2/5/2026 - AI Agents, Job Redefinition, and the Forced Identity Crisis of the Programmer

Description

Chapters

Transcript

0:00 Campy, you ready to sing sing for your
0:03 dinner?
0:35 You're all feeling nervous, aren't you,
0:38 boy?
0:42 And impeccable style.
0:47 Don't ever let them steal your joy
0:51 and your gentle ways to keep them from
0:54 running wild.
0:57 They kick dirt in your face, dress you
1:00 down, and tell you that your place is in
1:02 the middle when they hate the way you
1:05 shine.
1:08 I see you tugging on your shirt,
1:12 trying to hide inside of it. hide how
1:15 much it hurts.
1:18 Let them laugh while they can.
1:23 Let them spin.
1:26 Let them scatter in the wind.
1:33 I've been to the movies. I've seen out
1:36 hands
1:38 and the jokes
1:42 on them.
1:45 Voice is still not good.
1:49 You're feeling nervous, aren't you,
1:51 girl?
1:53 It's your brother's world for a while
1:56 longer.
2:02 Roy Harrison in the house.
3:39 Wow.
3:50 Woohoo!
3:58 Woohoo!
4:22 What's that, Vicki? You thought today
4:24 was Friday.
4:29 You like the shades? I think the shades
4:31 are cool. These are my new driving
4:33 glasses, but they're also my new jamming
4:36 glasses. You got to jam out. Got to jam
4:40 out on a guitar.
4:42 Get a little Jack Black.
5:01 Woo!
5:28 Wow!
5:37 You look cool. But how does your eye
5:40 feel? I I uh it is not just a like block
5:45 tear duct or whatever I thought it was.
5:46 I actually have pink eye, which Brandon
5:49 thinks I got in DC. I got too close to I
5:53 got too close to the muck and it rubbed
5:56 rubbed off on me.
5:59 Uh, no, I actually think I got it. I
6:01 went for an eye appointment at the eye
6:03 doctor and two days later.
6:06 So, but I'm on the correct drops now, so
6:08 I should be good.
6:13 Thank you for asking. But yeah, look how
6:15 cool I look. I think I look pretty
6:17 swell.
6:22 You and I here all alone.
6:26 Sunday morning here at home.
6:30 Sky's blue and a coffee strong. It's
6:34 true.
6:38 Then I open my eyes to a dream realized
6:42 in front of me
6:46 and I haven't got a clue when the world
6:48 is happening to me.
6:53 Think I think I'm happy
6:56 like the first day of summer vacation.
6:59 Happy got to get a little rest and
7:01 relaxation. Happy
7:04 like the choir on Sunday morning singing
7:07 true.
7:20 Woohoo!
7:40 Um,
7:43 what's going on?
7:45 There's a lot of stuff happening. I
7:47 don't know if you know this, in the
7:49 world of artificial intelligence,
7:51 there's a lot going on.
7:53 There's things being announced every
7:55 day.
7:57 Agents are about to take over the world.
7:59 Everybody pay attention to the agents
8:02 cuz where you think you have control,
8:04 you're about to lose it.
8:09 It's It's a [ __ ] [ __ ] show.
8:13 It is a train wreck. a travesty or the
8:19 most exciting thing to happen since
8:22 we didn't have agents.
8:28 Oh, good lord.
8:30 There's a new Claude 4.6 Opus out. Seems
8:34 to be impressing people.
8:38 There's a codeex 5.3 from Chat GPT.
8:42 There's a Cling 3 on the video front.
8:45 There's just everything here for the
8:47 music. I'm so sorry.
8:49 It's only me doing the singing. That's
8:51 the That's the problem with coming here
8:53 for the music.
9:06 Sitting in this lonely town.
9:11 Clean your guitar strings so you don't
9:12 crosscontaminate. Oh, that's a good
9:14 idea.
9:17 When you get actual doctors in the feed,
9:22 you can get some advice.
9:26 Sitting in this lonely town,
9:29 wonder things are going to change.
9:34 Dream my life away.
9:37 Dream. Seems these dreams have turned to
9:39 a bunch of dust clouds.
9:43 getting my nerve up. But my Corey likes
9:46 my round glasses. I knew Corey would
9:48 like my round glasses cuz they're hip.
9:52 She'd like them better if one of them
9:53 was square.
10:06 Dr. J. I'm a shrink. Lol. Just ate
10:08 germs. Shrinks are doctors. You got to
10:11 go to some sort of school to be a
10:12 shrink.
10:20 Yeah, I'm happy with the new glasses. I
10:22 I So, you kind of have a choice, right?
10:25 As you get older,
10:28 you can like you can go I'm going to age
10:31 up or I'm going to go more hip.
10:34 And and the the more hip I'll I'll show
10:36 you an unfortunate attempt at going more
10:39 hip.
10:40 That was not the right choice.
10:43 The These were my driving glasses for
10:45 about a week and then I realized what a
10:49 douchebag they made me look like. Look
10:50 at these things.
10:58 Oh, they're awful. They to go with my
11:01 hip haircut. Exactly. Yeah, these are
11:03 awful. These I I I bought online and
11:08 then the minute they showed up I'm like,
11:10 "Oh, those are kind of awful."
11:16 So those those those just sit in the the
11:19 back corner. So we'll go with the we'll
11:22 go with the round ones.
11:26 A Kelly Bes character
11:35 Can we go back to the rounds? Yes, we
11:37 can.
11:39 They go in the shelf next to the rabbit
11:41 in terms of uselessness. I think that's
11:43 a good idea. Producer Brandon, we'll
11:44 just put them right up here. Right next
11:46 to the rabbit R1.
11:49 They go right there. That was a sad
11:52 attempt. A sad attempt at hypn.
11:57 sad. But
12:00 are you having a midlife crisis?
12:03 If only are midlife.
12:34 Sitting this lonely town. Yeah.
12:38 Things are going to change.
12:42 Dream my life away.
12:45 Seems these dreams have time to bump
12:48 those clouds.
12:51 Get my nerve up, but my past is pulling
12:55 me down. Sing along, Chappie.
12:59 Wondering how long
13:02 this black sheep going to stick around.
13:10 Somebody told me once before, he said,
13:13 "You can never go home again.
13:16 Won't you leave?"
13:19 and a thing to steer me away
13:23 from the truth of who I am and what I
13:25 believe. So I thanked him for his two
13:29 sins with a handshake and some sympathy.
13:33 Yeah.
13:35 Packed on my blue jeans and headed for
13:39 this big price
13:42 of my freedom.
13:45 Bye-bye
13:47 black sheep to the black sheep of the
13:51 family.
13:53 Bye-bye.
13:57 Oh, I mean so very much to me and
14:01 bye-bye
14:03 black sheep to my friends and my family.
14:10 Bye-bye.
14:13 I going to set my soul.
14:17 Set it free.
14:26 I hate to interrupt the concert.
14:28 >> Yes,
14:28 >> I have some breaking news. Speaking of
14:30 the Rabbit R1,
14:33 >> uh they have are coming out with a
14:35 dedicated vibe coding machine soon
14:39 called Project Cyber Deck. Project
14:41 Cybert Deck
14:44 Project. Okay, we are now officially
14:47 Okay, I'm going to declare it. Mark this
14:48 date, February 5th, 2026.
14:52 We are entering AI being overhyped.
14:58 Project Cyber Deck. Come on. Could they
15:02 come up with a more horrible name?
15:03 That's really bad. All right, but what
15:05 is it? It's a It's a hardware device you
15:08 vibe code on.
15:09 >> Yes.
15:11 Nice.
15:12 >> Uh, so be purposely designed for vibe
15:15 coders to run things like cloud code CLI
15:17 and the upcoming rabbit command line
15:19 interface.
15:20 >> Ah,
15:22 all right. We're we're officially in the
15:24 in the hype cycle. Agents are taking
15:27 over. Everyone's freaking out.
15:33 I think Anthropic announced something
15:36 about
15:37 agents today.
15:40 Open AAI announced some new enterprise
15:43 thing. We should go look at that. I
15:44 haven't I didn't look deeply enough into
15:46 it. So, so we'll go look at that and see
15:49 if that's anything interesting.
15:55 Massive layoffs in January. Yeah. But,
15:58 you know, Todd, I here's the here's the
16:01 thing about the layoffs. The it
16:04 corporate America can't have it both
16:07 ways. Okay. corporate America can't have
16:11 there's no return on AI. AI is a waste.
16:14 We we don't know how to use it. No one's
16:16 adopting it. And we're laying off 30,000
16:19 people because of AI. You can't have it
16:21 both ways. America, let's just let's
16:24 just
16:27 admit
16:28 that we're going to cut some costs so we
16:31 can drive a little more profit to the
16:33 shareholders. Okay. Okay. and let's hope
16:36 that the remaining people can figure out
16:38 how the [ __ ] to use AI to make up for
16:40 the shortfall of the ones you just let
16:42 go. That's what's going on in 2026 in my
16:46 humble opinion because there's not
16:48 enough adoption.
16:51 No one knows how to [ __ ] use it yet.
16:53 We don't know how to use it as
16:54 individuals.
16:57 Companies don't know how to use this
16:58 [ __ ] yet. They're not training their
17:00 staff on how to how to do these things.
