4/20/2026 - Testing Local AI Performance on the M5 MacBook Pro with Qwen 3.6 and OpenClaw

Description

Chapters

Transcript

0:00 Oh, champ. Oh, champ.
0:05 [music]
0:13 [music]
0:18 Come here, Champy. You want me to put
0:20 you little your little bed over here?
0:24 Come here.
0:26 There you go. Lay down there. Now, let's
0:28 do a little singing. What do you think?
0:31 You ready to sing?
0:33 [music]
1:04 >> [music]
1:19 [screaming]
1:23 [singing]
1:23 >> Happy Monday everybody.
1:27 Welcome.
1:28 And by everybody I mean Steo and
1:31 Nicholas
1:33 and
1:34 Dr. J. I think that's it. Oh, I show
1:39 eyeshow's here. All right.
1:44 >> [music]
1:49 [music]
1:54 >> Woohoo!
2:06 [music]
2:07 Wow!
2:13 >> [music]
2:18 >> Woohoo!
2:20 [music]
2:26 [music]
2:28 You're stinky, champ.
2:31 You're a little stinky.
2:33 He's a little stinky. [music]
2:37 You know when their glands get a little
2:39 funky?
2:41 You smell like bad salmon.
2:44 [music]
2:52 [singing and music]
3:01 >> [singing and music]
3:02 >> Come on.
3:04 Come on.
3:07 [music]
3:11 [music]
3:18 [music]
3:24 [music]
3:31 >> [music]
3:41 [music]
3:47 >> Source camp is on time.
3:51 She set the alarm. Here's the thing
3:52 about intentionality, people.
3:55 You set the alarm. You wake up from your
3:57 nap. Why did you take a nap? Because you
4:00 were out educating the people about AI.
4:02 It's exhausting. They don't want to hear
4:04 it.
4:05 They don't want to hear it. I don't I
4:07 don't want to. I don't want to. I don't
4:09 want to. So, you get home from your long
4:11 day of I don't want us. You take a
4:13 little nap. You set the alarm. You show
4:14 up to AI learning lab. That's how it
4:16 works. That's how professionals do it in
4:19 this world.
4:21 Kyle bugging my house. [laughter]
4:28 >> [music]
4:34 [music]
4:40 [music]
4:49 [music]
4:50 >> Boo. Cool.
4:57 >> [music]
5:01 [music]
5:07 [music]
5:12 [music and singing]
5:20 [music]
5:25 >> I have to listen with headphones
5:27 otherwise your dog will start howling.
5:28 It happens. Happens to the best of them
5:31 doggies.
5:33 [music]
5:37 [music]
5:43 [music]
5:48 [music]
5:50 Um, so so tonight we'll just have some
5:54 fun tonight. If if anyone has any
5:55 questions they want to ask about
5:58 stuff or nonsense, feel free to feel
6:00 free to pop a question or a thought or a
6:03 comment into into yield old
6:06 yield old chat. Um,
6:10 this week, as you may remember from last
6:12 weekend, I know I know you're all uh,
6:16 you know, trying to run your schedules
6:17 around mine, so you all know this
6:20 already, but um, tomorrow morning at
6:24 like 7:30, I leave for Washington DC
6:30 and then, uh, do a kickoff of a AI salon
6:34 flyin in DC. Um,
6:39 so that'll be nice. [music] Welcome
6:41 welcome some companies in.
6:49 [music]
6:57 [music]
6:58 Um, and then the other thing that's
7:00 going to happen this week, so, so I go
7:02 to I go to DC tomorrow, and then
7:04 Wednesday morning I fly to Boston, do a
7:07 kickoff of an innovation summit, uh,
7:10 Wednesday night, and then it's all day
7:12 Thursday and some activities on Friday.
7:15 So, my week this week, I'm pretty much
7:17 out. And this is also the week where it
7:20 looks like we're going to get a new
7:21 model from OpenAI.
7:24 We just got 4.7 from Anthropic. [music]
7:30 But wouldn't surprise me if we got
7:32 something from Google as well. [music]
7:35 Although I haven't heard any buzz about
7:37 Google dropping anything.
7:41 [music]
7:44 What do I think about Perplexity
7:45 Computer?
7:47 Um, I have only played with it a little
7:51 bit and I haven't quite distinguished
7:58 >> [music]
8:01 >> if it's any good or bad or or or
8:03 different. Like right now
8:06 for the Agentic stuff, I'm putting my
8:09 attention into my Open Claw
8:11 installation, which was on a PC. Now
8:14 it's on a Mac. That that transition took
8:18 some time.
8:20 And he sort of works now. And he's
8:22 running on an all local model, which is
8:23 super cool. I'm running Quen 3.6. 6. So,
8:26 that's pretty cool. Um, I can probably
8:29 want run Perplexity Computer. And then
8:32 I'm also seeing now there's people that
8:34 are running uh what the hell is that
8:36 thing called? Um,
8:39 there's OpenClaw. Hermes. And there's
8:42 Hermes. And Hermes is like OpenClaw, but
8:44 different. And now there's people
8:45 running Hermes and Open Claw on the same
8:49 machine where they check each other's
8:51 work. [snorts]
8:53 So I it's the agentic stuff is such a
8:58 different discipline
9:01 that I don't I don't quite have my head
9:03 around it yet. [music]
9:10 [music]
9:15 The thing about agentic work,
9:19 [music]
9:20 about agentic systems like openclaw
9:25 is that
9:27 you can just do stuff, right? You can
9:29 just have it go do [ __ ] Make me a
9:31 report. Make me a this. Make me a that.
