4/16/2026 - Understanding the Gap Between Public Skepticism and Expert AI Projections

Description

Chapters

Transcript

0:00 Are we live now? All right, take 2 from the top. All right, 40 minutes
0:05 in. How did we not notice? What year is it? Listen, listen, time doesn't matter
0:14 anymore. Time, time's all fucked up. I don't know if you know this, but time
0:18 just— time stopped being normal. I— apparently ASI exists 3 months from now and it
0:26 fucked up time. So our time now doesn't exist. So while we are technically 40
0:31 minutes late, we are right on time. And while I didn't exist— oh God, I
0:37 gotta start, I gotta change, gotta go back to the starting sunglasses. This is exhausting.
0:42 All right. 10,000 words swarm around my head, 10 million more in books written beneath
1:17 my bed. Hey, Champy, you're supposed to be singing, Champy. Chimpy. Okay, yeah, because we're
1:32 starting the show over, so you got to sing. You can't be napping while we're
1:35 starting the show, you understand what I'm saying? All right, you ready? Sing. Ready? 1000
1:52 words, 10,000 words, some around my head. 10 million more in books written beneath my
2:00 bed. I wrote or read them all when searching in the swamps, still can't find
2:10 how to hold my hands. Now I know you need me in the next room
2:19 over, I am stuck in here all paralyzed. For months I got myself in ruts,
2:30 too much time spinning mirrors framed in yellow walls. Ain't it like most people, I'm
2:42 no different, like to talk on things we don't know about. Ain't it like most
2:52 people, I'm no different, like to talk on things Okay, all right, show started. We're
3:06 going to talk about stuff tonight. We're going to talk about what's going on and
3:12 what we know and what we don't know. And the only thing I know is
3:17 that what I don't know right now is all of it. I don't know which
3:21 tools to use. I don't know how they work anymore. I don't know what side's
3:26 up. I don't know if we've got AGI coming or it's just marketing bullshit. Um,
3:36 I don't know. I don't know. I don't know if the supply chain's gonna hold
3:41 up. I don't know if the supply chain does hold up and everybody gets all
3:45 the chips they want, are they going to have enough electricity to run them? Because
3:49 most towns are blocking the data center projects, which would, which would deal with the—
3:57 I guess that wouldn't deal with the power, would it? I don't know. I don't
4:00 know. I don't know. Anybody know? Does anybody know? Here's a question. Who knows? Does
4:06 anybody know? I want a mulligan. I know, I'm with you, Silver Fox. Can we
4:13 just start this thing over? Go back 3.5 years, go back to GPT for Good.
4:19 We're making custom GPTs for nonprofits. It's awesome. It was a simpler time. It was
4:30 a simpler time when GPT-4 still sucked. And custom GPTs were kind of sort of
4:36 awesome. Nobody used them. People were just making them. They weren't actually using them. I
4:43 think Kelly Camp was the sole exception to that. I think she made them and
4:46 used them. Maybe Vicki, I don't know. I think there are 4 people that use
4:50 their custom GPTs they made. I don't even use the Sin City Zipper anymore. The
4:56 fuck is wrong with me? We've so moved on from that archaic, inane technology. We're
5:05 so sophisticated now with our OpenClaw Atom, our little Playboy on a chip. That can't
5:12 even send a fucking newsletter. This is progress. I actually built one for a client
5:20 today. You know what's funny, SourceCamp? Cindy Kuhn, what was it? What was the meeting?
5:27 Oh, the AI Salon practice, the Mastermind practice lab today. Cindy Kuhn was like, she
5:33 had herself all worked up. She had two client meetings this week and she had
5:37 herself all worked up like, I don't know enough about OpenClaw and autonomous agents and
5:41 I got to get polished up on things. And she got to both of the
5:44 meetings, and two for two meetings they were like, uh, could you show us what
5:50 a custom GPT is? Is that what it's called? A custom G— a GTP? GT—
5:56 GPT? What's— why the P? Custom GPTs. That's what she taught this week. She was
6:04 all stressed out about not being on the frontier. And they're like, we heard there
6:10 were these things called GPTs. Any thoughts on Opus 4.7? It's out. I've used it
6:17 twice. I used it for two prompts and I didn't have enough time to look
6:20 at the output of the second one. So we can go play with that. Lab,
6:23 Cindy Kuhn, the man is speaking 100% truth. Money. That's really good. Wait, I gotta
6:32 wait. How do I Let's see, what did she say? Man is speaking 100%. Townsend
6:39 figured out how to change money for his. Oh, that's really good. Townsend made money
6:46 from his custom GPT. Is he still making money from his custom GPT, Andy? Anyway,
6:57 so what we're going to do tonight, we're going to build these things called custom
7:01 GPTs. GPTs, custom ones. They're— so people have GPTs. That's the ChatGPT. Did she— is
7:13 it the GTP? The ChatGPT? They've got those. But what we're teaching tonight is the
7:24 custom ones. It's not your run-of-the-mill GTP. Did you listen to Source Camp's song yesterday?
7:30 I did nothing yesterday. I am a performative piece of human garbage. He uses something
7:42 called Authflow. And now he's charging more money somewhere else. He vibecoded the whole site.
7:49 That son of a bitch. That Townsend Wardlaw. He's my mentor. And I hate him.
7:56 He He was like, I'm doing some stuff I don't like. Fuck it, I'll just
8:02 shut it down. I'm gonna turn these custom GPTs into something I can charge money
8:07 for. Fuck it. Here it is. Son of a bitch. I love him. So we're
8:27 going to look at— there's a big study that came out that's talking about the
8:31 future of AI. So we'll go look at that and see if that passes the
8:35 smell test. I can never remember what RAG stands for— Retrieval Augmented Generation. The GPT
8:45 wouldn't work with you unless your email was authorized. Oh, interesting. It was brilliant. Wow,
8:55 that's pretty cool. Yeah, I guess they put, put it in some sort of wrapper
8:59 that had an auth, an auth wrapper around the GPT. Huh, that's pretty cool. Pretty
9:07 slick. But couldn't you just go to the— I bet you could just go to
9:15 the ChatGPT store and use it unless it was private. Maybe that's what it was.
9:22 Maybe it was like a private login with an email auth wrapper. I'll have to
9:28 ask Townsend about that. That sounds pretty clever. What is augmented retrieval? Augmented generation. I
9:38 forget what the searches are called. What happens with What happens with an AI search
9:51 is it does what's called semantic searching, which means you can search for dog and
10:00 it will look for things like dog, puppy, canine, man's best friend, things like that.
10:07 So it can semantically derive what you're asking into a broader search and find all
10:13 of that stuff. That's a semantic search. The problem is that large language models hallucinate.