17:03 I mean, it's a handful of companies
17:05 represented, you know, in this audience,
17:08 in this community that that are starting
17:11 to get their [ __ ] together. They're
17:12 starting to figure it out.
17:20 It's like buying calculators and get
17:22 getting rid of your accounting
17:23 department. That's exactly what it's
17:25 like.
17:27 Yeah. And we know stuff. Exactly, Corey.
17:30 Like we know it's not ready for prime
17:32 time. We know most people don't know how
17:34 to use it at all. We know that it's
17:37 chuck full of security risks and
17:39 [ __ ] Like it it just Anyway, I I
17:43 just
17:45 it's exhausting right now. It really is
17:48 [ __ ] exhausting right now.
17:53 Is it?
17:54 >> Yeah.
17:55 >> Can Can I talk about my prompt of the
17:57 week? I'm really excited about it.
17:59 >> Yes. Jump in. Jump in. I I I got
18:02 nothing. I got I got nothing going on
18:04 here. So, I think prompt of the week
18:06 week is awesome.
18:08 >> Yeah. So, if you're not sure where we
18:09 are, we are at community.thesalon.ai,
18:12 which is an extension of the AI learning
18:14 lab. And what we've started out this
18:17 month is a new prompt of the week. And
18:20 so, each week we're going to come back
18:21 and share on Thursdays a different
18:23 prompt for you to riff off of. And so
18:28 since the theme of the month inside the
18:30 AI salon is signal, the prompt of the
18:35 week this week is frequency modulation
18:37 against an analog backdrop.
18:39 >> Oh, nice.
18:40 >> So your challenge is to create anything
18:43 that explores that idea. an image, a
18:45 song, a short story, a workflow, a
18:48 conversation. Interpret it literally or
18:50 metaphorically, digital versus human,
18:53 precision versus texture, change
18:55 happening inside something familiar. And
18:57 what we want you to do is go over to
18:59 look what I made and share your results
19:01 and maybe even add one sentence about
19:03 how you interpreted the prompt. So this
19:07 was my example image
19:09 of a frequency modulation against an
19:13 analog backdrop. This was done in CH GPT
19:17 >> enhanced imaging.
19:18 >> That's very cool.
19:21 >> That is very cool.
19:22 >> I would encourage you to head over to uh
19:25 the AIM and share what you think of
19:27 that.
19:27 >> I love it. And and the the to to uh to
19:30 pump it up a bit more the the theme of
19:32 the month. I'm really excited about it.
19:34 This is Brandon's Brandon's baby. And
19:37 there's a there's a group out there uh
19:39 called uh creative mornings and they
19:42 they've got they do these meetings once
19:44 a month where they meet in all different
19:46 cities and just have creative
19:48 conversations and they have a theme of
19:49 the month and so all of the
19:51 conversations are geared toward toward
19:53 that theme. And so we're starting that
19:56 inside the uh the AI salon community. So
19:59 signal is the uh is the theme for
20:01 February. So go check out that prompt.
20:04 That's that's a so so within
20:08 within the AI salon, if you go just to
20:10 the welcome page, you'll see there the
20:12 cycle of AI readiness. Play first,
20:15 create excellence, generously lead. That
20:18 prompt of the week is a really good
20:22 opportunity for you to go play, take
20:24 that theme and go play in some piece of
20:27 AI land, some tool that you haven't used
20:31 before and come up with a way to
20:34 interpret that. If you haven't ever made
20:36 an app, go to lovable.dev and make an
20:38 app. If you haven't ever made a movie,
20:40 go make a movie, you know. Yeah. Really
20:44 cool. Really cool producer Brandon. I
20:46 like it. I like it a lot.
20:48 >> And when you're done doing that and you
20:51 need some comic relief, I would
20:54 encourage you to head over to the
20:55 Irregular's channel to see what digital
20:58 gods whipped up.
20:59 >> Why do I have a
21:01 files?
21:04 >> So now you get to hear a Gen Xer get
21:06 pissed off because some dweebs in some
21:09 [ __ ] lab somewhere decided that
21:12 images needed to be 3% more efficient.
21:15 So they created an entirely new [ __ ]
21:18 file format that no one [ __ ] uses.
21:23 So what we're gonna do because we've
21:25 experienced this before is we're going
21:27 to whine. Okay? We're going to whine and
21:30 we're going to [ __ ] and then we're
21:32 going to go fix it because we know how
21:34 to fix it. But someone,
21:37 some little punk ass 25year-old, you
21:40 know, if if we if we uh encoded the the
21:43 the pixels, the the data image and the
21:46 pixels, the color space, we could and
21:48 the nobody cares.
21:52 The internet's fast enough.
21:56 I just want [ __ ] to work.
22:02 Thank you for sharing that, producer
22:04 Brandon.
22:06 That was It was not one of my proudest
22:08 moments,
22:10 but now it's been memorialized
22:13 with a talking lemon. Listen, we we went
22:16 in heavy on talking hot dogs early in
22:19 this channel, so I think having me as a
22:20 talking lemon is is is awesome.
22:29 It was one of your better rants. I
22:31 wasn't actually really good when I when
22:33 I went to do the uh the the the dweeb
22:36 talking about the algorithm. My brain
22:38 went blank.
22:40 But, you know, but I was pissed off. So,
22:43 there's that. It's just like I mean,
22:45 it's been three years and
22:48 nobody still use like like chat GPT
22:52 outputs image file formats that it can't
22:55 import.
22:58 It's like, stop it. Just use the [ __ ]
23:02 that works.
23:04 God damn,
23:08 you get me all fired up again. Tik Tok,
23:10 dude. Love Lord Digital. God's acidic
23:13 Kyle.
23:21 night after night.
23:26 Thank you, Lord, digital gods, for
23:28 memorializing my my moment of shame.
23:33 Oh, man.
23:35 All right, I want to go look at
23:36 something. I want to go We're going to
23:38 read something together. We're going to
23:39 have a little reading moment.
23:43 this one
23:45 cuz I saw this today and I didn't read
23:47 it
23:49 because I don't know if you know this
23:50 but my eyes are melting out of my
23:53 [ __ ] head.
23:56 Um
23:59 Oh, listen to the article. We could
24:01 listen to it. Yeah, let's listen to it.
24:04 [ __ ] it. Let's listen to it.
24:08 We live in a world now where you can
24:12 You can just listen to [ __ ]
24:15 Everything's an audio book now.
24:27 Okay, there we go.
24:32 And this is why we come back night after
24:34 night.
24:38 Uh, okay.
24:43 >> AI has let teams take on things they
24:45 used to talk about but never execute. In
24:48 fact, 75% of enterprise workers say AI
24:51 helped them do tasks they couldn't do
24:53 before. We're hearing this from every
24:56 department, not just technical teams.
24:58 The way work gets done has changed, and
25:00 enterprises are starting to feel it in
25:02 big ways. We've seen this in action with
25:04 over 1 million businesses over the past
25:07 few years. At a major manufacturer,
25:10 agents reduced production optimization
25:12 work from 6 weeks to one day. A global
25:15 investment company deployed agents endto
25:17 end across the sales process to open up
25:19 over 90% more time for salespeople to
25:22 spend with customers. And at a large
25:25 energy producer, agents helped increase
25:27 output by up to 5% which adds over a
25:30 billion in additional revenue. This is
25:33 happening for AI leaders across every
25:35 industry and the pressure to catch up is
25:37 increasing. What's slowing them down
25:39 isn't model intelligence. It's how
25:42 agents are built and run in their
25:45 organizations.
25:52 What's slowing them down isn't model
25:54 intelligence. It's how agents are built
25:57 and run in their organizations.
26:00 I don't think that's actually [ __ ]
26:02 true.
26:06 What is the definition of agent?
26:11 Are they talking about automations
26:16 or autonomous agents?
26:20 I think this is just horseshit marketing
26:22 right here. But anyway, let's continue
26:24 on, shall we?
26:26 Wait, why can't where?
26:30 >> Today, we're introducing Frontier, a new
26:32 platform that helps enterprises build,
26:34 deploy, and manage AI agents that can do
26:37 real work. Frontier gives agents the
26:40 same skills people need to succeed at
26:42 work. Shared context, onboarding,
26:45 hands-on learning with feedback, and
26:46 clear permissions and boundaries. That's
26:49 how teams move beyond isolated use cases
26:51 to AI co-workers that work across the
26:53 business. Okay, I think this sounds
26:56 cool. I think this sounds right. I think
26:58 this sounds good. But this setup
27:00 paragraph is absolute horseshit.
27:04 Organizations are not struggling with AI
27:06 because they can't figure out how to
27:08 build agents better. They're struggling
27:10 with AI because they don't know how to
27:11 deploy it. No one in the organization
27:13 knows how to use it because they're not
27:15 train anyone training anyone on what the
27:17 [ __ ] it is.
27:24 I'm telling you, man. Open AAI needs an
27:25 angry demer.
27:28 Oracle, State Farm, Thermoffisher, and
27:30 Uber are among the first to adopt
27:32 Frontier. And dozens of existing
27:33 customers, including BBVA, Cisco, and
27:36 T-Mobile, have already piloted
27:38 Frontier's approach to power some of
27:39 their most complex and valuable AI work.