9:33 It'll go off and do that [ __ ] The
9:35 problem with it is,
9:40 do you actually need that [ __ ] Because
9:43 if it's just off making [ __ ] that you
9:45 got to now pay attention to, it's not
9:48 actually solving a problem, right? Well,
9:51 so then you go back to, okay, well,
9:56 what am I trying to accomplish? And I've
9:57 got different companies and different
10:00 projects going. Do I know exactly what I
10:03 want to accomplish?
10:06 Maybe. But I don't know how much time I
10:08 want to put into training a shitty AI
10:10 employee to halfass do the work that I'm
10:13 just going to have to go delete and deal
10:14 with at some other point. So I am having
10:17 I am having birthing problems with my
10:19 agents.
10:21 Not because [music]
10:24 they're so hard to understand. They're
10:25 actually decently easy to understand.
10:27 But the question of what do you want
10:29 them to do is actually kind of a meaty
10:32 question. When people tell me they want
10:35 an agent, I ask them what they'll use it
10:37 for. I usually get, "Huh?" Yeah,
10:39 exactly. I I gotta have an agent. Are
10:42 you talking about a custom GPT? Are you
10:44 talking about an an agent like Manis or
10:49 GenSpark? Are you talking about have a
10:51 suite of autonomous agents running on a
10:53 chip locally on your device? Or or are
10:56 you are you willing to run up a $2,000 a
10:59 month bill to run this on the most
11:02 powerful model out there? What?
11:07 I want I want an agent for my business.
11:09 Okay, define agent.
11:13 You're the expert.
11:17 How do I feel about putting OpenClaw on
11:18 a personal computer? So, I just I just
11:20 bought I'm I I basically came into a
11:25 little bit of money. Um my stepmother
11:27 passed away. Didn't have much money, but
11:30 she gave us a little bit. Some for the
11:31 kids, some for the wife, some for me.
11:33 So, I basically took as much as I could
11:35 of that. And um bought myself a MacBook
11:40 Pro, an M5 MacBook Pro.
11:44 So,
11:45 this is this is fascinating.
11:48 So Brent Peterson, who if you don't
11:50 know, he's he's a guy that hangs out on
11:53 office hours. He's got a a company
11:55 called Signal and Cipher, and they've
11:56 got this whole he's been developing like
11:59 an agent governance platform for two and
12:02 a half years. So like his company is
12:05 basically poised for this moment. And at
12:08 the time I ordered the MacBook Pro,
12:10 Brent Peterson said, "Don't bother
12:12 getting a powerful machine because all
12:15 the models suck. they like they're not
12:18 none of the open source models are good
12:19 enough to actually run something like
12:22 openclaw all locally.
12:25 By the time my computer got here, which
12:27 was like a week and a half later, um
12:30 Quen 3.6 35 billion parameter model came
12:33 out and it's really good. And Brent
12:35 Peterson said, "Oh, this this is
12:37 actually good enough to run run these
12:39 agents on." So, we have officially hit
12:42 the tipping point where at least one
12:45 person I deeply respect thinks that
12:47 open-source local models are good enough
12:50 to run something like OpenClaw without
12:52 having to go to the to the to, you know,
12:56 OpenAI or Anthropic. Now, if you want to
12:59 do kick-ass high-end coding, great. Go
13:02 do that. But like I I'll give you an
13:04 example. Um, one of the things that Adam
13:07 did is I had him create a daily
13:13 The Great Repurpose signal, which was
13:16 basically go find five news articles
13:18 that are related to The Great Repurpose.
13:20 Put it in a nicely formatted email where
13:24 you talk about here's the five articles.
13:25 Here's why I picked them. Here are the
13:28 the five stages of the Great Repurpose
13:30 that you know are are relevant here. um
13:33 and include a picture from the from the
13:35 article in the email. Um the quality of
13:39 that email when I was using um
13:44 chat GPT 5.4 and when I was using Quinn
13:47 3.6 locally, it was the same. Like I
13:50 didn't see a degradation in quality. Um
13:53 I'm sure there's areas where you see
13:55 lots of degradation in quality, but it's
13:57 pretty flipping good. So he So I'm I'm
14:01 writing an article right now. Uh Audra.
14:03 So is it Wait, it's Audra. Audra.
14:07 Audrea. Audrea. Audrea.
14:12 [laughter]
14:13 I don't know how to say your name.
14:16 Um I'm actually writing an article about
14:19 it right now. Here's how I equate it.
14:22 When I was a kid in the 70s, you'd go to
14:25 the arcade and you'd play Asteroids and
14:27 you'd play Space Invaders and you you'd
14:29 play the video games, right? Play
14:32 Pac-Man.
14:34 And then there were all these rumors
14:37 that they were coming out with home
14:39 video game consoles and the Atari 2600
14:43 came out
14:45 and I got myself an Atari 2600 and my
14:49 favorite game was Asteroids and
14:50 Asteroids was amazing. The physics were
14:53 amazing and it it was spooky and it was
14:56 dark
14:57 and then Asteroids came out for the
14:59 Atari 2600. I'm like, "Oh my god." And
15:01 and if you remember back then, the box
15:04 art for Atari 2600 games were was done
15:09 by illustrators who would illustrate
15:12 their interpretation of what the game
15:14 looked like. So you'd see the game in
15:16 the store and you'd be like, "Well, it
15:17 looks so good." And you get it home and
15:19 you plug the cartridge in and it was
15:21 this stinking piece of [ __ ] these
15:23 blocky pixels. And the the ship barely
15:26 turned around. It was I think it was
15:27 like eight points. It wasn't smooth. It
15:29 was like D and then the the sounds were
15:32 all crappy and the graphics were all
15:34 crappy and the gameplay was all crappy.
15:36 It was just horrible. So the gap between
15:40 your home video game and what you got at
15:42 the arcade was massive.