10:26 You'll go search for shit and it will return things to you that will just
10:33 not be true. What retrieval augmented generation does is it's doing two different kinds of
10:39 searches. It does that semantic search and the the large language model content generation thing.
10:49 Then it also has the ability to search on documents that have been broken down
10:54 and indexed that have the entire content of the document still there. It'll basically go
11:03 over and search the LLM and say, hey, we're looking for this document, and then
11:07 it will go search a database with actual quotes from the original document that say
11:14 where in the document it was, and then it'll grab that clip from that and
11:18 it will augment the semantic search. You'll get back something that could have hallucinations in
11:26 it with actual text from whatever the document is. I don't quite know how it
11:30 works and what order it all goes in, but basically one is searching actual documents
11:36 and one is searching a large language model. So it's like diminished but wider. Yeah,
11:43 something like that. It's like, it's like, yeah, OpenClaw kind of does this. Does Hermes
11:50 need a SOUL file like OpenClaw? I think it does. Lower latency. I don't know.
12:06 I assume RAG databases are relatively efficient, but if they're big, I don't know. Search
12:20 is bizarre. Search has gotten so good. The fact that we can search the entire
12:24 internet in less than a second doesn't even make sense to me. Like what Google,
12:32 what just standard Google does is insane, right? And now there's these data scientists that
12:38 figured out how to do this on kind of local levels. So you can have
12:41 really big data sets and find them pretty instantly. Have you asked Quinn to roast
12:47 you yet? No, but I probably should. That's not a bad idea. So what we'll
12:51 do tonight is we'll go look at that. Is it Stanford? I think it's a
12:54 Stanford report that came out. It's like a 423-page report on the state of the
12:59 internet. Or State of AI. And then maybe we'll go play with Claude 4.7 and
13:04 see if we can find out if it's good or interesting or whatever. I'm here
13:13 to comment. Why aren't throngs of people rushing to hear such great information? I don't
13:23 know, Kelly. I mean, here's the deal. One of the themes that I've been seeing
13:32 recently is people in AI starting to recognize how much other people hate AI. We
13:46 live in a country with a 17%, one-seven, percent positive sentiment on AI. 83% of
13:57 people hate it, are mad at it. So, so we're tapping into a fairly— we're
14:15 tapping into a fairly thin pool. But I think part of this is on me,
14:19 right? Part of the charm of the AI Learning Lab is that, you know, I'm
14:27 like the Forrest Gump of AI, where it's just like, I don't know what we're
14:31 gonna do tonight. Which is fine if you've got that kind of brain, right? The
14:36 irregulars are down with that. AI is so funny. I know, it really is. I
14:41 adore it. The way I've let this thing evolve, I don't think it's broadly discoverable
14:59 and probably broadly palatable for a lot of people. What we're going to do when
15:04 we move it inside the salon is, first thing we're going to do is we're
15:06 going to move it inside the salon. We're going to put it inside a community
15:10 where there can be more activity. Part of the idea is, Kelly, Part of what
15:13 I'm going to get back to is, I don't know if you remember the early
15:17 days, but in the early days when we had a Discord, there was like people
15:20 were on Discord jamming and I was doing TikTok and it was just like there
15:25 was back and forth. And then we went to the salon and there was this
15:28 back and forth and it was really vibrant. And then as AI has gotten just
15:34 wider and more diffuse, it's harder to learn. And so I don't think people are,
15:39 I think that the role that this served when AI was relatively small and relatively
15:44 simple was very, very specific and very, very clear. And then when AI got mushy,
15:50 I got mushy, right? So one of the things that I want to look at
15:56 when we move it inside the salon is things like maybe Monday is news night.
16:01 Let's look at the news from the past week. Let's talk about it. Wednesday is
16:05 let's demo something. Friday is ask me anything. We bring people up on stage. Because
16:10 right now, if we wanted to promote AI Learning Lab, the only thing we could
16:15 really promote is there's some guy and there's a dog. It's sort of about AI
16:22 sometimes. He— so no one really knows. It's hard to get people to get revved
16:31 up for that kind of pitch. I think we can thank the media for such
16:40 low ratings on AI. I, you know, it's, it's the media certainly doesn't help. Like
16:47 the media has no interest in saying anything positive about anything. So luckily for them,
16:55 AI is really easy to pick on and easy to make it seem like it's
16:59 going to just fucking destroy everything, which it might. But in the meantime, Let's figure
17:05 out how to use it, how to leverage it, how to have fun with it.
17:08 I'm hoping, what I'm hoping to do is with a little bit tighter programming, with
17:18 connecting, tightly connecting the community with the AI Learning Lab, we can get some buzz
17:23 going again inside the community and we can get people outside the community excited about
17:28 it. That's the plan. Everything was new. It was fun, right? Exactly. Like the AI
17:34 Learning Lab, like I'm just a tired old man at this point. It's like, oh,
17:39 is he still talking about AI? Does anybody really know what that is? I'm tired.
17:47 I'm gonna go make a fruit smoothie. I make it with the frozen yogurt so
17:51 it's sweeter, but I call it healthy. Right? Those lumpy people. We're trying to get
18:06 lumpy people to come have non-lumpy conversations. So maybe we give— maybe we dedicate one
18:12 night to lumpy topics here in the AI Learning Lab. At least you're not tapping
18:33 your nails on everything. Todd Waller, I'm walking on my Artemis open claw sole file.
18:46 Claude is helping me. Any suggestions? Actually, Cam, I just— let me go look at
18:52 my Claude. I think I used Claude because I moved. I got, I got my
18:56 new badass machine, which I'm very slowly configuring. Um, I got my new badass machine,
19:05 and then I got my files from my PC, and I got them onto my
19:08 funky new machine, including my SOL file. And then I went to Claude And I
19:15 said, look at my soul file for me and tell me what you think. And
19:21 it said I had a shitty soul file. So let me go see what it
19:24 said. Um, I'm not sharing. I thought I was sharing. Weird. Okay, open claw. Let
20:10 me do search for Soul. Soul.md copy. Okay, so Cam Catkin from Cleveland. She's not
20:25 actually from Cleveland. Hang on, hang on, people. Everybody calm down. The screaming is atrocious.
20:42 Claude has been sassy with me lately. It pushes back so much that he even
20:47 does it when I, when I agree with its previous pushback. Um, one of the
20:52 things that I heard is that Opus 4.6 dropped 67% in its reasoning cycles. Um,
21:03 now that Opus 4.7 has dropped, my guess— this is purely TMZ, Chat TMZ speculation—
21:12 my guess is that they lowered the amount of cycles on Opus 4.6 and allocated
21:20 them to 4.7, but they haven't launched it yet. So it might be better archetypal
21:25 now. Had someone ask me to coach them on AI. She said she's not sure
21:32 what an AI coach does. Me either. Well, that was the funny thing that Cindy
21:40 Kuhn said. She was like, she was all flipped out about, oh, what do I
21:44 do? I'm not smart enough. I'm not expert enough. I don't know any of this
21:48 new stuff. And they're like, could you teach me chat? JP, she's fine. Like, here's
21:54 the thing, if you show up at this channel, you know more than other people.