27:42 The AI opportunity gap. Companies are
27:44 already overwhelmed with the
27:46 disconnected systems and governance
27:47 spread across clouds, data platforms,
27:50 and applications. AI made that
27:52 fragmentation more visible and in many
27:54 cases more acute. Agents are now getting
27:57 deployed everywhere and each one is
28:00 isolated in what it can see and do.
28:02 Every new agent can end up adding
28:04 complexity instead of helping because it
28:06 doesn't have enough context to do the
28:08 job well.
28:10 As agents have gotten more capable, the
28:12 opportunity gap between what models can
28:14 do and what teams can actually deploy
28:16 has grown. The gap isn't just driven by
28:19 technology. Teams are still building the
28:21 knowledge to move agents past early
28:23 pilots and into real work as fast as AI
28:26 is improving.
28:27 >> That's an understatement.
28:28 >> At OpenAI alone, something new ships
28:30 roughly every 3 days. And that pace is
28:33 getting faster.1.
28:34 Keeping up means balancing control and
28:37 experimentation, and that's hard to get
28:39 right. Enterprises are feeling the
28:41 pressure to figure this out now because
28:43 the gap between early leaders and
28:45 everyone else is growing fast. Open AI
28:48 Frontier. We've learned that teams don't
28:50 just need better tools that solve pieces
28:52 of the puzzle. They needed help getting
28:55 agents into production with an
28:56 end-to-end approach to build, deploy,
28:58 and manage agents.
29:07 >> Really?
29:09 Really? Is that is that what you're all
29:11 hearing?
29:15 that the big the big holdup is we can't
29:17 figure out how to deploy our agents.
29:21 What are they actually trying to say?
29:22 Here's what they're trying to say.
29:24 Source camp, we have a new thing that
29:26 makes it easier to deploy the things
29:27 that were just hidden in our dev tools
29:30 for two years that no one can figure out
29:32 how the [ __ ] to use. So, we've we've
29:35 made it easier for corporations to use
29:37 them. And we're going to talk about it
29:38 as if everyone's trying to use these
29:41 things. That's what they're trying to
29:43 say. They've just collected sound bites.
29:46 Yeah, this is pretty bad. I This the
29:49 tool actually sounds kind of cool. They
29:52 asked chat GPT. Yeah, exactly. This
29:55 totally sounds like a chat GPT written
29:57 thing. You know, one of the great
29:58 challenges of AI adoption is that they
30:00 can't just figure out how to get these
30:02 darn agents to behave, if you know what
30:04 I mean.
30:08 Hey, I really like the way you're
30:09 running your organization there. Maybe
30:11 if you could get some agents going,
30:12 things would be better, huh?
30:22 Okay, let's let's let's continue on,
30:24 shall we? We only have six more minutes
30:26 of this to go. My son is in community
30:29 college and he's having to find his own
30:33 way trying to learn technology outside
30:35 of college. I don't get me started on
30:37 [ __ ] higher education m okay
30:42 I it's going to be another [ __ ] rant
30:44 night. I can't. It it it makes me insane
30:48 that our educational institutions, we've
30:52 got the the the the most profound
30:57 knowledge technology in the history of
30:59 humanity
31:01 at our fingertips
31:03 and your sons having to [ __ ]
31:07 moonlight
31:09 his [ __ ] AI career
31:12 because they're like, "Oh, well, you
31:14 know, artificial intelligence just
31:16 allows students to to to write their
31:18 papers without even trying. So, we've
31:20 got detectors for that. There were we
31:22 are we still there?
31:25 It's astounding to me. It's astounding
31:27 to me. You know, people ask me like, you
31:30 know, should my kids go to college or
31:32 whatever? I mean, I guess maybe for the
31:35 social thing certainly. Um, I think they
31:39 should be studying philosophy more than
31:41 computer science, quite frankly. Um,
31:45 and if if I were interviewing schools,
31:47 like I would be interviewing the heads
31:49 of the departments to say, you know,
31:52 what's your stance on AI? And if they
31:53 were not aggressively into it, I
31:55 wouldn't [ __ ] go there. It's
31:57 astounding to me. Anyway,
32:02 all of the schools are behind. They're
32:04 willfully behind.
32:06 They're [ __ ] willfully behind.
32:09 It's irresponsible.
32:16 I wanna I wanna I want to play a video
32:19 that I saw today.
32:22 So So you know, we just launched this
32:24 thing called the great repurpose where
32:25 people are going to be losing
32:29 their sense of meaning if they've got
32:32 their identity too tight tightly tied to
32:35 their work. And as their work as the
32:38 tasks of their work gets automated and
32:40 stripped away, what are what are they
32:42 going to be left with if if their
32:44 meaning was tied to that?
32:46 Um Mark Andre, there's a there's a a
32:50 clip of him in an interview
32:54 that is talking about this and
32:58 I want you to hear it if I can find it.
33:02 Is this it? I think that's the interview
33:04 tab. Oh yeah, thanks.
33:12 >> Yeah,
33:13 >> on that AI coding actually abstracts the
33:15 way that
33:16 >> Okay,
33:18 listen to this. So this is Mark Andre.
33:21 He's the one that
33:23 he he invented Netscape, you know, he
33:26 created the first modern browser. He's
33:27 the head of A16Z
33:31 VC firm.
33:32 >> There's and then to your point like AI
33:34 coding is the next layer on that. AI
33:35 coding actually abstracts away the
33:37 process of actually writing the
33:38 scripting code, right? And so in one
33:40 sense, this is a really big deal for all
33:42 the obvious reasons, but on the other
33:43 hand, it's like, okay, this is the next
33:45 layer of the task redefinition under the
33:47 job of programmer, right?
33:49 >> Wait, listen to listen to those words.
33:53 He's smart. You we we've got to decipher
33:56 the words cuz cuz he's got the big brain
34:00 in the big giant head. This is the next
34:02 layer of the task redefinition under the
34:05 job of programmer.
34:08 >> I got I gota hear it one more time.
34:11 Thank god we have it. The lay. Let's
34:13 see. This is the next layer
34:16 of the redefinition
34:19 under the job of programmer. Wait the
34:22 wait. This is the next layer of the task
34:26 re the task redefinition.
34:28 This is the next layer of the task
34:31 redefinition of the job of programmer.
34:35 >> Now what's the job of the programmer?
34:37 It's to your point. It's not necessarily
34:39 to write the code by hand. But what it
34:41 is now is all right. Now you know if you
34:42 talk to the world's best programmers
34:43 today, what they'll tell you is oh my
34:45 job is I'm sitting there and I'm
34:46 orchestrating 10 code bots. What my job
34:49 is now is I'm a manager.
34:54 I'm a manager. I'm no longer a coder. If
34:57 you're a coder that loves coding,
35:01 you didn't ask for this.
35:06 You got a promotion. One of the things I
35:08 learned when I was doing agency.com is
35:10 there were sometimes people were really
35:12 awesome at their job and we would
35:14 promote them to manager and they would
35:16 either quit or they would ask for a
35:19 demotion.
35:21 I remember Paul Galley. I I remember it
35:23 distinctly. He he was employee number
35:26 two at agency.com
35:29 and he was a really good site builder
35:33 and it and everyone was getting promoted
35:35 around him and he didn't he didn't want
35:37 a promotion but at some point it just
35:39 became weird that like all of his
35:41 colleagues were getting promotions and
35:43 he wasn't. So we gave him a promotion
35:46 and two months later he walked into my
35:48 office and he said can I not do this? I
35:52 don't like managing people. This is not
35:56 what I enjoy doing. So like programmers
35:59 right now are in the middle of a forced
36:02 redefinition. I'm sorry. Wait, what did
36:04 he say? A forced
36:08 a forced
36:10 uh this is the next layer of the task
36:12 redefinition under the job programmer.
36:18 Right.
36:20 I want I want to read you something. I I
36:24 read I uh I did some uh I did some
36:27 groing today. How where do I where's my
36:30 history? Oh, history.
36:37 So, there was a Reddit Reddit post. We
36:39 we read this in here um back when it
36:43 happened. This was in March of 2023. If
36:45 you if you were here back then, you'll
36:47 remember this.
36:51 This was the first
36:56 this this th this post was the canary in
36:59 the coal mine to what's coming.
37:04 All right,
37:07 here's the full text. Okay,
37:13 so this is a this is a guy working at a
37:16 a 10-person game company in March of
37:18 2023.
37:20 I lost everything that made me love my
37:22 job through midjourney overnight.
37:26 I'm employed as a 3D artist in a small
37:28 games company of 10 people. The art team
37:31 is two people, me and my friend. He's
37:35 been working there longer than me and I
37:37 am a junior artist. Last week, my boss
37:39 was away on a trip and me and my friend
37:42 were left to work. We were given the
37:44 task of creating three graphics for an
37:46 advertising campaign. We finished them
37:48 all in two days and were left with
37:50 nothing to do. My friend showed me
37:52 Midjourney and we were messing around
37:54 with it for the rest of the week. When
37:56 my boss came back, my friend told him
37:58 about Midjourney and how good it was. My
38:00 boss got excited and asked him to show
38:03 it to him. My friend showed him how to
38:05 how to prompt and my boss was amazed. He
38:08 asked us to create graphics for a new
38:10 campaign. My friend prompted it and it
38:12 came out great. My boss was happy and
38:15 said, "This is the future." He then
38:17 asked me to create a graphic for another
38:20 campaign. I prompted it and it came out
38:22 okay. My boss was happy and said that
38:24 I'm a natural. He then asked us to
38:26 create graphics for all the campaigns
38:28 they have. We did and they were all
38:31 great. My boss was ecstatic and said
38:33 that he was going to save us so much
38:35 time and money. He then said we can now
38:38 focus on other things. I was happy at
38:40 first, but then I realized I'm not an
38:43 artist anymore. I'm a guy who types
38:45 prompts and waits for the computer to
38:47 generate the art. I lost all the joy I
38:50 had in creating art. I lost the feeling
38:52 of accomplishment when I finish a piece.