15:46 And then over the years it it kind of
15:48 caught up. And then the era where the
15:50 Nintendo 64 came out and Sega Genesis,
15:55 there was an there was a point at which
15:57 the video games that you could play on
15:59 your TV were were like at or above what
16:02 you could play in the arcade and and you
16:04 know so so as the as the quality of
16:07 games goes up, the arcades started to
16:10 die off, right? And went away and then
16:12 you know video games just got better and
16:13 better and better. I feel like we're at
16:16 a really similar point with open-source
16:18 models and hardware right now. So,
16:23 running any sort of open-source model
16:25 locally over the past 3 years was just
16:28 kind of it was kind of like playing the
16:30 Atari 2600 compared to the the video
16:33 game. Um, I feel like this Quen 3.6 six
16:37 model is the first one that's kind of
16:40 like, huh, that's that's like a similar
16:43 experience as to what I was having with
16:45 these big big models. Now, I'm just
16:47 using it informally, so I'm not in any
16:50 way comparing Quen 3.6 to one of the big
16:52 models. It's not, but like it's good
16:55 enough to run
16:57 uh run OpenClaw, run my my agent, Adam.
17:00 So, I can talk to Adam and he talks
17:02 right now and I got my my machine set up
17:04 and so I'm I'm a fan of it. Why can't we
17:08 load local LLM like the old Microsoft
17:10 CDs? Exactly.
17:12 [music]
17:18 [music]
17:20 Jeez seems to work better for me.
17:22 Fantastic.
17:25 Have you tried 3.6 Joker? 3.5 was good,
17:28 but 3.6 six. I'm using the 3.6 the six
17:33 the Q6
17:36 um K version the quantiz 6
17:47 [music]
17:55 >> [music]
17:59 [music]
18:08 [music]
18:15 [music]
18:19 >> Yes, but I'm just now experimenting.
18:21 Haven't had the GPU long. Yeah. So, the
18:23 the Mac I got I got an M5 Max with 64
18:26 gigs of RAM and I got the 14inch MacBook
18:30 Pro. It is it's it's a really sweet form
18:32 factor. I like it. I was I was kind of
18:35 bummed like there was a part of me that
18:36 was like wait for the new round of of uh
18:40 Mac Minis, the M5 Mac Minis, which I'm
18:43 sure are going to be baller like for,
18:44 you know, $600, you'll probably be be
18:47 able to get something that just destroys
18:48 everything that's out now. Um, but the
18:52 M5, the the architecture of the M5, I
18:54 thought my philosophy on Max has always
18:57 been just buy the most expensive thing
19:00 you can afford. You just you won't
19:02 regret it. It'll last you a while. And I
19:05 feel like I've got a machine now that
19:06 that I I'll be able to like as these
19:08 models get better and better, like it's
19:11 it's pretty solid right now. [snorts]
19:13 And it's a really cool form factor. I
19:15 don't need to plug in a monitor to it. I
19:16 can just carry it around if I want to or
19:18 I can just leave it. I've got it set up
19:19 so I can close it and it still is
19:22 running as a server and then I can um I
19:25 can remotely log into it from this
19:26 machine that I'm I'm broadcasting from.
19:29 It's pretty sweet.
19:31 Buy once, cry once. Yeah, exactly. Hey
19:34 everyone, sorry I'm late. What's up
19:35 tonight? Um it's it's going to be
19:37 tonight's going to have to make up for
19:39 the week. So because I'm I'm traveling.
19:41 I got a I got a really busy week this
19:43 week and next week. So it's going to be
19:45 it's going to be thin here in the AI
19:46 learning lab. Um,
19:49 so anything you want to talk about, we
19:51 can talk about. Um, anything. Maybe I'll
19:54 demo something. Maybe we'll just have
19:55 some fun tonight. I don't know.
20:01 It looks like we're going to get some
20:02 new models this week. Um, so that that
20:05 is probably going to be something
20:06 interesting to deal with.
20:09 Is that your Mac Mini? No. But I can
20:12 show you. You want to see something
20:13 cool?
20:16 And by the way, the the URL on screen,
20:18 community.thesalon.ai,
20:20 if you've not been to that, um, please
20:22 go join the AI salon. It's where we all
20:25 hang out in between these meetings.
20:33 They always do cool stuff when you
20:35 leave. [laughter]
20:37 Um,
20:38 okay.
20:41 So, you can see my this is where I just
20:43 ordered business cards for social media
20:45 marketing week, which is next week. Um,
20:49 but this I wonder if this is going to
20:51 work. Why?
20:57 Hang on a sec.
21:13 Black Bar takeover. I can't keep up with
21:16 all the cloud features. Can we review?
21:23 [music]
21:24 >> Yeah,
21:26 I think black bar has uh consumed your
21:30 entire display. Oh,
21:34 [laughter]
21:35 >> I know. I'm Hang on. Let me try
21:37 something here. Let me try to relog into
21:40 it. It could just be a screen redraw
21:42 issue. Hang on. I'm going to connect to
21:45 server. I'm going to go here.
21:55 Archetypal Architect on YouTube
21:57 meanwhile has a uh has a good
21:58 suggestion.
22:00 >> Okay.
22:11 Okay. Yeah, that's I'm in.
22:18 Yeah, I'm in. But am I still sharing?
22:21 Yes, I am. Okay. So, what you're looking
22:23 at here,
22:29 what's this?
22:36 What you're looking at here is no black
22:40 bar
22:43 mods. Can you please [clears throat]
22:47 pin my welcome comment to Dr. J.
22:53 Mary says, "Hello, Dr. Jay."
22:56 Okay, Eric, how good is that M5?
23:01 >> Please welcome my good friend,
23:02 >> not Dr. J.
23:03 >> Oh, not Dr. J. DJK to a regulars. Hey,
23:06 DJK. Welcome, welcome, welcome, welcome.