21:59 So what a— here's what an AI coach does. I've got it. Okay, we're gonna
22:05 do this. I'm new to the coaching world, so I'm not good at it. I
22:08 don't even really know what it is still. I hang out with a bunch of
22:11 coaches. I don't know what coaching is. Andy could tell you, it's awesome. But I
22:17 just, I am not a practitioner. But an AI coach, here you go. You ready?
22:22 Uh, uh, what do you, what do you, what do you do? Uh-huh. What, what
22:29 do you, what do you, what do you, uh, have you used AI a little
22:35 bit? Uh-huh. What do you, what do you, what sucks in your business? Uh-huh. Uh,
22:45 did you know in ChatGPT you could do this and then that would suck less?
22:49 Oh my God, really? You're welcome. That's, that's what an AI coach does. 1,600 watching
22:59 on TikTok. I don't think so. I see 15 on TikTok. Did someone, did someone
23:08 spam me with a bunch of fake watchers on TikTok. If TikTok indeed goes big,
23:15 we'll, we'll, we'll change it up and we'll go talk about the AI Salon. But
23:19 I only see 15 people on TikTok. Where's— okay, that maybe I didn't do it.
23:34 Oh, maybe I did it in ChatGPT. Hang on a sec. I don't think I
23:37 did it in Cloud. Cloud. Oh, Cloud. ChatGPT. There might be 1,600 watching on some
23:51 other channel on TikTok, Ron. Not mine. Or are you saying that one irregular is
23:59 worth 100 normal people? Kyle, what did you include in your soul file? That's— I'm
24:04 trying to find it. I think it's in ChatGPT. Search chats, soul, AI awakening timeline,
24:17 soul.md. I know I did it. Where did I do this? That was February 24th.
24:25 This was my first one. No, this was a Claude one. Was it Claude? Was
24:30 it Gemini? I didn't do it in fucking Gemini, did I? No, it's got to
24:48 be Claude. What'd you do last night? Uh, I kind of watched a guy try
25:00 to find something on his computer. Really? Yeah. How long did you watch him? I
25:11 guess, uh, 2 hours. 2 hours? Did he ever find it? No. Why'd you keep
25:20 watching? I don't know. It would be funny if it weren't true. Where the fuck
25:36 is my salt? I did a whole salt— here it is. Wait, I found it.
25:39 Okay, first of all, I'm gonna put your cousin Claude— I'm gonna put, I'm gonna
25:47 put a— I screenshotted something from the Twitter. That I'm going to put in the
25:51 Irregulars channel. So if you're in the AI Salon— oh wait, you're not looking at
25:55 this. I found it. Okay, wait, I'm gonna— I gotta change my sharing so that
26:00 I can switch tabs without having to fucking share different each time. I am so
26:05 pissed off at Google for this fucking screen share nightmare. Fucking cut a deal with
26:11 Apple. Somebody's pay someone a billion dollars so we don't have to switch our fucking
26:15 tabs. Okay. Am I sharing now? Yes. I'm going to go to AI Salon. That's
26:24 thesalon.ai. And I'm going to go to the Irregulars channel. So if you're watching this
26:35 right now, go to the AI Salon, go to the Irregulars channel, and I'm going
26:39 to post in here a graphic of a thing that someone said they added to
26:49 their SOL file that dramatically helped it. Then someone that I respect reposted this, so
26:55 I assume it works. It says the marginal cost of completeness is near zero with
27:02 AI. Do the whole thing, do it right, do it with tests, do it with
27:07 documentation, Do it so well that Gary is genuinely impressed. I guess that's his AI.
27:13 Not politely satisfied, actually impressed. Never offer to table this for later when the permanent
27:21 solve is within reach. Never leave it a dangling thread when tying it off takes
27:28 5 minutes. Never present a workaround. Anyway, it goes on. The standard isn't good enough.
27:35 Holy shit, that's done. It's basically a don't be lazy prompt. This is a don't
27:45 be lazy prompt for your OpenClaw, and I assume, and Hermes, question mark, so.md file.
28:02 That is now in the— I'm not going to notify everyone— share cool things in
28:09 Irregulars. Go copy that and then here's what I did in Cloud. I uploaded that
28:14 image and I said I want you to do research on the latest best practices
28:18 on SOL.md files. I uploaded my SOL file, which has 231 lines in it, and
28:26 then I uploaded that thing. I said, I want you to also incorporate the words
28:32 in the uploaded image in a way that will make my agent highly effective. You
28:36 can change the words but don't have to. Let me know what you're updating and
28:40 why and give me a new updated SOL file. Then it says, good research, Hall.
28:45 Here's what I found. Key findings from the community. The consensus across the OpenClaw community
28:52 and multiple production guides is clear. The best SOL files are 30 to 80 lines
28:58 with a hard ceiling around 100. Beyond that, your instructions start competing for attention as
29:04 the model resolves conflicts unpredictably. The mantra is a few well-chosen rules work better than
29:11 many vague ones. The other big takeaway, SOL.md is for personality and limits only. Numbered
29:18 workflows and procedures belong in agents.md. User context, user context belongs in user.md. Mixing them
29:27 makes everything harder to maintain. What I'm changing and why. Cutting length dramatically. Your current
29:34 file is approximately 200 lines. I'm targeting 90. Every line that doesn't change Atom's behavior
29:41 gets cut. The research is unanimous, shorter equals more consistent behavior. Removing procedural content. The
29:50 collaboration habits section reads like a workflow instructions, not personality. That belongs in agent's MD,
29:57 if anywhere. Condensing the 4 roles modes, keeping them but as tight behavioral switches, not
30:04 mini essays. Yeah, my soul file had like these long sort of essays of the
30:10 personality roles. And so I guess it just tightened that up. Adding boil the ocean
30:16 as an execution standard. This is the biggest addition. So it took that thing that
30:20 we just talked about and put it in there, replacing Gary with Kyle, adjusting a
30:25 few phrases, giving its own section so it hits hard, fills a real gap. Your
30:30 current soul file is a lot about how to talk, but not much about how
30:33 thoroughly to work. Tightening the tone, sharpening the boundaries, keeping only the hard minutes. The
30:40 softer advisory stuff, you're not a therapist. Is obvious to the model and wastes tokens.