38:54 I lost the feeling of pride when I show
38:56 my work to others. I lost everything
38:59 that made me love my job.
39:02 My friend is happy because he hates
39:04 creating art. He just wants to get the
39:06 job done and go home. And he's now the
39:08 lead artist because he's better at
39:10 prompting than me. I am now the junior
39:12 prompter. I am thinking of quitting my
39:15 job. I don't know what to do.
39:21 That's about to happen at scale. I think
39:23 in in engineering it's happening right
39:25 now.
39:27 Um,
39:30 this unvoluntary
39:35 redefinition of the tasks that make up
39:37 our job is about to happen. And and the
39:42 the
39:45 agents, the agent swarms, right? The the
39:48 the open claw kind of things, they're
39:51 going to start rolling out from the big
39:53 frontier companies over the next two
39:55 months.
39:59 And the implication of that is so what
40:01 Mark Andre was talking about is as a
40:03 programmer
40:05 you're no you're no longer new no longer
40:08 doing coding what you're doing is you're
40:10 managing 10 agents that are doing the
40:12 coding right. So so you've gotten a
40:14 promotion you didn't ask for. Well the
40:16 agents get good
40:18 even there's there's going to be
40:20 coordination agents that are going to do
40:22 the coordinating.
40:25 So, there's going to be fewer people
40:28 that have to look over, have to manage
40:31 the agents because they're going to get
40:32 better and better and better.
40:35 Um,
40:38 and then we've got this this vacuum,
40:40 this vacuum of meaning where to the
40:42 extent that you've got yourself
40:44 self-worth and your identity tied to
40:46 your job as it gets ripped away, what
40:49 are you left with?
40:50 And I think for all of us, it's an
40:53 incredible opportunity. I think the
40:55 people that hang out on this channel,
40:57 you're already thinking this way, right?
40:58 You're already thinking like, "What do I
41:00 want? What do I want to do? What can I
41:02 do now that I have all these skills?"
41:04 But a lot of people are not going to
41:05 feel that way. So, the youngans will not
41:09 be tied to their work identity. I don't
41:11 think certainly if they can't get entry-
41:13 level jobs, they won't be.
41:18 It's going to be an introverts world. I
41:20 don't know. Mary Mary, I think I think I
41:23 think that um
41:29 I think there's going to be a a return
41:32 to analog. I think there's going to be a
41:34 return to in-person meetings. I think
41:37 there's going to be a slowing down. I
41:38 think the only way through an identity
41:41 crisis, a forced identity crisis, is to
41:45 slow down.
41:48 and and and probably get your ass in
41:50 some sort of community, right? Because
41:52 there's going to be a lot of people
41:53 going through this. So,
41:56 the more we can hear stories of people
41:58 that are
42:00 navigating their way through this, I
42:03 think the the better it's going to be
42:04 for everyone. But it's it's crazy, man.
42:09 Crazy.
42:11 Work-wise, it's going to be an
42:12 introverts world. Well, yeah. Work's
42:15 going to get weird.
42:17 I just think it's Yeah, it's going to
42:19 get more and more and more isolating.
42:21 Like at some point it's just going to be
42:23 some guy just some guy looking at a a a
42:26 green light and if the light turns red
42:28 he's got to go restart servers or
42:30 something like that.
42:36 The same thing happens when you retire.
42:37 It doesn't last long. Well, but you know
42:40 the thing about retiring to block Tom is
42:43 you kind of know it's coming, right?
42:45 when when you retire, you've got years
42:47 to say, "Okay, I'm not going to be doing
42:50 this job anymore." And and generally,
42:53 you know, if you're approaching
42:55 retirement at your job, you're getting
42:56 sideways promotions or you you know, you
42:59 know, you're just doing your job and at
43:01 some point they're like, "Ah, you're
43:02 going to be retired in a year. Just
43:04 just, you know, slow it down." You've
43:07 got some sort of transition.
43:09 >> The thing that I think is going to make
43:10 this different is
43:12 you're 40 years old. You're 20 years
43:14 into your career, you're kicking ass,
43:17 and all of a sudden,
43:20 AI automates your job and it automates
43:23 your sector.
43:26 So, you can't go get that job somewhere
43:28 else.
43:32 And you're 40
43:34 and you're like, I went to school for
43:36 this. I got a master's degree.
43:38 I worked at the best consulting firms in
43:40 the world. I'm doing all this
43:42 consulting. And then
43:46 right you're in some sort of analytical
43:48 role because you love analysis and you
43:50 love supporting the team. You're not the
43:52 front guy, right? The front guys, the
43:54 people that have the human relationship,
43:57 you know, uh interface, they're they're
44:00 probably going to be fine because you're
44:02 going to need people. People are going
44:03 to need to trust people. Tik Tok pin. I
44:05 built a GPT for my exec and transition
44:08 group. add LinkedIn profile
44:12 and it Oh, that's really cool. Source
44:15 Camp. Um,
44:17 ping ping me and let's talk Yumi and
44:20 Andy and Brandon about that because I I
44:24 want to start taking the great repurpose
44:26 and
44:28 finding human beings that are doing
44:31 stuff with this. I want to start finding
44:32 tools where people can sort of self-n.
44:36 So, so let's let's talk about that. um
44:39 and put it in there. That's really cool.
44:42 You need to retire to something, not
44:44 from something. That's exactly right,
44:46 Steo. And so,
44:49 so it's just going to be this it's just
44:51 going to be a lot of people that are
44:52 just going to have this this like
44:54 band-aid ripped off
44:57 to a wound that wasn't a wound, right?
45:01 That is now a wound. like like in a
45:04 society where we tie our work to our
45:07 self-worth and our identity, it's okay.
45:10 At some point it may break down, but for
45:12 the most part you've got time to
45:15 navigate that. But if all of a sudden
45:18 lots and lots of people have that those
45:20 tasks stripped away and they haven't had
45:24 time to process this, they haven't had
45:26 time to prepare for it. And now all of a
45:28 sudden they're just like all that [ __ ] I
45:31 did for 20 years is valueless.
45:38 What's you know like the self-esteem
45:40 there is just going to be brutal.
45:43 The fact of the matter is
45:46 there are things you do when you do your
45:48 tasks like how you do your tasks and the
45:52 integrity that you do your job with and
45:54 the amount of give a shittness you have
45:56 about your job that is the stuff that's
45:59 valuable val valuable about who you are.
46:03 But if you haven't done that work to
46:05 figure that out and and you just have it
46:07 tied to the tasks and those tasks get
46:09 ripped away, that's the thing that's
46:11 gonna be [ __ ] brutal.
46:15 The great repurpose could really be
46:17 aimed at retirees. I think it can be it
46:19 can certainly be aimed at retirees, but
46:21 I think it can be aimed at at really
46:23 anyone that's going through this
46:25 transition. That's the thing. I don't I
46:26 just I don't think this is just about
46:28 retirees. It's about retirees right now,
46:30 but like in the next three years,
46:33 there's going to be a lot of people
46:35 offered early retirement
46:39 that perplexity model context video.
46:45 What is that
46:49 model? Oh, model council. All right,
46:52 we'll go look at that
46:54 anyway.
47:06 I'm hopeful that the meaningful work
47:08 will come back once we figure out
47:11 what counts as an economy.
47:19 Not working for dollars, but working for
47:21 for fulfillment. I I think that's the
47:24 transition.
47:26 We're at the beginning of
47:30 it's the transition part that I think is
47:31 going to be painful.
47:34 All right, let me see this thing. What
47:35 is this thing? Is this at all
47:37 interesting?
47:42 Run three Frontier models at once.
47:44 Compare outputs. Oh, that's kind of
47:46 cool. Let's watch that video. That's
47:47 worth looking at.
47:51 Does this have audio with it?
48:35 You know, it's funny. I was thinking
48:37 like last week I was thinking about um
48:41 let's go to Perplexity and see if it's
48:43 live. It might not be. Um,
48:50 do we have three models?
48:53 Upgrade for the latest models. No, it's
48:57 not live yet. Or it's it's at least not
48:59 live for me.
49:02 >> It's under the plus sign if you have it.
49:05 >> Oh, is it?
49:08 >> Model council.
49:09 >> Oh, max. You have to you have to be a
49:11 max. $271 a month for the model council.
49:16 I'm good.
49:18 Yeah, I was thinking I was thinking that
49:20 um
49:24 there's probably going to be a lot of
49:28 money to be made if someone can figure
49:30 out
49:33 some sort of AI interface
49:37 that maintains your memory independently
49:40 of the models you use. and it it
49:42 dynamically figures out all the
49:44 different tools you could use. Um, and
49:47 it maintains your your your memory
49:50 because right now all the frontier
49:52 companies are trying to get you to, you
49:54 know, imprint on on their on their
49:57 particular flavor of AI.
