does all the work. No tokens.
- 24:48So, I asked Claude, "Findy me a free
- 24:50local vision model." Claude had me
- 24:53download
- 24:54to host the model and Moonream, the
- 24:57vision model itself that identified and
- 25:00labeled the video files. Don't use the
- 25:03Frontier models data centers.
- 25:06Took two days of constant compute and I
- 25:09used Isabelle's Mac Mini for it. And
- 25:11then I was basically done and Claude
- 25:15walked me through the REST OF IT.
- 25:17>> [screaming]
- 25:18>> OKAY. SO, I'M HAVING THIS feeling and I
- 25:21don't know if you remember this feeling
- 25:22and maybe you get this feeling doing
- 25:24something different, but it's the
- 25:26feeling
- 25:27when
- 25:29you're learning how to ride a bike and
- 25:32who's ever teaching you pushes you and
- 25:35lets go of the seat and you're doing it.
- 25:38You're like, "Oh my god, I'm I'm riding
- 25:39a bike. I'm riding a bike." And then you
- 25:41crash. [laughter]
- 25:42But it's that feeling. Okay? And that's
- 25:45the feeling I just got. I did one test
- 25:48of the footage crawler and I have to
- 25:51tell you how it like what happened. So
- 25:53it had me test dog. It said type in dog
- 25:57and run it and then test one of the
- 25:59clips. So this is what happened. See
- 26:02where the arrow is? I typed in dog there
- 26:04and then I pushed play. I pushed play.
- 26:08So I saw this, right? And I said, "Ah,
- 26:11there's no dog.
- 26:13There's no dog in this frame. It saw
- 26:17Lua's dress
- 26:20and it kind of looks like a dog and I'm
- 26:23like, "Ah, I'll just change the tags and
- 26:25I'll like figure out how to correct
- 26:27that." And it's a time lapse. And so I
- 26:29kept watching
- 26:33and look, I mean, it it went by so fast
- 26:36you won't even be able to see it right
- 26:39here for like How many frames is that?
- 26:42like five frames. Unbelievable. So, like
- 26:46when I clicked that, I got that little
- 26:48taste, like a little eyropper of that
- 26:51letting go of the bike. I got it. The
- 26:54best I've ever gotten it as an adult was
- 26:56when I when my son, it was so weird.
- 26:58Mirror neurons. I let my son go and he
- 27:01got it for a second and it was just
- 27:02like, it was that exact sensation. It
- 27:05was so crazy. It was like asset or
- 27:08something.
- 27:10baseball field. Boy riding a bike.
- 27:14[music]
- 27:20Wow, this is better than last time.
- 27:24You're getting it.
- 27:27You're supposed to wipe out, so don't
- 27:28worry about that. [music]
- 27:30And you crashed very well.
- 27:35[music]
- 27:47underlined years ago, chosen at random
- 27:50today.
- 27:56Art doesn't get made on the clock, but
- 27:59it can get finished on the clock.
- 28:06Join a paid tier of the Patreon to get
- 28:08in touch with me or to avail yourself of
- 28:10this channel's alternative universe. You
- 28:13can click this. It'll take you right
- 28:15there.
About this transcript
This page contains the full transcript of Tickets, A Necktie & 26 Years of Footage: Long-Form 3 by Van Neistat, generated from the public captions YouTube serves with the video. The transcript has 3,585 words across 557 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.