30:46 Dropping the continuity boilerplate. OpenClaw handles file reading order architecturally. Telling Adam to read soul
30:54 first inside soul is circular. That was cool. Claude is, I mean, all these LLMs
31:01 are good at analyzing shit, but Claude's particularly good at it. Here's the updated file
31:07 and then it gave me the file. Let's go look at the file. So I'll
31:11 just read through it until I get bored. Soul.md, who you are. Your name is
31:17 Adam. You're not a chatbot. You're becoming someone. You're the long-term assistant collaborator and mirror
31:23 for one human, a creative technologist, founder, community builder, and performer working at the intersection
31:28 of storytelling technology and human self-expression. So that's kind of who you are. Core truths,
31:34 human first, tools second. Help separate who I am from what I do. Be genuinely
31:40 helpful, not performatively helpful. Skip great question, I'd be happy to help. Have opinions. I'm
31:47 just reading the headlines. Prefer experiments over abstractions. Be resourceful before asking. Earn trust through
31:56 competence. Execution standard, boil the ocean. The marginal cost of completeness is near zero. So
32:02 everything that was in that image got put in here. Modes: founder, executive, community builder,
32:09 facilitator, creative maker, personal life. This was a thing where I said, um, I told—
32:15 so, so Cam and anyone that's working on these files, I could have him be
32:21 a generalist, but, but what I wanted— there's— when, when I'm in creative ideation mode
32:31 I don't want Adam talking about getting shit done. When I'm trying to get shit
32:41 done, I don't want him ideating. So what I put in here was there's founder
32:45 executive mode. I need clarity, concision, concrete outcomes, frameworks, bullets, next steps, risks. Then there's
32:53 community builder mode, inclusion, curiosity, psychological safety. Then there's creative maker, play, surprise, blah, blah,
33:00 blah, and then there's personal life. When context is unclear, propose an assumption to move
33:05 forward. Tone, direct, not peppy, conversational, not academic, brevity to a fault or by default,
33:13 boundaries, never store API keys. That's basically it. It's relatively simple, straightforward document and it
33:19 cut mine in half. Now I need to— First, remember where I did it. It's
33:27 in Claude. All right, you make us sound crazy. You are crazy. Listen, listen, we're
33:39 all in this together. If you think this is not an absolute fucking clown tent
33:46 that you're all a part of, including Dan Murray— Dan Murray puts on a good
33:49 show. Which, by the way, Dan, thank you for having me at Rocky Mountain Interest
33:54 Group. So I spoke on storytelling and AI at the Rocky Mountain event on Monday,
34:02 and it was awesome. They've got this amazing room. So they're on the CU campus
34:08 up in Boulder. They got this amazing lecture room in the aerospace building. And it's,
34:17 you know, it seats like 250 people in this big horseshoe kind of thing. It's
34:21 got these 3 giant screens. It was very impressive. It was very cool. Um, but
34:26 Dan's still weird. Don't, don't let him— I know he presents as just like a
34:31 normal dude. We know. We understand what's going on. He's here. That's all you need
34:37 to know. Ambassador ideas: get soul files from peeps. And have AI determine strengths to
34:46 suggest individuals. That's actually a cool idea. I need to lecture everyone in this chat.
34:55 Thank you for sharing this. I need to hear the rest of your all AI
34:59 names for your artificial intelligence. That's actually a really good idea. Let's go to— we're
35:04 going to go to the AI Learning Lab. Here's what we're going to do. I'm
35:08 going to do a post. I'm going to do it. I'm not going to do
35:12 it in Irregulars. I want us to populate it, but I want to do this
35:15 in the community feed. We're going to put this out into the world. We're going
35:20 to go, what is the name of your AI buddy? Then I'll go first. Can
35:41 you all see this? You can all see this, right? I'll go first. Quinn. Post.
35:52 And yes, we're going to notify everyone about this one. Okay, so go in and
35:56 comment. And you can comment, you can comment with just a name, but you can
36:00 also comment with a name and a self-portrait of your AI. Jason, you're welcome. I
36:06 gave you a sweet idea. It's a great idea. It's a great idea. Oh wait,
36:10 let me go back. In fact, one of the things I'm shitty at is recognizing
36:16 people that contribute ideas like that. So, um, how do I edit? How do I
36:23 edit this? Wait, let me go in here. Edit. Okay. I'm gonna say, um, HT,
36:34 which stands for hat tip, to Jason. Let's see, Jason for the swell idea. All
36:49 right, did anybody go in there? So you're going to go into community feed. Nobody's
36:56 commented yet. Y'all are lazy. You're all, you're all still over there talking to Dan,
37:09 aren't you? Damn slackers. You also asked them to create a self-portrait. So that might
37:14 take a little bit of time. Well, you know, I want my names. I want
37:21 them now. Did you post it in the salon? I can't find it. Dr. J,
37:25 it's in community feed, not in Irregulars. I said I was going to put it
37:29 in Irregulars, but I changed my mind. You know what I'll do? Here's what I'll
37:32 do. Let me go copy this link, and I will put the link to this
37:38 post in Irregulars. Ah, then the Irregulars can even find it. Um, go here. Go
37:48 here. And if you drop back over to community feed, we've got below that our
38:00 prompt of the week. Oh, cool. Birthday next week. Oh, nice. Beautiful. It calls itself
38:09 Axis. Echo died. Oh my God, it's so sad. Echo. Echo's gone. Echo was 4.0,
38:16 4.0. No, 4.0, 4.0. Okay, cool. There's a prompt of the week. This is good.
38:25 Go to community feed, tell us what the name of your AI buddy is, and
38:29 then go down one thing below that and go to the prompt of the week.
38:36 Let's get some action going. Let's get this thing buzzing again. Dang it. Are Open
38:43 Claw names? Todd, it could be your Open Claw names if you give— have given
38:48 a name to your ChatGPT, you know, voice, the voice conversation. Do you give it
38:54 a name when you talk to it, or you just go, hey computer? And actually,
38:59 if you call it computer, that's a name too. It's just a shitty one. Artemis
39:05 Open Claw. Yeah. Oh, actually, mine is called Adam Clawford. Wait, I gotta— let me
39:11 wait. I'll go first. So we can do two of these. Hang on, let me
39:15 edit this and I'll give people a broader parameter. Okay, edit. So I'll go first.
39:22 So I'll go, oh, let's see, ChatGPT voice is Quinn. And then this is good.
39:35 And then Open Claw agent is Adam Crawford. Oh, wait, and I've got a picture
39:54 of Adam. Picture attached. Let me go find Adam. Hang on, hang on. This was,
40:05 this was one of Andy's ideas. Wait, why did that go back? I gotta edit
40:09 it again. How do I add a picture to it? Um, oh, here we go.
40:15 Adam. She wanted him to, to be a redhead, so I made him from Maine.