50:01 Oh man, what's happening in DevO?
50:05 We're shaking.
50:09 Um,
50:11 M I also thought it might be interesting
50:14 to just go through Twitter and just see
50:16 if there's anything interesting to look
50:19 at.
50:24 Picture of a dog getting a haircut.
50:26 That's always nice.
50:32 Everyone's talking about Cling 3, but
50:34 the weird thing about Cling 3 is you
50:36 can't get it at Cling yet. You can get
50:38 it at a bunch of other sites, but not at
50:39 Cling.
50:44 It's Nikki Weiss. Hey Nikki, what's
50:46 happening
50:47 then? What happens to your memory? You
50:50 mean your real memory?
50:56 Like what I'm talking about is right now
50:59 I've got by far more of my interactions
51:02 in chat GPT than in any of the any any
51:05 of the other models. And as a result, I
51:07 don't really unless I have a specific
51:09 thing where I want to try a different
51:10 model. I don't just in an adventury kind
51:14 of way go off and use other models
51:17 because I don't I don't want my memory
51:20 fragmented. I want chat GPT to know as
51:22 much about me as possible so I can say
51:24 based on what you know about me, you
51:25 know, write a little bio for me or write
51:28 a little something, write a little
51:30 article and it it'll understand how I
51:32 write and all that sort of stuff.
51:40 Yeah. Yeah. Yeah. Exactly. Um,
51:46 so if there was some sort of central
51:47 service that had hooks into all those
51:50 other services that was basically
51:54 like what would be cool like I'm trying
51:55 to figure out where this would live.
51:56 This would probably live as like a like
51:59 a an operating system
52:03 extension
52:06 where it understands that you're going
52:07 to go off and use a bunch of different
52:09 AI things and it's sort of sucking in
52:11 your interactions with them into a
52:12 single into a single memory. I don't
52:15 know. I just it feels like there's an
52:17 opportunity there.
52:19 Feels like there's an opportunity there.
52:23 All right.
52:25 A lot of people quote tweeted this.
52:29 Oh wow. Andre Karpathy on the one-year
52:32 anniversary of of him coining the term
52:34 vibe coding.
52:39 That's great.
52:43 Cling three.
52:46 Instead of watching a three-hour movie,
52:48 watch this master class instead.
52:52 This was the uh the new what was it
52:55 called? Clawcon
52:57 for for you know in San Francisco.
53:00 They're all going crazy over open claw.
53:04 I wonder if has anyone in here actually
53:06 installed openclaw yet? I was thinking
53:08 about doing it but I don't I just don't
53:10 have two days to spend geeking out and
53:13 being frustrated that I'm not a better
53:14 coder. GPT 5.3 codeex with image gen
53:19 skill.
53:20 I'm loving this model so much. This is a
53:23 oneshot platformer.
53:26 That's cool.
53:28 I've left anthropic to start something
53:30 new. Cool.
53:38 I'm considering
53:41 an app aggregator. It's sort of an app
53:44 aggregator, but it's a it would be a
53:46 smarter
53:47 it would be smarter than that.
53:50 Like the thing I'm envisioning is
53:54 you probably have what it would probably
53:56 look like is this. It would probably be
53:58 a small
54:00 a small language model that's running
54:02 locally on your machine
54:06 that's got a really slick interface on
54:08 it. So, it can render images. It can be
54:12 a chat interface. It can be a voice
54:15 interface. It can it can shape shift
54:19 and then it's got hooks into all of your
54:21 accounts for all the other things. So,
54:24 you can just give it a prompt and you
54:25 can let it figure out which tool is the
54:28 is the right one to do or if you want to
54:30 if you want to just focus on one, you
54:31 can select one. You can let it select.
54:34 You could do multi-tool, you know, you
54:36 know, send this out to Perplexity,
54:38 Claude, Chat, JPT, and Gemini, send the
54:41 same prompt to all of them, get the
54:43 responses back. So, it would be a
54:45 manager kind of layer, but then the
54:47 important piece of it is it would be
54:50 capturing all of your prompts and all
54:52 the responses into into a memory
54:56 that it maintains that you maintain your
55:00 memory, not chat GPT.
55:06 Couldn't you just store it on your hard
55:08 drive? Possibly. But if you're just
55:12 storing it like a rag um you know
55:16 embedded embedded database model like a
55:20 vector database locally maybe I don't
55:22 know like I don't know I don't know
55:24 enough about it. I just know as an
55:26 enduser
55:28 it is frustrating to me right now that
55:30 I've got all this memory in OpenAI and
55:34 none of it in these other places. Like I
55:36 can't easily leverage the work that I've
55:40 put in
55:42 in Open AI. I can't easily leverage that
55:45 in Claude.
55:47 So I don't know.
55:51 It's not a huge deal right now, but I
55:53 think it's going to be a huge deal over
55:54 time. Ethan Mullik had early access to
55:56 Opus 4.6.
55:58 Here's the result of the same prompt to
56:01 create a control panel for a spaceship
56:02 in the distant future
56:05 done using the standard clawed
56:07 interface, not code. I had it zoom into
56:10 a few subsystems for the details
56:15 here. Okay. I I had
56:18 Okay. Did Did he give the prompt? Kind
56:20 of amazing
56:28 code. Play it.
56:36 Kind of amazing. So, let's see.
56:42 Does he give us the prompt? We just go
56:45 make up our own.
56:47 Let's go play with Claude. Okay.
56:52 Had early access to Opus 4.6
56:56 for the same prompt. Create the control
56:59 panel.
57:01 Create the control panel for a spaceship
57:04 in the distant future.
57:10 All right, let's go play with this.
57:17 So, I'm going to Claude cla.ai.
57:21 We now have a new thing, CL Opus 4.6.
57:26 So, I'm going to say um create, let's
57:30 see, I'll just paste in what he wrote.
57:32 Create a control panel for a spaceship
57:34 in the distant future. I want it to look
57:40 like a functional
57:43 control panel
57:45 and even have interactive
57:49 elements.
57:52 I can
57:56 play with and
58:01 affect the interface.
58:08 I'm not sharing my my tab.
58:11 Now I am beautiful. Fantastic Bob. Use
58:14 blockchain interface so you can
58:17 container control everything through
58:19 encryption. That's not actually a bad
58:21 idea. Like I think I think having it on
58:24 the blockchain makes sense.
58:26 Having it be trustless makes sense.
58:30 Having it tokenized makes sense. Having
58:32 it distributed makes sense.
58:36 Um yeah, I think that's cool. I think
58:39 that's right. That feels right.
58:44 And like if it's living at kind of an
58:46 operating system level, then it doesn't
58:49 matter if I'm using codeex or or cloud
58:52 code or a browser or an an AI browser or
58:58 some application.
59:01 It would it would be able to pull all
59:03 that stuff.
59:10 Um, I've installed it on my VPS. Oh, uh,
59:14 Claude Claude Code I mean uh, OpenClaw.
59:18 It's not user friendly and you have to
59:20 put it together from scratch. Very
59:22 dangerous for people that just want it
59:23 to work out of the box. The UX sucks.
59:25 Good to know. Yeah, that's I I mean
59:29 it's a security nightmare right now.
59:31 It's a usability nightmare right now. If
59:33 you're not geeky, it it just doesn't
59:35 matter. Like just just ignore it. Like
59:38 we're
59:40 wait two months. In two months we're
59:42 gonna have the consumer version of
59:44 OpenClaw. And it might even be from
59:46 OpenClaw. But right now it's it's you
59:48 know if you're not a Lord Digital gods
59:51 then don't don't even play with it.
59:59 Um,
1:00:01 so what Opus is doing right now is
1:00:03 writing us code for our little spaceship
1:00:06 interface.
1:00:09 Want to meet up in Denver at ETH Denver?
1:00:11 Possibly. Drop me a DM in the AI salon.
1:00:18 I'm a little I'm a little um I'm a
1:00:20 little bitter about the whole NFT world.
1:00:23 I you know I I think I think the NFT
1:00:27 world got appropriately slapped back
1:00:31 um because it's just so full of fraud
1:00:34 and it's like it it hasn't gotten
1:00:36 better. Like it's still a it's still a
1:00:39 it's still a disaster.
1:00:42 Like every time I talk to anyone that's
1:00:44 got any sort of NFT project or
1:00:45 blockchain project, I'm like, "But
1:00:47 what's it actually do?" They're like,
1:00:48 "Well, it's it's on the blockchain." I'm
1:00:50 like, "Yeah, I know." They're like,
1:00:51 "Well, you can get tokens." "Yeah, I
1:00:52 know." What do you do with those tokens?
1:00:54 Well, you can you can spend them on what
1:00:56 on the thing that it is. What?
1:01:00 It's always a circular conversation.
1:01:04 I believe in the core technology. I just
1:01:06 like I just feel like the the toxicity
1:01:09 of that whole [ __ ] world is uh is
1:01:12 alive and well.
1:01:16 Got a LOL from Endo.
1:01:22 I mean, it's just like, [ __ ] man. Just
1:01:25 just
1:01:28 don't scam your friends,
1:01:31 you know? Let's start there. Let's just
1:01:33 start there.