23:10 Um,
23:14 so what you're looking at here is this
23:16 is my um
23:20 my MacBook Pro.
23:23 And so here's a little this. And let's
23:25 see, here's my terminal. So there's my
23:29 little terminal window. So So basically
23:30 the way the way I work with it is is uh
23:35 I've got Claude right here. This is
23:37 Telegram. So this is Adam. So I can say,
23:38 "Hi, Adam. How are you?"
23:45 And then you'll notice up here at the
23:46 top right underneath Adam, it says
23:50 typing right there.
23:52 Um, and so, so Adam's thinking he's
23:55 probably waking up
23:59 and then, uh, and then, you know, if I
24:03 need to do any
24:12 Let's see. Hey, Kyle, I'm here. I'm
24:14 doing all right. Bit annoyed at the
24:17 busted YouTube Digest migration, but but
24:19 [laughter] otherwise, Steady. How are
24:21 you? I'm good.
24:24 Um, let's see. Um, Lord
24:28 Digital Gods
24:31 has sent us a migration
24:36 uh instruction
24:39 set.
24:42 So, we should be good to go as soon as I
24:48 get off
24:50 my butt.
24:51 >> [laughter]
24:53 >> Okay. So, so I'm just over here talking
24:56 to him. But what's cool about this is
24:58 I'm, you know, I'm traveling for two
24:59 weeks, so I can just remote into this
25:04 dude, this this uh M5 Mac and do all the
25:08 configuring I need to do. So, I've got
25:10 I've got a lot of things I want to play
25:11 with. Um, did you get rid of something
25:14 there? No. YouTube comment archetypal.
25:19 How about setting up an AI avatar to run
25:22 your channel while you're gone? That
25:25 should be a quick project.
25:28 [music]
25:33 [music]
25:36 But
25:40 I don't know that there's any AI
25:42 avatars. Yeah, I lost my screen share.
25:44 That's okay. I don't know that there's
25:45 any AI avatars
25:48 that you can turn into just like a
25:50 full-on interactive talking dude.
25:54 Champion responds to human Kyle. That's
25:56 probably true. Um Eric was asking about
25:59 the M5. Do I like it? Um I I got to say
26:03 that the the MacBook Pro M5 in its in
26:08 its baller configuration is an
26:10 incredibly small machine. like it's way
26:13 smaller than my my M1. It's way thinner.
26:16 Um, it's very fast. When I was running
26:20 [music]
26:23 I I was running something on it. Oh, I I
26:27 was running LTX Studio. I was running
26:29 LTX Studio locally and it got really
26:32 hot. [music] So So if I'm going to be
26:35 doing a lot of compute on it, I would
26:37 probably get get it some sort of cooling
26:39 fan base.
26:41 um just to just to improve the the the
26:44 cooling of it. Speaking of it, did you
26:46 see Tim Cook is stepping down? I think
26:49 thank god um [music]
26:52 Apple needs to start innovating again.
27:03 Um but but to your point, I think Apple
27:06 Silicon,
27:08 they've been they've been very behind
27:10 like like laughingly behind on the AI
27:13 stuff, right? Like just like it's
27:16 ridiculous how far behind they are. And
27:19 and like Siri has been a laughingtock of
27:22 the AI agent world for more than a
27:25 decade, right? Siri's just an
27:26 embarrassment. Um [music] and so you
27:31 have this product that's kind of an
27:32 embarrassment and then everyone else is
27:34 doing this AI stuff that's remarkable
27:36 and you're [ __ ] Apple. So, like
27:38 you've got cash, you should be able to
27:41 do this as good as anyone on the planet.
27:42 You understand h the humanities more
27:46 than most other tech companies and and
27:48 they just they just missed the ball on
27:50 it. That said, their hardware division
27:53 is is making like the thing about the M5
27:56 chip is that it's got 18 or 20 GPU cores
27:59 and each one has its own AI accelerator.
28:02 On the M4, you had a single AI
28:04 accelerator that then routed it to all
28:06 the different GPUs. This now has an AI
28:09 accelerator on each GPU. Like, it's it's
28:11 incredible. It's incredible
28:13 architecture. I don't know enough about
28:15 how this [ __ ] works to know if it's
28:17 good. But what I can tell you is the the
28:19 Quen
28:20 35 billion parameter, the 3.6 version
28:24 with the with the quant the quant 6, the
28:27 quantiz 6 version is fast and good. It's
28:32 got reasoning. It's it it's can it can
28:34 see images. You can send it images. It
28:36 can see them. Um it's really quite good.
28:39 So yeah, so I think Apple's going to
28:40 catch up by by having computers that are
28:44 really strong inference machines and
28:47 then a lot of AI inference is going to
28:49 move off of OpenAI and Anthropic as as
28:52 these models get better and better and
28:54 as they get easier to install, right?
28:56 which is again Apple should [ __ ]
28:58 destroy it. Like like what Apple should
29:01 have is like you turn on your Mac, it
29:03 says, "Do you want AI?" "Yes." "Do you
29:06 want me just always to give you the
29:07 latest model?" "Yes." "Do you want to
29:09 make pictures?" "Yes." "Do you want to
29:12 make music?" "Yes." Right? Do you want
29:14 to do programming? Ah, maybe. Right? And
29:17 then it just goes out and finds all this
29:19 [ __ ] What are your priorities for these
29:20 things? And then it's just constantly
29:22 updating them. Like that would be
29:25 something Apple could kill
29:28 and people would buy things just for
29:30 that and then you'd run all your all
29:31 your AI locally.
29:35 An Apple harness running all that stuff
29:37 would be amazing.
29:40 [music]
29:42 But here's the thing. Tim Cook didn't
29:44 call me.