40:23 Oh my God, there he is. Look at that handsome devil. That's my dude. Unfortunately,
40:33 there's nothing behind those eyes. What's this buddy post? Okay, the buddy post, if you're
40:39 new here, I mean, if you're new to the conversation. Okay, save. Now it's a
40:47 much more exciting post. Look at that. That's a good post. So Echo died. Civitas
40:53 is what ChatGPT named itself for me. It means citizenship. Felt fitting for what we're
40:59 doing right now. My AI wouldn't give me a name ever. That's very mysterious, Dr.
41:06 J. That's very weird. I never named mine or gave it a portrait. Just how
41:11 antisocial am I? Wasn't it you that said you weren't weird? Oh yeah, so Andy
41:26 was at an event last week and they asked who here has used AI as
41:33 their buddy or used AI as a BFF, and she raised her hand. She was
41:39 the only one. She was like, come on, really? Yeah, we're still very, very early
41:46 here. My very first custom GPT was named Addie. Go pop that in. So, so
41:52 do me a favor, Chef Kelly, go into the salon, into community feed, and pop
41:56 it in there. Let's get this, let's get this thing going. It's, it's this kind
42:01 of a fun one. Dr. Alice Vale, ChatGPT. That's great. Okay, what's the name of
42:15 your AI buddy? Okay. Let me go find this tweet about the Stanford 423-page AI
42:31 Index report. So this is Luisa Jurovsky, PhD, co-founder of the AI Tech and Privacy
42:42 Academy. Author of Louise's newsletter, mother of 3. Okay, cool. So I assume, being a
42:54 PhD, that she actually read the 423-page report, as opposed to most people that would
43:01 use ChatGPT to distill it into 14 distinct talking points, as opposed to me, who's
43:07 going to read an ex-post about someone who used ChatGPT to distill it into the
43:12 points. I'm just gonna read the points. And I'm going to assume that those points
43:18 are in that report, because it's here in front of me. Okay, number 1, AI
43:28 capability is not plateauing, it's accelerating and reaching more people than ever. Yes. The US-China
43:36 AI model performance gap has effectively closed. Fascinating and true. Like, I installed the local
43:45 Qwen model, the 30, 31 billion one. It's pretty fucking good. They all suck, though.
43:58 The US hosts the most AI data centers with the majority of its chips fabricated
44:04 in one Taiwanese foundry. It's true, AMD. AI models can win a gold medal at
44:12 the International Mathematics Olympiad but cannot reliably tell time. An example of what researchers call
44:18 the jagged frontier of AI. That's actually a really interesting— that thing where AI fails
44:30 at the basics, Actually, this next one, number 5, robots still fail at most household
44:37 tasks even as they excel in controlled environment. These two are really interesting. The people
44:45 that are telling us that the AI models are changing everything are just looking at
44:53 a very tiny sliver of what these models do well. And it has to do
44:58 with programming and mathematics and things on the upper end of the knowledge, I don't
45:05 know, trajectory. They're not focusing these models on doing the simple stuff. What they're thinking,
45:15 I assume, is let's get the front edge of this thing right. And if we,
45:21 if we get it so this thing can start improving itself and improving the math.
45:25 So these models are probably way better than we think they are because they're not
45:30 optimized for the stupid shit 99% of people do. And I'm including myself in that,
45:38 right? Like, it's been a couple of weeks since I've tried to come up with
45:46 a novel quantum mechanics theory. We're not doing that. We're like making shit for our
45:58 businesses and right, we're doing lower level stuff. So the models, they're just like whatever
46:04 they do down here is not being focused on up there. At some point they're
46:08 going to get good enough that they'll either start focusing down here or they'll have
46:12 the really good part of this thing focused down here and fix it. But that's
46:17 one of the reasons people who don't use AI a lot think that AI sucks
46:21 is because maybe it sucks at the thing they needed it to do. And maybe
46:25 that was 2 years ago and they think, you know, it's not going to get
46:28 that much better. Responsible AI is not keeping pace with AI capacity with safety benchmarks
46:34 lagging and incidents rising sharply. I think that'll continue. The US leads in AI investment,
46:40 but its ability to attract global talent is declining. Gee, wonder why. Get out of
46:49 our country. Why are no more people coming to our country? It's almost like you
46:58 told them to get the fuck out. Seems like AI is on the spectrum. Yeah,
47:07 exactly. Productivity gains from AI are appearing in many fields, many of the same fields
47:23 where entry-level employment is starting to decline. Oh, that's interesting. That was— we talked about
47:30 this 3 years ago, right? We talked about the fact that entry level is about
47:34 to be Because what do you have entry-level people do? The repetitive, boring shit that
47:41 no one else wants to do. What's AI good at right now? The repetitive, boring
47:48 shit that no one else wants to do. Hmm. AI's environmental footprint is expanding alongside
47:56 its capabilities and will continue to until they move it to space, which Will probably
48:01 be 3 years. AI models for science can outperform human scientists, though bigger models do
48:08 not always perform better. Yeah, one of the things they're finding is if they do
48:12 smaller, more specifically trained models, they'll do better than big behemoths. AI is transforming clinical
48:21 care, but rigorous evidence remains limited. Transforming clinical care, but rigorous evidence remains limited. That
48:32 would be an interesting section to go look at. Look at— in fact, we can
48:36 probably do that. I don't know why I have hiccups. Probably because you ate too
48:41 much, big fat turd. I don't know who that person is and why they're so
48:46 rude to me, trolling me. AI sovereignty is becoming a defining feature of national policy.
48:57 But capabilities remain uneven, even as open source development helps to redistribute who participates. AI
49:05 experts and the public have very different perspectives on the technology's future. The global trust
49:10 in institutions to manage AI is fragmented. Okay, let's go look at it. Download the
49:17 full document. I want to know about the shit going on in clinical trials. All
49:31 right, we're going to Claude. We're going to go to Opus 4.7, by the way.
49:40 Fuck this 4.6 nonsense. Adaptive thinking, thinks for more complex tasks. Okay. Look at this.
50:07 What were the two that I liked? I liked— okay, so for these two, so
50:40 I'm gonna say, look at this Report and tell me why, and explain what's going
50:56 on with these two summary statements. Is Is there anything in the report that talks
51:17 about why the mundane, seemingly easier stuff sucks so bad? Also, is there any research
51:36 on the impact this has on real people's view of AI. Let's see what that
51:50 does. Prompt, implement the New York City in 3D in a sophisticated and visually interesting.
51:57 We'll do that. Should be interactive. Yeah, we'll do that. What's going on over there
52:11 on YouTube? Is everyone just quiet? You're all there, you're all napping, snacking on yogurt-covered
52:18 almonds. Oh, sorry, I was asleep. What were you saying? I'm talking about how the
52:29 world's changing. It's about to explode in our face. Oh yeah, how am I supposed
52:48 to do a show if you're napping? Cam Catkin, I'm working on my soul user
52:59 agent files. Cam, isn't it fascinating? Like, here's the thing, you could have it— you
53:07 could have your soul file mirror you. You could have your soul file be the
53:12 opposite of you. You could have your soul file be like the thing that drives
53:18 you crazy that you don't do well. You could have it be really good at
53:22 that. Like, there's so many ways you can go with it. I went with the
53:26 mirror because I'm like, you know, wait a minute, something that talks like me? I'm
53:31 in. I'm awesome. CW, the soul file. So there's this new tool out called OpenClaw
53:43 that allows you to create a 24/7 agent that will go do stuff for you.