1:01:35 Don't rug pull your friends.
1:01:44 At least with AI, I can [ __ ] up my own,
1:01:47 you know, my own stuff.
1:01:53 Oh, look at that. That's so cool. All
1:01:55 right. What can we do? Can we click
1:01:57 [ __ ]
1:01:59 Oh, wait. What did I do?
1:02:17 Why is
1:02:21 My four four-year-old would love this.
1:02:23 Yeah, but wait, next time try this in
1:02:25 co-work. No, [ __ ] you. Okay, let's say
1:02:29 um good start,
1:02:32 but most of the screens are
1:02:37 Wait, let's see. But the
1:02:40 but the main screen
1:02:44 is black. And when I flip
1:02:49 switches,
1:02:51 nothing happens.
1:02:53 This looks more like a corporate
1:02:57 dashboard
1:03:00 than
1:03:01 the controls of a futuristic
1:03:07 uhistic
1:03:09 spaceship.
1:03:12 Um,
1:03:14 shouldn't there be be more quantum
1:03:20 drive
1:03:22 kind of things and fractals
1:03:26 and frequencies? See what I did there,
1:03:29 Brandon? Frequencies and signal
1:03:35 from the latent space.
1:03:39 Uh, I'm going to yell at a little bit
1:03:41 here. Um,
1:03:44 you created
1:03:46 um a 2002
1:03:51 um finance dashboard.
1:03:56 I want
1:03:59 I want a [ __ ]
1:04:03 spaceship.
1:04:08 um control. Okay, let's let's see how it
1:04:12 responds.
1:04:14 Thinking about redesigning this.
1:04:17 Architecting immersive spaceship
1:04:19 controls with quantum effects, dynamic.
1:04:21 You're absolutely right. That was too
1:04:23 tame. Let's build something that
1:04:25 actually feels like piloting a vessel
1:04:27 through spaceime. Fractals, quantum
1:04:30 field visualizations, living waveforms,
1:04:32 the whole thing breathing and reacting.
1:04:35 Yes.
1:04:43 Oh my god, man.
1:04:48 NFT and blockchain have very narrow
1:04:50 usability. I actually think they've got
1:04:52 broad ones. I mean, as as much as uh
1:04:56 World ID got some [ __ ] for for being,
1:04:59 you know, kind of circular in in its
1:05:02 logic, like human identity actually
1:05:06 feels like that's a really important
1:05:08 potential use for blockchain. something
1:05:11 like we just talked about. We're
1:05:13 maintaining my memory across AI
1:05:15 interactions in something that I control
1:05:20 um and is distributed and you know non
1:05:24 uh you know you can't destroy you can't
1:05:26 overwrite your file basically like that
1:05:28 feels valuable to me um but we'll see
1:05:32 the whole idea of decentralized anything
1:05:34 was by definition going to be a scam
1:05:38 especially in metaverse remember where
1:05:39 the where AI was 10 years ago and watch
1:05:41 blockchain in the next two to four
1:05:43 years. Yeah, I don't I don't think
1:05:44 blockchain's not here for the for the um
1:05:49 I think so. I think blockchain is going
1:05:51 to be actually I think AI is going to
1:05:54 drive blockchain adoption and and NFTTS
1:05:58 too. We just won't call them that,
1:06:00 right? It'll just be tokenized
1:06:02 ownership, things like that.
1:06:06 Oh, twin. Yeah, twin. It's twin.so. I
1:06:10 think they they just dropped their new
1:06:13 agentic thing, right?
1:06:15 What did they drop last night? Are you
1:06:18 sure you want other apps to know how you
1:06:21 treat your poor chatbt?
1:06:27 All right, I guess this thing's coding.
1:06:29 Is it Is it coding?
1:06:34 What's it doing?
1:06:37 The code's just sitting here. It's
1:06:39 thinking. Creating a truly immersive
1:06:41 thing.
1:06:43 Hi, I'm Claude. How can I help you
1:06:45 today?
1:06:47 Huh? All right. Well,
1:07:06 yeah, I used that that twin. I think
1:07:09 it's twin.so. Well, while this thing's
1:07:11 doing its thing, let's go check it out.
1:07:14 Um, twin. Yeah, twin.so.
1:07:19 the AI company builder. Try now.
1:07:26 I had it
1:07:29 I had it go
1:07:32 find me 10 tweets a day
1:07:35 that were like people talking about
1:07:38 human centric interaction with AI.
1:07:41 And so I get an email digest every
1:07:43 morning um
1:07:46 with 10 tweets. I don't read them,
1:07:50 but it worked.
1:07:52 It's like all the [ __ ] I tried with the
1:07:54 the Open AI agent and when you could do
1:07:57 automated uh prompts and chat GPT. I set
1:08:00 all those things up on this channel and
1:08:03 then I get emails every morning from all
1:08:04 those things I set up a year ago and
1:08:06 don't ever look at them.
1:08:13 Oh, good lord. Good lord. Good lord.
1:08:16 What would you like to automate? I don't
1:08:18 know. What would we like to automate?
1:08:20 Um,
1:08:27 I'd like you to find
1:08:36 research papers
1:08:40 that
1:08:42 generate buzz.
1:08:48 Every day
1:08:50 you find the sources
1:08:55 of the buzz,
1:08:58 then identify
1:09:02 the paper
1:09:07 and then
1:09:12 um
1:09:14 consolidate
1:09:16 the buzz into
1:09:19 why the community
1:09:23 feels this paper is a big deal.
1:09:29 Then do the
1:09:32 analysis
1:09:35 of the paper.
1:09:38 Tab tab tab. Oh, sorry.
1:09:40 Thank you. Then do I thought I thought
1:09:44 you were actually writing me a big long
1:09:46 screed of something and I looked over
1:09:48 just tap tap tap. Uh I'm at twin.so by
1:09:52 the way.
1:09:54 Um then do the analysis of the paper
1:09:58 to tell me if you think
1:10:05 the
1:10:06 it it matches the hype.
1:10:12 Pick one paper a day and email me
1:10:22 the results.
1:10:26 The structure
1:10:30 of the email should be
1:10:34 overview
1:10:35 paragraph
1:10:38 um hype
1:10:41 hype highlights
1:10:49 easy to understand overview
1:10:54 of paper for nontechnical
1:10:58 me
1:11:01 followed by
1:11:03 conclusion.
1:11:05 All right, let's see what this does.
1:11:11 That's a fantastic idea. Thanks, AI.
1:11:15 Thanks for telling me I'm awesome. I
1:11:17 didn't realize I was that awesome. I
1:11:20 have fantastic ideas. Huh. So cool.
1:11:24 A daily hype check on research papers.
1:11:26 Let me figure out what we're working
1:11:28 with and get the details from you.
1:11:30 Great. This agent has everything it
1:11:33 needs. Web search, deep research,
1:11:35 scraping platforms, Reddit, Twitter, and
1:11:38 email. Perfect combo for hunting down
1:11:40 hyped papers and doing a reality. Couple
1:11:43 of quick things I need from you. What
1:11:45 kind of research papers? AI and machine
1:11:48 learning.
1:11:50 So, that's cool. It's given me options.
1:11:52 Okay, perfect. AI machine learning is
1:11:55 the loudest buzz anyway.
1:11:58 Okay, here's the goal I'd craft for this
1:12:02 agent. Oh, good. It's going to write
1:12:05 It's going to write my goal. Daily AI
1:12:07 paper hype check every day. Search
1:12:10 Twitter, Reddit,
1:12:12 our machine learning, our local llama,
1:12:15 hacker news, and AI newsletters
1:12:18 for the single AI machine learning
1:12:21 research paper generating the most buzz.
1:12:24 Identify why the community is excited.
1:12:26 Collect the key hype highlights and
1:12:28 notable takes. Then read and analyze the
1:12:31 actual paper to assess whether the hype
1:12:33 is warranted. Email kyle@stofind.com.
1:12:37 Cool. So, it remembered who I was. Nice.
1:12:41 With the results structured as overview,
1:12:44 hype highlights, plain English
1:12:47 breakdown, and a conclusion.
1:12:50 Um, go.
1:12:53 All right, let's set it up.
1:12:59 Okay, back to Oh,
1:13:02 now we're talking.
1:13:07 All right. Can I what can I
1:13:10 What can I do? What can I move? Tab. Oh,
1:13:13 tab.
1:13:15 Let's see. How do I make this full
1:13:16 screen? Do I I can do that. That's cool.
1:13:23 Is this having any sort of effect?
1:13:28 No.
1:13:49 What's that? Throttle. Oh, that worked.
1:13:53 Okay, so throttle worked
1:14:00 except
1:14:02 now it's now we're going too fast.
1:14:05 Okay. Uh let's say this is better,
1:14:10 but other than throttle,
1:14:15 I can't see the effect of
1:14:20 any of the switches.
1:14:24 Everything should do something visible.
1:14:33 Um,
1:14:37 let's just give it that. All right,
1:14:40 let's go back to twin twin.so. Let's see
1:14:43 what's happening here.
1:14:46 Task in pro in progress.
1:14:49 So, I'll start by researching across
1:14:51 multiple platforms. Authenticated X
1:14:56 and it authenticated X because I've
1:14:58 given it that access before. That's
1:15:01 fascinating.
1:15:03 Um, let me check Reddit.