29:46 All he had to do is say, "Kyle, what do
29:48 you think we should do?" I'd told him.
29:52 >> [music]
30:00 [music]
30:03 >> Um, but yeah, I think their hardware is
30:06 pretty amazing.
30:11 It's such a slick form factor. It really
30:13 is. And they're they're completely
30:16 redoing the MacBook Pro line
30:18 um in the fall. So, they're going to
30:20 have touchscreens now. They're going to
30:22 be thinner. They're doing a whole new
30:24 architecture. I think that this this is
30:27 probably the 10th,
30:29 maybe not 10th, maybe sixth or seventh
30:31 [music]
30:32 version in this same chassis. Tik Tok
30:34 question. Kyle, what's your theory about
30:37 why Apple set sat back on AI? Um,
30:41 [music]
30:42 well, as innovative as Apple is,
30:46 sorry, as innovative as Apple was, um,
30:52 they tended not to jump on new
30:54 technology early. They would let they
30:57 would let people sort of [ __ ] around and
30:59 find out and then they would take what
31:00 worked and and run with that. So, they
31:03 would they they're what I would call a
31:05 fast follower or a true innovator,
31:07 right? They're doing stuff in their R&D
31:09 labs. They come up with, you know, the
31:11 the the iPad and the iPhone and and put
31:15 that stuff out in the world and change
31:17 the world with it. Um,
31:25 I I honestly don't know.
31:31 They're a consumer
31:36 they're a consumer computer company
31:40 even though a lot of engineers, you
31:42 know, use them. So So they're not
31:44 they're not really targeting engineers.
31:48 Um
31:50 AI isn't ready for prime time yet. I
31:52 would argue that if they had gotten in
31:54 the game early, like if if Apple were
31:56 one of the frontier model companies
31:58 along with OpenAI, Anthropic, and
32:01 Google, like they would just be doing
32:06 large language models and and probably
32:08 multimodal models on on just a they
32:11 would basically be coming at it from a
32:13 very philosophically different approach
32:15 where they would be balancing art and
32:17 science where the others are just going
32:19 full-on science.
32:23 So, it frustrates me that we don't have
32:25 Apple's voice in that conversation. That
32:29 said,
32:31 the models are just now getting good
32:33 enough where Apple can sit back and go,
32:36 okay, maybe we'll go with open source.
32:39 We'll take some open- source thing. Like
32:41 the the Apple operating system is based
32:43 on Linux, right? It's it's BSD.
32:46 So they can they can take an existing
32:48 open-source framework and and build, you
32:52 know,
32:54 consumer grade stuff on top of it. So So
32:56 they know how to do it.
32:59 I don't know. I I honestly don't know.
33:02 It just, you know, it was either too
33:05 early and unproven. It didn't really
33:08 they in the end they've got to sell more
33:11 AirPods and Macs and Apple watches and
33:15 right they've got to sell more [ __ ] So
33:18 if they made the calculation that AI
33:20 features are not going to sell more
33:22 [ __ ] Yeah, we'll just wait three years.
33:24 We'll we'll let these guys go battle it
33:26 out. We'll just pay attention.
33:29 But you know the fact that Tim Cook is
33:30 leaving, you know, I think I'm I mean
33:33 I'm not the only one bitching about the
33:35 fact that they missed it. Hey Kyle, I'm
33:37 building an app to automate social
33:39 marketing for authors.
33:41 What do you think I should do?
33:44 Well, the first one is build it.
33:47 [laughter]
33:49 The next one, okay, so here's what you
33:52 should do, archetypal.
33:54 You're building a a social marketing.
33:57 You're going to automate social
33:59 marketing for authors.
34:01 Automate social marketing for the social
34:04 marketing
34:06 automated social marketing tool for
34:07 authors. Use the tool to market itself
34:14 you know
34:20 something like that. Oh, you he thinks
34:22 he means like lovable. Oh, I'm building
34:25 an app to automate social marketing.
34:28 What do you think I should do in terms
34:31 of what?
34:33 Yeah, that's in the queue. Are have you
34:35 already built it? Are you building on on
34:36 lovable? Make videos and smoking jackets
34:39 and tobacco pipes. Yeah, Chef Kelly,
34:43 that would be good. Hey, Chef Kelly, you
34:44 want to hear something funny? So, you
34:47 know, when you did the podcast and you
34:48 sent me um you sent me snacks to to eat
34:52 on. Actually, your Amazon order got
34:55 split into two orders. One half of it I
34:58 got the day that I talked to you, but
35:00 the other half has been sitting in in my
35:03 company's um little post office boxes
35:06 for like a month and they're like, "Hey,
35:08 you've got a package down here no one
35:10 picked up and we're about to we're about
35:11 to send it back." I was like, "Oh,
35:12 okay." So, I went down there and it was
35:14 a whole bunch of candy.
35:16 Thank you. [laughter]
35:19 Apple played AI like it was another nice
35:21 upgrade while everyone else realized it
35:23 was the whole next game. Yeah, I don't I
35:25 honest to God I can't I can't for the
35:28 life of me figure out why Apple didn't
35:31 pounce on AI. I don't get it. I don't
35:36 get it.
35:38 Hey, you want to know? So, Chef Kelly
35:40 sent me So, she's doing this thing with
35:43 um helping big companies replace
35:46 artificial dyes with natural dyes. And
35:49 so she sent me the these sample candies
35:52 that are that use natural dyes and
35:54 they're tasty. They're good. But but
35:56 here's the problem, Chef Kelly.
36:00 [laughter] Because because they don't
36:02 look at as as bright and because they
36:04 use natural coloring, I'm like, "Oh,
36:07 these must be healthier for me." So I I
36:09 ate like eight of those little packs
36:10 today of like fake Skittles and fake
36:14 Starburst shoes, things like that.