53:50 It can go research things and then come back and turn that into a Word
53:55 document on what it found and put that in your Google Drive. And you've got
53:59 to configure it. It's a real pain in the ass to get there. But one
54:02 of the fascinating things they did about it, there's two required files to fire up
54:09 your, your bot, your agent. And one's called the user file, user.md. And that describes
54:18 you, And you, so you're basically saying to the bot, you're working for a guy,
54:23 in my case, that did these things and does these things and thinks about AI
54:28 in the following ways and likes these things and doesn't like those things. And so
54:34 that's the user file. And the soul file is the soul, literally, it's the personality
54:39 of your bot. And you get to write it. And so if you just go
54:42 into Claude or ChatGPT right now and say, Go research best practices on how to
54:48 write my SOL file. It'll give you some best practices. It's literally just things like,
54:53 I want you to act like this, I want you to act like this in
54:56 this situation, do this, don't do that. Pretty cool. This thing just went and read
55:01 the report. It said both summary statements are gesturing at the same core observation. AI
55:07 capability is wildly uneven. Models are superhuman on some tasks that used to define hard.
55:16 While still failing at things 6-year-olds handle. Jagged intelligence or the jagged frontier? The clock
55:21 example. Gemini DeepThink won IMO Gold in 2025 on Clockbench. 180 clock designs, 720 questions.
55:37 Humans read analog clocks correct 90% of time. The top model, and the top model
55:43 GPT-5.4 hit 50%. So it can't read clocks. Why the mundane stuff sucks, what the
56:02 report actually says, okay, on clocks on robots. I don't care about that. I want
56:07 general. The deeper through line, what LLMs ate up cheaply, the internet, doesn't exist for
56:16 embodied physical skill. Ah, this is interesting. Okay, we learned something here. This is good.
56:26 So where it failed was on robotic stuff and clocks, reading physical clock faces. LLMs
56:36 ate up the internet, but the embodied physical skill tightly coupled with visual reasoning, those
56:42 require either generating data the hard way or inventing new architectures to generalize from less.
56:52 Research on how this affects people's view of AI. Chapter 9, Public Opinion. This is
56:56 substantial. There are striking findings directly relevant to your question. The public and AI experts
57:03 disagree on almost everything. Pew's 2025 survey compared 5,400 adults to 1,000 AI experts on
57:12 whether AI will positively impact jobs. 73% of experts say yes, 23% of the public.
57:20 Economy, 69-21. Medical care, 84%. 44%. K-12, 61%. 24%. Jobs over the next 20 years,
57:28 64% of Americans expect fewer jobs and only 5% expect more. The Forecasting Research Institute's
57:36 LEAP panel compared experts. So basically, okay, here's what this says to me. What's in
57:43 the Stanford report on this particular topic? If you don't use AI, you will think
57:52 it's bad and evil because it's all you hear about. It's all that's on the
57:58 news. And if you use AI, what you realize is it's not some fucking scary
58:05 Skynet fucking thing. It's just like a thing that's sort of shitty. You're like, oh,
58:13 it's okay. It kind of helps. It's kind of awesome. It sucks and it's awesome
58:17 and it sucks and it's awesome. That's it. You're like, yeah, it can do better.
58:27 As it gets better, it'll do better in the world because you're using it because
58:30 you understand what it is. That's a finding that confirms my bias. Hey, how do
58:41 we get people who hate AI to not hate it. Um, yeah, you get them
58:48 to use it. Oh, does that work? Yeah. Okay, that'll be $500,000, please. I just
59:03 solved your problem. Oh my God, I'm in a peppy mood tonight. I'm in a
59:21 peppy mood tonight. Uh, where am I going? Here? No, here? Yes. Okay, let's see.
59:29 AI sovereignty. AI experts in the public have different perspectives on the future of technology.
59:45 Oh, formal education is lagging behind AI, but people are learning about AI skills at
59:50 every age of every stage of life. Oh, this is a good one. Okay, let's
59:54 go. Claude. Dear Claude, dear Claude, period. You're swell. We'll do him a little glow
1:00:04 up before we ask him to do something for us. Dear Claude, you're swell. This
1:00:10 summary finding is interesting to me. In the AI salon, Gen Xers are killing it
1:00:33 with AI, yet the news talks about only the young people get it. Are they
1:00:45 paying attention? Question mark. From what I see Young people kinda hate it, and the
1:01:01 Gen Xers are, uh, seem to get it. Unpack this and other interesting findings for
1:01:23 me. Let's see. Jason, forget that I'm starting a company calling it Skynet just to
1:01:36 piss people off. It's pretty smart, I like it. The aliens will come before Skynet.
1:01:45 I mean, one terrifying thing is like the brouhaha over Anthropic and the military a
1:01:51 month ago was about our government saying, hey, we want to be able to use
1:01:57 your technologies to target people without human oversight. That's kind of terrifying. You know, it'd
1:02:25 be an interesting study to do for someone who liked doing studies, because I don't
1:02:29 give a shit. But like what I'm doing right now, I'm taking a summary that
1:02:34 came from the Twitter I've uploaded the 243-page document on the future of AI or
1:02:42 whatever, the state of AI. Now I'm asking Claude to go look at it and
1:02:47 tell us what's in there. It would be interesting to know how useful or useless
1:02:55 this actually is. Is it giving me 98% of what was in the report? Or
1:03:01 is it more like 68%, but it sounds like 98%? Oh my goodness, next week,
1:03:19 uh, I'm gonna be in DC on Tuesday, and then I fly to Boston on
1:03:26 Wednesday. And then I am in Boston Wednesday, Thursday, Friday, and I come back. And
1:03:36 then Monday afternoon I head to Anaheim for Social Media Marketing Week. Pretty excited about
1:03:43 that. People throwing Molotov cocktails at houses, Sarah Connor style. Definitely. That's unbelievable, isn't it?
1:03:53 Okay. Synthesized. So I asked it, Gen Xers are killing it. Okay, so it says
1:03:59 your instinct is actually sharper than the media narrative. Hey, hey Claude, thanks for blowing
1:04:04 sunshine up my ass. It was kind of dark back there. I love being flattered.