1:15:11 The EDOS deep think paper is generating
1:15:13 significant buzz. That's This is
1:15:16 actually a cool little agent.
1:15:19 Found the paper. Now, let's read it.
1:15:21 Now, we have the full paper. Let me
1:15:22 extract the key sections.
1:15:26 All right, it's doing its thing. I'm
1:15:28 digging it.
1:15:30 Let's go back here to code. Come back
1:15:33 out here to this thing.
1:15:55 All right, that's still doing its thing.
1:16:00 Now I have comprehensive data. Let me
1:16:01 also check the hacker news discussion
1:16:03 about this paper specifically. Then
1:16:05 compose the email. I now have
1:16:08 comprehensive data from all sources. Let
1:16:10 me comp compile the full analysis and
1:16:13 draft the email.
1:16:15 It's pretty slick, huh?
1:16:26 First dinosaurs. Do we remember before
1:16:28 the worldwide web when decentralized
1:16:30 ideology was niche?
1:16:34 Huh?
1:16:48 I now have comprehensive data from all
1:16:50 sources.
1:16:53 All right, that's still doing its thing.
1:16:54 Let's go back here. See if this thing's
1:16:58 done anything new. It's pretty [ __ ]
1:17:01 slick, you got to admit.
1:17:05 And Murphy,
1:17:06 what's shaking, lady?
1:17:10 We're making we're making quantum drive
1:17:13 um spaceship
1:17:15 um control panels with with Claude Opus
1:17:20 4.6.
1:17:22 I'm also going to make a uh an asteroids
1:17:24 clone because I I made one earlier and
1:17:26 it it it's about as close as I've seen
1:17:29 one of these things make to the original
1:17:30 game. It's pretty amazing.
1:17:36 Ah, okay.
1:17:38 Email sent successfully. Now, let me set
1:17:41 up the data interface and finish. Let's
1:17:44 go see if it did send the email. In
1:17:46 fact,
1:17:49 Kyle Shannon via Twin Daily AI paper
1:17:52 hype check.
1:17:55 We're still in outer space. Sorry.
1:18:00 Today's paper, semiautonomous
1:18:02 mathematics discovery with Gemini, a
1:18:05 case study on the air problem. So, erdos
1:18:08 are the 1500 math problems that no one's
1:18:11 been able to figure out, right? Or, you
1:18:13 know, whatever that they're they're
1:18:15 hard. Overview. Google deep mind
1:18:17 researchers built an AI agent called
1:18:20 Althia powered by Gemini Deepthink to
1:18:23 systematically attack 700 open problems
1:18:26 from the famous Aerdos problems
1:18:28 database, a collection of mathematical
1:18:30 conjectures
1:18:32 posed by blah blah blah blah. Okay. Hype
1:18:34 highlights. The headline claims Google
1:18:37 agents solved 13 new airdos problems.
1:18:43 This is
1:18:46 the plain English paper down. The erdos
1:18:48 problems are famous collection of
1:18:50 mathematical conjectures essentially
1:18:51 well-defined puzzles that mathematicians
1:18:54 haven't haven't yet solved or in some
1:18:56 cases haven't confirmed whether existing
1:18:59 solutions are correct. Paul Erdos was a
1:19:02 prolific 20th century mathematician. the
1:19:04 Google deep mind blah blah blah blah
1:19:06 blah. Okay. Conclusion. Does it live up
1:19:08 to the hype? This is a case where the
1:19:10 paper itself is far more careful than
1:19:13 the hype surrounding it. The authors
1:19:15 explicitly warn against overexitement.
1:19:18 This is actually a useful little [ __ ]
1:19:20 bot.
1:19:23 This is cool.
1:19:27 The actual contribution is modest. Five
1:19:29 genuinely novel solutions to the
1:19:31 problems. the authors themselves say
1:19:35 could have been easily dispatched by the
1:19:37 right expert. The remaining eight were
1:19:40 literature searches, not original
1:19:42 mathematics.
1:19:44 The paper's real value lies in its
1:19:46 honest methodology and candid discussion
1:19:48 of AI's limitations in research
1:19:51 mathematics, including a 76% failure
1:19:53 rate, even when the solutions appeared
1:19:56 correct. Social media transformed
1:20:00 this cautious empirical study into
1:20:03 breathless claims like AI cracking Aeros
1:20:06 problems, often conflating different
1:20:09 systems
1:20:10 and overstating mathematical
1:20:12 significance. The competitive framing
1:20:14 between Google and Open AI further
1:20:17 muddies the assessment. The genuine
1:20:19 advance here is method methodological
1:20:22 methodological. I think that's right.
1:20:24 Showing how AI can narrow search spaces.
1:20:26 This is pretty cool. I actually like
1:20:29 this,
1:20:32 you guys. I I could sell this as a
1:20:34 thing. You could subscribe to my daily
1:20:36 AI scientific paper. Hype check is twin
1:20:40 free. I don't think I've paid for it.
1:20:42 Maybe I have. I don't know. Have I? I
1:20:45 don't think so. Yeah. No, it says trial.
1:20:49 I'm in a trial.
1:20:51 I have 354 credits left.
1:20:56 I don't know what that means.
1:20:58 All right, let's see. Is this thing
1:21:00 working now? No, this is still this is
1:21:02 still uh this thing is still coding.
1:21:08 I I have one of the most powerful um
1:21:11 computing systems in the history of
1:21:13 humanity working on a toy for me.
1:21:24 Oh my god, man.
1:21:27 It's crazy. It is crazy.
1:21:40 What is this thing talking about right
1:21:42 now?
1:21:46 Oh, so here's the It just gave me what
1:21:49 it sent me.
1:21:53 Ah, here we go. Excellent first run.
1:21:55 That's media analysis. I like that it
1:21:58 caught the gap between the papers on
1:22:00 careful framing and the social media
1:22:01 frenzy. The author's literally warning
1:22:04 about misleading hype while getting
1:22:06 hyped is the chef's kiss. The workflow
1:22:10 is solid. It cast a wide net across
1:22:12 sources. Identified clear winner.
1:22:15 Exactly what you asked for. You'll get
1:22:17 one of these in your inbox every weekday
1:22:20 morning at 8:00 a.m. If you want to
1:22:22 tweak um
1:22:25 I'm going to send a message. I'm going
1:22:26 to say not just weekdays.
1:22:32 There's no sleep in AI
1:22:38 7 days
1:22:41 a week
1:22:44 and then a
1:22:48 weekly
1:22:52 wrap up report on Sunday.
1:22:59 That is a highlight.
1:23:05 Let's see. A highlight
1:23:07 roll for the weekly
1:23:11 for well let's see for the most
1:23:14 interesting
1:23:16 insights
1:23:18 from the weeks
1:23:21 analysis.
1:23:26 All right.
1:23:29 Love it. AI never rest. Neither should
1:23:31 the hype check.
1:23:37 I [ __ ] hate AI. I [ __ ] hate it. I
1:23:41 [ __ ] hate it. Like like
1:23:45 you know that it's like it's like a
1:23:47 young person, right? Put code in there.
1:23:49 Not even code. They wrote in there
1:23:53 be swell
1:23:57 blow sunshine up the user's ass.
1:24:02 Oh my god. Sunday bonus. Okay. They just
1:24:05 said
1:24:07 next run in 10 hours and 30 minutes.
1:24:11 All right. Well, that's cool. Let's let
1:24:14 me go see if there's who who sent that
1:24:16 to me. Twin.so. Let me go look for other
1:24:19 twin. So things twin.So
1:24:27 human plus AI collaboration digest. So
1:24:30 this was from two days ago.
1:24:37 So it looks like it stopped sending
1:24:42 this. So here's tweets.
1:24:47 It's kind of cool.
1:24:50 future work.
1:24:52 Oh, I'm not sharing. I'm a dick. Sorry.
1:24:58 All right, let's go back here. Is this
1:25:01 done yet?
1:25:04 I think it's done.
1:25:23 All right, it's doing something when I
1:25:25 flip my switches.
1:25:28 Oh, that one's cool.
1:25:50 All right. It's at this point it's a
1:25:53 total waste of time. Here's a good
1:25:55 lesson is I had no real
1:25:59 I had no real goal here other than make
1:26:03 something cool. So if you don't know
1:26:05 what you want, you get generic crap,
1:26:07 which I got. All right, let's do a new
1:26:09 let's do a new chat and then we'll get
1:26:11 out of here. We'll have it. Dream bigger
1:26:13 with Opus. Try Opus 4.6. Okay,
1:26:18 here's we're gonna Okay, Opus 4.6. Okay,
1:26:21 I'm going to say uh make me a clone of
1:26:26 the classic
1:26:31 Asteroids
1:26:34 arcade game
1:26:38 complete with analog
1:26:43 CRT
1:26:44 visual
1:26:46 effects.
1:26:50 Make
1:26:52 it as close to the original as possible.
1:27:02 All right, off we go.
1:27:09 Good evening, Dr. Shannon. Would you
1:27:11 like to play a game? Any idea on how
1:27:14 much it costs for Claudebot? Clubbot's
1:27:16 free,
1:27:19 but it uses your LLM to power it. Yeah,
1:27:21 it's you've got to hook your API to it,
1:27:23 so it's going to cost you API tokens.