36:17 [laughter]
36:20 >> [music]
36:21 >> Okay. Oh my god.
36:28 [music]
36:38 Another big food ruse.
36:48 I gave it a very simple prompt and it
36:50 spun up agents to build up much more
36:52 complex business. Already got payment
36:54 set and cold outreach.
36:59 [music]
37:03 I mean I think a thing to do archetypal
37:07 here here's
37:11 here's a thing that I think we could all
37:12 start experimenting with.
37:15 is
37:18 whether you use Perplexity Computer or
37:22 Manis or Gen Spark or Open Claw, whether
37:26 it's local or in the cloud, doesn't
37:28 matter.
37:30 Once you start getting your head around
37:32 agents and and to be clear, the thing
37:34 that I'm about to say, I'm not doing,
37:37 but I think this is a thing to do. So,
37:40 you've come up with this
37:44 automated social media marketing for
37:46 authors platform, right?
37:49 I would have I would go to
37:53 Claude probably put Claude into deep
37:56 thinking mode
37:58 and have Claude tell you,
38:02 give me a whole strategy for how I can
38:04 market this thing and break it down into
38:07 components. Here's how I find authors to
38:10 target. here's, you know, the the
38:13 campaigns I put together. Here's, you
38:15 know, you need to come up with a
38:16 creative campaign and you need to have a
38:18 creative brief for that. Have it come up
38:20 with a whole plan for how you're going
38:21 to do this. And then, you know, once you
38:24 get people to a landing page, how are
38:26 you converting them? Have it give you
38:27 the whole strategy for all that stuff.
38:30 And then just see if you can use agents
38:34 to do all of the components that it it
38:36 designs. Like I kind of feel like this
38:39 idea of a single a single operator
38:43 doing all of the all of the execution of
38:46 of a full strategy as if you had a 20 or
38:48 30 person company. Uh might be might be
38:52 a uh
38:54 might be worth I I think it would be
38:56 worthwhile if nothing else just to
38:58 learn, you know, just to learn [music]
39:04 because Apple sucks. That's why they
39:06 didn't punch on AI,
39:09 but really they're trying to save money.
39:11 Well, this is Listen, here's the deal.
39:14 Steve Jobs put Tim Cook,
39:18 you know, at the helm of Apple, you
39:21 know, on his deathbed. And I I
39:25 understand it, right? You have a like
39:27 Tim Cook was a CFO, right? So, he's a
39:29 numbers guy. So, what Steve Jobs, I
39:32 assume, was banking on was, "Hey, Tim,
39:34 you know, go keep the company running."
39:37 And and and Tim Cook's done a good job.
39:40 Like, basically, all of the innovations
39:41 that they did, they've maintained.
39:43 They've done a couple of, you know,
39:46 decent innovations along the way, but
39:49 Apple has been in a in a maintaining
39:51 mode for since Steve Jobs died.
39:55 Tim Cook stepping down makes room for
39:57 someone to come in. I like I don't know
39:59 who's coming in but I hope who they put
40:01 in as CEO is someone who's got some
40:03 [ __ ] vision
40:06 right [music]
40:08 you know there's maintaining and you
40:09 know maintain the growth maintain the
40:11 company build the cash reserves
40:14 so that they can survive anything
40:18 so they've done that well it's just that
40:20 you know the the the quality of their
40:22 products has gone down the quality of
40:24 their software has gone down and their
40:27 innovation pipeline is just crap.
40:30 [music]
40:35 Making data centers is different
40:37 business than selling hardware. Did you
40:40 see that? Um who was it that went into
40:42 business with um CBRE, the the big uh
40:47 commercial real estate company? They're
40:49 they're now training people to install
40:51 fiber optics in data centers.
40:54 Was it Amazon? Might have been Amazon.
40:57 or [music] maybe it's Meta. I think it's
41:00 Meta. Meta and CBRE are are training
41:04 people how to be optical cable
41:07 technicians for data centers. [music]
41:17 [music]
41:21 I'll tell you why I don't like Tim Cook.
41:22 Same iPhone every year. Change one or
41:24 two things. Call it a new model. This is
41:26 but this is what I'm saying.
41:29 That was Tim Cook's job. Tim Cook's job
41:32 was maintain what we've got. Don't lose,
41:35 don't blow the lead we have.
41:39 Except the cost of it is when something
41:41 like AI comes along, for whatever
41:44 reason, they chose not to pivot. They
41:46 chose not to embrace it. They chose not
41:48 to
41:50 lean into that.
41:53 And you know, maybe it was a calculated
41:55 bet to say that all this generative AI
41:57 stuff is just going to be interesting to
41:59 engineers and we're a consumer company,
42:03 but I think they could have driven the
42:05 acceleration of AI by getting in early.
42:08 And they still may. I mean, here's the
42:10 thing.
42:12 If you if you're if you're talking
42:14 metaphorical comparisons to the early
42:16 days of the worldwide web,
42:19 the first three years of this channel
42:22 was like 1995, 1996.
42:26 The last 6 months is like 1997. The
42:29 difference between 96 and 97 is that 97
42:33 is when the hype really started with the
42:35 worldwide web.
42:38 So AI has taken a little bit longer to
42:40 sort of buzz up. Like I know it's people
42:43 are talking a lot about it, but it's
42:45 just now starting to get stupid,
42:49 right? Just stupid marketing campaigns
42:51 and stupid just it's it's getting stupid
42:53 now.
42:56 [music]
42:58 So we're very early.