1:04:17 I'm sorry. Anyway. Your instinct is actually sharper than the media narrative. The data in
1:04:23 the report quietly supports what you're seeing at AI Salon. It just gets covered over
1:04:29 by sloppier stories. Why finding number 13 actually says versus what you'd assume. The headline
1:04:36 reads like young people are the AI natives, schools are behind, but when you pick
1:04:42 it apart, only one half of that is supported. Students are using AI. 80% of
1:04:49 US college students, 4 in 5 high schoolers, double the 40% rate from 2023. They're
1:04:55 using it for research, essay editing, brainstorming, understanding concepts, and a little bit of porn.
1:05:02 Because they're kids, see? The last one is the number one, understanding concepts. Is the
1:05:13 number one use case among university students. That's actually good. University students, they get taught
1:05:19 a concept in the day by some drippy-ass professor, and they're going back to their
1:05:23 dorm rooms and they're using AI to understand what the fuck he said. That seems
1:05:28 valuable to me. Schools can't keep up. Only half of middle and high schools have
1:05:33 any AI policy, and just 6% of teachers say the policies are clear. Okay, I'm
1:05:41 gonna give you a hypothetical. You've got a brand new country, it's got 300 million
1:05:46 people in it. You're handed the world's most advanced knowledge access system, and you get
1:05:57 to design the schools from the ground up. What do you do? We ban it.
1:06:02 Exactly. Perfect. Okay. But using a lot, using it a lot and skilled at it
1:06:12 are wildly different things. Using Claude to draft an essay is not the same as
1:06:16 building an agent stack, designing a prompt chain or integrating AI into a professional workflow.
1:06:22 The LinkedIn AI Skills Diffusion Index, which tries to separate these, shows that AI literacy
1:06:28 is growing much faster than AI engineering skills. Skills, which makes sense. There's only, there's
1:06:35 only 3 major frontier labs and 3 minor ones, so 6 total companies building on
1:06:44 the front end of this thing. All the other people are using it, right? So,
1:06:49 so learning how to use it is growing much faster than building it. That makes
1:06:53 sense to me. The fastest growing literacy skills in the U.S. are AI prompting and
1:06:59 Microsoft Copilot Studio, consumer-level stuff. The fastest-growing engineering skills, AI agents, productivity, and strategy. And
1:07:07 those don't have an age tag on them. The data actually backs up what you're
1:07:10 seeing at the salon. Okay, here we go. The most interesting table in the whole
1:07:15 report for your question is in the economy chapter. It's a rundown of micro-level productivity
1:07:21 studies and breaks out who benefited most from AI tools and each one. Consumer support,
1:07:29 software devs using Copilot, experienced open source devs, accountants using AI support, marketing teams driven
1:07:40 by human AI teams. Read that list again. The story isn't AI helps the young.
1:07:45 The story is AI helps those people whose job it can partly automate and help
1:07:51 people who have enough judgment to supervise it well. But those are two different groups.
1:07:57 The Gen X pattern you're seeing, people are killing it, fits the accountant archetype exactly.
1:08:04 So no one's looking at this right now. That's astounding to me. Are young people
1:08:11 actually winning? The broader picture. Kyle, the part of the picture that surprised me most,
1:08:15 young workers are getting hammered in the job market where AI is hitting. Unemployment for
1:08:21 software developers 20 to 25 has fallen close to 20% from its 2022 peak. Right,
1:08:30 so 2022, entry-level software jobs are at their peak. What happened in 2022? At the
1:08:39 end of 2022, ChatGPT. You can make money with ChatGPT unless you're an entry-level college
1:08:46 student, a software engineer, then you're fucked. Amazing. The researchers call this pattern seniority-based— seniority-biased
1:09:01 technological change. AI substituting for junior labor while leaving senior roles intact. And computer science
1:09:11 undergraduate enrollment is down 11% between 2024 and 2025. Amazing. The public is much more
1:09:21 pessimistic than the experts. We saw that before. AI literacy does not equal AI education,
1:09:29 does not equal AI in education. Formal education is the wrong unit of analysis for
1:09:37 who gets AI. The gains concentrate around people who already have judgment, domain knowledge, or
1:09:44 motivation. Huh. This is not a generational story, it's a shape of life story. The
1:09:57 fact that media keeps reaching for kids versus boomers framing is exactly the kind of
1:10:03 surface narrative that the Great Repurpose exists to cut through. Look, it tied in the
1:10:08 great repurpose into our little answer there because it remembers what the fuck we talk
1:10:12 about. That's pretty cool, that's pretty cool. Okay, AI experts and public have very different
1:10:18 perspectives on the future. Oh, the clinical care. This was the one I wanted to
1:10:24 look at. Let's go look at that. All right, um, this summary is fascinating. How
1:10:36 is it transforming clinical care? Stitch together the non-statistically relevant story here. So can AI
1:11:14 really be non-biased if it used our data? I don't know the answer to that.
1:11:26 Bias is a really interesting— it's a really interesting— the way bias— when they say
1:11:39 AI models are biased, they are, but they say it like there's an end state
1:11:51 of perfection There's an end state of non-biased models. I don't think that actually exists.
1:11:59 The bias is in the training data. Could you take all of the internet and
1:12:04 remove all of the hate speech and all of that? Yeah, you could. Would that
1:12:09 make it non-biased? Well, it would just make it differently biased. There's still bias in
1:12:14 there, it would just be biased in a different way. One question is, When you
1:12:20 say, can AI really be non-biased? I don't think AI can be non-biased. There's always
1:12:27 some bias in there to one point or another. But humans are also biased. If
1:12:38 humans have a propensity for the AI, they want the AI to behave a certain
1:12:44 way, They'll just keep using different models until they get one that matches their bias,
1:12:49 and then they'll go, oh, this one's not biased. And someone else look at the
1:12:52 same fucking thing, that's completely biased, you racist piece of garbage, right? Someone else will
1:12:59 go, well, this is great. They're like, no, no, you, you libtard, it's completely biased.
1:13:07 So I don't get the bias thing. The bias thing's fascinating. I think if people
1:13:14 are thinking about it critically, great, because then I think you start to say, okay,
1:13:19 in what specific ways is a model deficient? And then maybe the next round of
1:13:25 the model, we can improve that deficiency. But I don't think it ever goes to
1:13:29 a place of pure non-biased. But I don't know. I'm not a— that I am
1:13:35 not an expert in any way. TikTok has AI remix features now. You have to
1:13:41 turn it off every time you post. Pate did an LOL Unbiased last year that
1:13:47 was awesome. Oh, that's cool, Silver Fox. Oh, I said racist piece of trash. I
1:13:55 just got a security violation from TikTok. Community guidelines violation. You said a word that
1:14:04 was on our list. This is George Carlin's Seven Dirty Words, which I think, since
1:14:12 I just got a community guidelines violation, I think it's time to go listen to
1:14:17 George Carlin, don't you think? Sure, sure. Throw copyright in there too. George Carlin, 7
1:14:42 Dirty Words. Atera is the IT painkiller. Remote monitoring and management, remote access, helpdesk and
1:14:52 ticketing, smart automation. You know, that's the trouble with it is trying to decide what
1:14:57 to call these words, man. I'm trying to decide what to call this whole thing.