1:27:26 Have you checked out Clawbot
1:27:30 name now changed to OpenClaw? Um, I have
1:27:33 not installed it. I was thinking about
1:27:35 it, but Lord Digital Gods did, and he
1:27:37 said it's if if you're not super geeky,
1:27:39 it's not worth playing with. Um, okay.
1:27:42 So, tomorrow office hours at 11:00 a.m.
1:27:45 So, if you go to my LinkedIn profile,
1:27:47 Kyle Shannon on LinkedIn, um you'll see
1:27:51 if you go to my events, you'll see uh AI
1:27:55 salon office hours. That's 11:00 a.m.
1:27:57 tomorrow. And then tomorrow here, Friday
1:28:00 night, date night, get your nacho
1:28:03 fixings ready. Go get yourself some Hot
1:28:05 Pockets for your dirty microwave. And uh
1:28:08 get yourself a date. or if you don't
1:28:10 have a date, I'll be your date. Come on,
1:28:12 hang out with me. We'll we'll do that
1:28:14 tomorrow. Is there a first Friday delay?
1:28:17 There might be, but I haven't heard any
1:28:20 any uh
1:28:23 rumblings from the misses about wanting
1:28:25 to go out. So,
1:28:32 let's see what twins up to. Has it done
1:28:34 anything? No. I guess it's just going to
1:28:36 do its thing.
1:28:38 Start AI amplification digest.
1:28:43 Memory updated. Task started. Test
1:28:45 completed. See you at 8. Done.
1:28:50 Um,
1:28:51 the last
1:28:57 of these I got was Feb
1:29:03 3.
1:29:05 What
1:29:07 happened to
1:29:10 yesterday?
1:29:13 And let me look at my 354 credits left.
1:29:17 All right. So, it's not I'm not blowing
1:29:18 through credits.
1:29:21 There's a run from yesterday that's
1:29:23 still active. It didn't finish. Let me
1:29:25 check what happened. Yesterday's run got
1:29:27 stuck. It authenticated with Twitter
1:29:30 successfully that stalled at step one.
1:29:33 Found the issue. yesterday's run.
1:29:37 Um,
1:29:39 I've nudged it to continue. Oh, all
1:29:41 right. There you go. Cool.
1:29:45 Oh, and I'm not showing my tab. Dang it.
1:29:51 Oh, for tomorrow office hours, I said
1:29:53 11:00 a.m. I didn't say what time zone.
1:29:57 It's uh 11:00 a.m. Mountain time. If you
1:30:00 go to the AI salon,
1:30:02 community.thesalon.ai AI. It's in the
1:30:04 events. So, we've got an events calendar
1:30:06 there. Um, and the links there as well.
1:30:08 Okay. Claude responded. All right, Kyle,
1:30:11 here's your faithful Atari.
1:30:16 There we go. Look at that thing. That's
1:30:18 slick.
1:30:20 That is slick.
1:30:26 That's pretty much what it looked like.
1:30:27 Push start. All right,
1:30:31 here we go.
1:30:35 It's got It's got a 90 degree problem. I
1:30:38 It didn't Oh, that's good. Okay. Uh you
1:30:43 have the ship
1:30:46 um oriented
1:30:50 90 degrees
1:30:53 off.
1:30:58 thrusting
1:30:59 makes it fly sideways
1:31:04 and the bullets
1:31:07 shoot out the side of the ship.
1:31:23 Uhoh, I've used up 90% of my session
1:31:26 limit.
1:31:30 Uh uh uh uh uh. The drawing rotation has
1:31:34 an extra plus math. Pi divided by two
1:31:39 offset. Quick fix.
1:31:47 Did you ever try Claude Opus 4.6 beta
1:31:50 with the 1 million token window? I did a
1:31:52 research with it earlier. Took 53
1:31:55 minutes. I I have not played with the
1:31:57 million token context window. That's
1:31:59 pretty cool though.
1:32:02 Um, is this done? I think it's done.
1:32:06 Okay, let's try again.
1:32:09 Yeah, there we go.
1:32:19 This is too fast.
1:32:22 Oh, you little [ __ ]
1:32:25 Oh, [ __ ]
1:32:30 It's too fast. Okay. Okay. It's too fast
1:32:36 and too aggressive.
1:32:43 Cool out. I'm old.
1:32:52 Oh my god.
1:32:55 I'm playing with Opus 4.6 right now. It
1:32:57 just made me my Asteroids game.
1:33:01 Hey Joy, what's shaking?
1:33:05 All right, let's see. Ha, fair enough.
1:33:07 Let me dial it back. Slower ship,
1:33:09 gentler ro gentle gentler rotation, more
1:33:12 chill asteroids and UFOs that give you
1:33:14 some breathing room.
1:33:26 If you've been watching this channel for
1:33:28 a while, I've been I've been doing this
1:33:30 Asteroids build thing every time one of
1:33:32 these new models comes out. And it's
1:33:36 like we're we're pretty much at the
1:33:37 point now where it just does it
1:33:42 like these things were were pretty bad a
1:33:45 year ago. like a year ago. This was, you
1:33:47 know, you could sort of see it was
1:33:49 asteroids, but it would make like square
1:33:51 square asteroids or just circle
1:33:53 asteroids
1:34:01 at work. Hope you're doing better, Kyle.
1:34:02 I'm doing okay, Joy. I'm still uh I I
1:34:06 got I got a I got a bad infection, but I
1:34:09 got new drops today, so I think I'm
1:34:11 good. I think I'll be good by the
1:34:13 weekend.
1:34:25 Recalibrated difficulty settings for
1:34:27 accessibility.
1:34:31 All right.
1:34:33 Here we go.
1:34:38 Are you going to reload the thing?
1:34:48 Okay, reloaded. Okay, here we go.
1:34:56 That was a little too easy,
1:35:02 but that it's got the good anxiety sound
1:35:05 of the UFOs.
1:35:20 It also has uh auto fire. It has machine
1:35:22 gun fire, which that's not accurate.
1:35:41 [ __ ]
1:35:52 Last little
1:35:59 I can't wait to see Cling 3.0, Kyle.
1:36:02 Yeah, same here, Joy. like but it's not
1:36:04 at cling.com. You have to get it at a
1:36:06 bunch of other sites that I just
1:36:09 always does that. I don't like it.
1:36:20 All right.
1:36:34 Anyway, phone now. It's free on
1:36:37 Higsfield. Oh, that's interesting. I
1:36:40 don't use Higsfield.
1:36:42 I don't think I have any even any uh
1:36:44 what you call it over there. I don't
1:36:46 think I have any credits on Higsfield.
1:36:51 Cling 3.0.
1:36:55 Higsfield plus your wallet. What the
1:36:57 [ __ ] is that?
1:36:59 Are they they're doing they're tying
1:37:01 cling to NFT wallets now.
1:37:06 Jesus.
1:37:07 Um
1:37:10 log in.
1:37:25 How many
1:37:29 things. Do I have 60 credits available,
1:37:33 nine credits to generate? All right,
1:37:36 let's generate something. All right,
1:37:38 we'll do we'll do my classic.
1:37:41 Oops, I got to share my tab.
1:37:46 Try it out. Okay. Oh, you have a CPP
1:37:49 with Higsfield. Way to go. That's
1:37:51 awesome. Okay, so we're going to do um
1:37:55 1970s
1:38:01 Uh, muscle car resto mod
1:38:08 in an abandoned
1:38:12 factory starts
1:38:16 and drives out of the
1:38:23 space.
1:38:25 Cling 3.0 0 5 seconds. Choose duration.
1:38:31 All right, we'll do 5 seconds. That's
1:38:32 fine. Oh, and you can do 1080p.
1:38:36 Wait.
1:38:38 Oh, you've got to you got to pay for it
1:38:40 to get it. God, I have 60 credits. Let
1:38:43 me spend my credits, you [ __ ]
1:38:48 All right. I can't play with it.
1:38:56 Yes, get smart. That is correct.
1:39:04 Four notes.
1:39:15 Oh, it looks like that hung again.
1:39:18 All right. Whatever.
1:39:20 These agents these days kind of suck. Uh
1:39:24 I'm gonna get out of here. Let me turn
1:39:25 this all off.
1:39:28 And uh I don't know. I hope you had fun
1:39:30 tonight. I was kind of all over the
1:39:31 place, but then we did some stuff. We
1:39:34 did some stuff. Got a little wax and
1:39:36 poetic, a little rage, little demo.
1:39:40 You know, it was it it was a uh an AI
1:39:44 learning lab smores board tonight.
1:39:54 I think there's a toggle to do free for
1:39:56 a few days. Oh, that's cool. All right,
1:39:58 I'll try that.
1:40:00 Um,
1:40:02 all right, everybody. Um, have yourself
1:40:04 a good night. I will see you tomorrow at
1:40:06 office hours, 11:00 a.m. Mountain time
1:40:08 on LinkedIn or I'll see you here
1:40:10 tomorrow for Friday night date night.
1:40:12 All right.
1:40:14 I'm not sure if they're Truman Capotes
1:40:15 or Harry Potter's glasses you're
1:40:17 borrowing. somewhere between the two.
1:40:24 Hello, I'm Truman Capot.
1:40:27 He had a wild voice, that dude. All
1:40:30 right. Um, have a good night everyone.
1:40:32 I'll see you tomorrow.