43:01 So here's the thing. Here's the thing
43:02 with Apple. It's possible,
43:05 highly unlikely. It's possible they've
43:08 been brewing up [ __ ] you know, in the
43:11 back 40 for a couple of years and
43:14 they're just going to come out with
43:15 something and it's going to be
43:16 remarkable. I doubt that's the case. I
43:18 think they just chose not to do it. But
43:21 if at some point Apple says, "We're
43:22 going to go all in. We're going to
43:24 become a frontier model company," um,
43:26 and they come out with a model that is
43:29 really good, everyone will forget about
43:31 the fact that they waited. They'll call
43:32 them, "Oh, they were really smart. They
43:34 let everyone else make all the mistakes
43:35 and they came in and figured out how to
43:37 do it."
43:39 I mean, I wouldn't be surprised if Apple
43:42 comes out with some sort of chips to um
43:45 to put in their own data centers to
43:47 train their own things. [music]
43:49 Amow to invest up to another 25 billion
43:51 in anthropic.
43:53 So amazing. Just amazing.
43:59 Personal models. Yeah, I think personal
44:01 models are going to be big. Or maybe
44:03 it'll be a shoe company like like Albert
44:07 Apple shoes. Yeah, apparently they've
44:09 got Apple glasses are still rumored to
44:12 be coming out like augmented glasses.
44:19 Posted the app in the salon and look
44:21 what I made. Okay, cool.
44:23 So, if anyone wants to play with um
44:25 Archetypal's
44:27 app in the salon, go to look at look
44:29 what I made. Let me jump over there.
44:36 [clears throat]
44:43 This is an automated social marketing
44:45 app specifically for authors built with
44:49 Pulsia.ai.
44:51 I've never heard of Pulsia.ai.
45:00 You direct it, the AI writes it. Nothing
45:03 posts without your say. Start free. No,
45:05 no card required.
45:08 Book title, book blurb, genre.
45:13 Notebook, LM audio. Oh, that's cool.
45:17 TJI, your voice. That's cool.
45:24 AI slop spelled backwards.
45:27 That's pretty That's pretty genius.
45:31 [laughter]
45:32 I slop spell backwards.
45:36 That's pretty good. Um,
45:39 one thing I would fix is your your
45:41 button in the top that says create
45:43 account. The color of the font. I can't
45:46 read the button. [clears throat] Make it
45:48 white like this.
45:53 Generate a week of posts. Tell us about
45:55 your book. We'll create seven days of
45:57 platform ready social media content
45:59 tailored to your genre and audience.
46:01 That's super cool.
46:03 Ink Signal. Good name. Ink Signal. Nice.
46:08 Oh, we lost your screen share. Nice.
46:12 Thanks, Apple. Here, case in point.
46:20 Hey Tim, do you think we need the the
46:22 quality control team here at Apple? Nah,
46:26 we're good. Our stuff's good enough. We
46:28 don't need QC anymore.
46:31 [laughter]
46:33 Oh my god. Oh my god. Yeah, right.
46:38 That's fantastic.
46:40 Yeah. No, you're all super good. Super
46:43 good. Yep. There you go.
46:47 Well, you're in. Look what I'm I just
46:49 lost the screen share again. God damn
46:51 it.
46:53 I really should restart my machine
46:55 before every uh
47:01 [gasps]
47:02 before every time I go live.
47:06 You know, this is maddening.
47:12 [whistles]
47:12 At least once a week. I know.
47:18 >> [music]
47:23 [music]
47:28 >> A Dr. J song in Look What I Made.
47:33 Let's go look at a Dr. J song, shall we?
47:36 [clears throat]
47:38 Bones by Untamed Acid House.
47:43 I love that. Dr. J untamed acid house.
47:47 Nice.
47:51 All right, here we go. Here we go. Good
47:53 people. Broken piano. Close mic.
48:06 I still have your sweater in the bottom
48:08 of a drawer. I tell myself tomorrow I'll
48:11 be brave enough to throw it. But
48:13 tomorrow keeps on coming and I [music]
48:15 keep on keeping it. And I lie to
48:18 everyone I [singing] love and tell them
48:20 I don't know it. The elephants they go
48:23 back to the bones of their dead. Stand
48:26 there in [music] the silence doing
48:28 [singing] nothing with their heads bowed
48:30 down. I didn't know. I didn't know. I
48:33 [music] didn't know. Grief had a body of
48:36 its own. They told me there were
48:38 [singing] stages. They told me there's a
48:40 door. [music] They told me one day soon
48:42 I wouldn't feel this anymore. But I keep
48:46 coming back to the bones. To the bones
48:51 like the elephants. I keep coming back
48:55 to the bones. [music]
48:57 To the bones. I don't know how to leave
48:59 you in the ground. I keep coming back. I
49:02 keep coming back. [music] I keep coming
49:05 back. I keep coming back. I keep coming
49:09 back. I keep coming back. [music]
49:12 I keep coming back. I drive past your
49:15 mother's house. And I don't know why I
49:18 go. She hasn't seen me since the
49:20 funeral. She wouldn't want me to, but I
49:22 sit [music] across the street and watch
49:24 her water all your plants. And I whisper
49:26 to the windshield [singing] everything I
49:28 never got to. The crows remember [music]
49:32 faces. I bet they remember [singing]
49:33 you. I bet you fed them something and
49:36 they kept their promise to remember. I
49:39 didn't know. I didn't [music] know. I
49:42 didn't know. Grief was something you
49:45 could grow. They told me [music] there
49:46 were stages. [singing]
49:47 They told me there's a door. They told
49:50 me one day soon I wouldn't feel this
49:52 anymore. But I keep coming back. The
49:56 bones.
49:58 The bones. Like the elephants. Keep
50:02 [music] coming back. The bones.
50:06 The bones. I don't know how to leave you
50:08 in the ground. [music]
50:09 >> I keep coming back. I keep coming back.
50:13 I keep coming back. I keep coming
50:15 [music] back. I keep coming back. I keep
50:19 coming back.