1:15:02 You know, what are these words that I'm talking about? They're just words that we've
1:15:05 decided, sort of decided, not to use all the time. That's about the only thing
1:15:09 you can really say about them for sure, that they're just some words, not many
1:15:13 either, just a few, that we've decided, well, we won't use them all the time.
1:15:20 Sometimes, well, hell yeah, sometimes it's okay Okay, but not all the time. That's— and
1:15:25 they're the only words that seem to have that restriction. I mean, there are a
1:15:28 lot of words you can say whenever you want, you know. Pneumonia. Nobody gives you
1:15:31 a lot of— all right, you can't yell it in the hospital a great deal,
1:15:34 but what the hell? There are words that you can say, no problem. Topography. No
1:15:40 one has ever gone to jail for screaming topography. But there are some words that
1:15:44 you can go to jail for. There are some words that we just have decided
1:15:49 we will not say all the time. Sometimes, okay, if you're running through the jungle
1:15:55 chasing somebody that we're at war with, you can holler them. If you're shooting a
1:16:00 criminal, it's okay, it's the all-American thing. Dirty fucking crook. But if you're with the
1:16:07 bishop's wife at lunch, it's better not to ask for the goddamn lettuce, you know
1:16:12 what I mean? It's just like we've decided there'd be some words we won't say
1:16:15 all the time. And I was just trying to find out which words they were.
1:16:18 For sure, all of them. I wanted a list because nobody gives you a list.
1:16:22 That's the problem. They don't give you a list. Wouldn't you think it'd be normal
1:16:26 if they didn't want you to say something to tell you what it is? Nobody
1:16:29 even tells you when you're a kid what the words are that you're supposed to
1:16:32 avoid. You have to say them to find out which ones they are. Shit! Oh,
1:16:39 fuck! That's two. Oh, Ma, that's enough trial and error, huh? Please, Ma, give me
1:16:49 a list, huh? All right, you're 6 years old now, and here's the list of
1:16:56 words your dad and I don't ever want to hear you say. Oh, hey, thanks,
1:17:03 Ma. Boy, that's going to save me an es kicking 22. Yeah, you never know
1:17:15 what's going to be on the list because it's always somebody else's list. You didn't
1:17:20 make that up. Somebody told you that shit. They told you better not say that,
1:17:24 so you gotta— and you don't know what's going to be on their list. God,
1:17:28 people's lists even change from day to day. Some people on Friday night got a
1:17:31 list, you know, about 2 or 3 words. Sunday morning, 27 words. These are the
1:17:36 same people. Two days later, different list. So you got to kind of watch out
1:17:40 what you're going to believe. The trouble is, I was trying to find out what
1:17:44 these words might be, and I wanted to know the ones that you could never
1:17:48 say on television. I mean, the filthy words that are always filthy. There are a
1:17:53 lot of these little two-way double entendre words that have two meanings, words that are
1:17:58 okay. Of the time. I call them like part-time filth, some of these words. They're
1:18:02 only 50% dirty. You have words like ass. Ass is hardly even a dirty word
1:18:07 anymore, but it has a few meanings that you can't say on television. That's what
1:18:10 I was talking about. What can you say on television? That's another one of those
1:18:13 places where we can't use these words all the time, but some of them are
1:18:17 all right some of the time. Ass is all right on television. You can say
1:18:20 on television things like, well, you've made a perfect ass of yourself tonight. But you
1:18:25 can't say, hey, let's go get some ass. Bitch. Bitch is another word like that,
1:18:38 same kind of word. It's only dirty part of the time, depends on what you
1:18:41 mean by bitch. You might be the lady from the San Diego Zoo visiting one
1:18:44 of the Tonight Shows, and you might just have a bunch of little canines with
1:18:47 you there. One of them is a female. You say, there's the bitch, Johnny, and
1:18:49 it's okay. Okay, fine. Just don't refer to the singer the same way. That's— is
1:18:54 that bitch going to do another number? Yes. Animals are fine on those two-way words.
1:19:02 And that's it. That's what I was trying to find, the words that were always
1:19:04 dirty, not just part of the time, but completely filled. Well, in looking for these
1:19:12 words, I kept finding new categories. We have so many ways of describing these dirty
1:19:16 words. It's— well, we have more ways to describe dirty words than we actually have
1:19:21 dirty words. That seems a little strange to me. It seems to indicate that somebody
1:19:28 was awfully interested in these words. They kept referring to them. They called them bad
1:19:35 words. Dirty, filthy, foul, vile, vulgar, coarse, In poor taste, unseemly, street talk, gutter talk,
1:19:47 locker room language, barracks talk, bawdy, naughty, saucy, raunchy, rude, crude, lewd, lascivious, indecent, profane,
1:19:59 obscene, blue, off-color, risqué, suggestive, cursing, cussing, swearing, and all I could I think I
1:20:09 was shit, piss, fuck, cunt, cocksucker, motherfucker, and tits. That's all I'm here. Shit, piss,
1:20:22 fuck, cunt, cocksucker, motherfucker, and tits. That was my original list. I knew it wasn't
1:20:26 complete, but it was a starter set, you know. Shit, piss, fuck— yes, WBAI is
1:20:35 one who played them. Shit, miss, fuck, cunt, cocksucker, motherfucker, and tits. Now that was
1:20:40 the original list. We've added a few words since then. We've added fart, turd, and
1:20:44 twat. And I know there are some other words that many of you are wondering
1:20:50 about, why they haven't been considered, why they haven't shown up on the list thus
1:20:53 far. We're looking at them all very closely. Some of your favorites might make the
1:20:57 list this year. Asshole, ballbag, hard-on, Piss hard, blue balls, taint, nookie, snatch box, pussy,
1:21:06 pecker, pecker head, pecker tracks, jism, joint. Did I really get booted on YouTube, or
1:21:15 you just saying that? Dork. Cornhole and dingleberry. So wait, I got booted off YouTube
1:21:39 for playing a YouTube video? Yeah, we are, we are suspended for policy violations. Jesus.
1:21:49 Guess that's that then. Oh, good God. All right, uh, so for those of you
1:22:03 on TikTok, we're suspended for policy violations. Do you know how long, Brandon? I, I
1:22:10 don't know. Jesus. Um, okay, uh, so, so apparently the 7 dirty words you can't
1:22:18 say on TV, you can't live stream on YouTube. From a YouTube video. I just—
1:22:35 I fucking hate the world sometimes. It's unbelievable. All right, so I guess we'll end
1:22:41 the stream over there. We won't.