50:21 I keep coming back. I keep coming back.
50:24 [music]
50:24 I thought
50:28 I came.
50:30 >> That's pretty good, Dr. J. It's really
50:32 good. Very, very good. Uh, Producer AI
50:34 is now Google Flow Music. That happened
50:36 a while ago, I think, Brandon, didn't
50:38 it? But did they sell to Google or or is
50:41 Google just using them? They bought it
50:46 and rebranded it. Oh, okay. Google Flow.
50:50 Hey, do you want to come up and talk
50:51 about So, so this past weekend, um Rick
50:54 Macaulay and producer Brandon were vibe
50:56 coding and uh
51:01 they got they got screwed out of out of
51:03 a top five finish by some crappy AI
51:06 judging system.
51:11 >> Tell tell us about your weekend. What
51:12 did you do?
51:14 >> Yeah, so uh we got the gang back
51:16 together. Rick McCaulay, the AI guy, and
51:19 a couple of other team members, we went
51:20 down to actually we got doubly screwed
51:23 because we had aligned on a healthc care
51:27 track. We were going to cure cancer. And
51:29 uh at the last second, while I was on
51:32 the plane on my way to Miami, they're
51:34 like, "Just kidding. We're not doing the
51:35 healthcare track anymore. We're doing
51:36 the enterprise track." And so in true
51:40 hackathon spirit, we pulled up our plan
51:43 B, which was Upskill USA,
51:46 >> and ended up vibe coding in uh nine
51:51 hours a complete portal for enterprise
51:55 companies to figure out how much money
51:57 they're leaving on the table by not
52:00 activating their employees with AI. and
52:03 uh thought we had a pretty good shot,
52:05 but then they decided to also let AI
52:08 pick the top five projects and we did
52:11 not place because we did not uh we did
52:15 not
52:17 put our efforts towards impressing AI
52:19 [laughter]
52:23 >> with a really cool product and a really
52:25 cool demo and um so we could maybe take
52:29 a quick look at that. But yeah, it was a
52:31 lot of I mean, those those hackathons
52:33 are a ton of fun. If you've never done
52:34 one, they're they're just wild because
52:37 you're just in the zone for nine hours.
52:39 And we put
52:41 in months
52:42 >> and build something that's actually
52:45 real. I don't know if you saw I put
52:46 StoryVine through um and uh you know it
52:50 actually does a free audit of their
52:52 website using firecrawl and uh some API
52:55 stuff and it's it's it's a pretty cool
52:57 program but not not worthy for this.
53:01 >> You want to demo something?
53:04 >> Uh sure.
53:07 >> We'll see if my screen share does better
53:09 than yours.
53:10 >> Yeah, really
53:13 crazy.
53:15 So share screen with audio
53:20 upskill USA. So it's
53:22 upskillusa.lovable.app.
53:27 And
53:28 this was actually Rick's original idea
53:30 is he wanted to help upskill and retrain
53:34 employees you to be ready for AI. And so
53:37 we took that and we pivoted towards a
53:40 system that allows
53:43 the AI to understand what you do
53:47 and where your gaps are and what you can
53:50 automate completely away, what you can
53:53 augment with AI and what you have
53:55 authorship over what is uniquely human
53:57 in your role. And so
54:00 >> you built the entire site and you get
54:02 your free AI readiness audit. So, we can
54:04 come here and do something like I don't
54:06 know, give me your favorite uh
54:08 enterprise company, big box company.
54:12 >> Um, let's go with um Salesforce.
54:19 >> So, I [clears throat] just type in the
54:20 website and we got a little lead magnet
54:23 here. So, it asks for your email address
54:24 so we can send you the report and
54:26 capture who's using it.
54:30 And so, it's analyzing Salesforce.com.
54:32 So we not only use firecrawl to
54:34 understand the website but we also use
54:36 an API to look at like the publicly
54:38 available how many employees do you have
54:40 that type of stuff and it figured out
54:44 that uh they're losing $2.8 8 million a
54:48 year by not fully implementing AI. And
54:51 even though they are 75% out of a 100,
54:54 they got a good workforce score. What
54:57 we're losing is sales representatives
55:00 spend an excess amount of time on data
55:02 entry and lead tracking. Customer
55:04 service agents are inundated with
55:06 repetitive inquiries.
55:08 So it does this whole analysis and then
55:10 what
55:11 >> sales sucks. [laughter]
55:14 Their own tool sucks.
55:16 >> Yeah. What you can do is you can log in
55:21 and then you can have your employees
55:23 take an assessment.
55:26 Uh and actually I'll just show you and
55:28 the um
55:31 and it would show you
55:35 let me log back out here.
55:40 It shows you for for as an employee.
55:44 These are the amount of how much time
55:46 you can save by automating invoice data
55:49 entry. Uh but things like vendor
55:52 relationship management should stay in
55:54 completely human.
55:56 So it shows you how much you and then
55:58 you can build your own
56:00 um emulator. So the core problem we're
56:06 trying to solve is everyone's trying to
56:09 do AI from top down and say here's cloud
56:13 code go use it in your workforce.
56:16 >> We're finding that the bottom up
56:18 approach is much more effective. Cisco
56:21 did a pilot program where they had 800
56:24 of their employees just use AI freely
56:28 with governance obviously, but the
56:30 low-level employees that were actually
56:33 on the front lines of the company found
56:35 250 novel use cases for AI that the
56:39 upper management team was completely
56:40 blind to. So, this idea of empowering
56:44 users to automate little parts of their
56:46 mundane day-to-day is a huge unlock. And
56:50 uh I'll play this. This is a two-minute
56:53 demo. And
56:55 we put this together
56:57 and in about 40 minutes.
57:01 And what's really cool about this is
57:03 that Sunno did the
57:06 audio for this. We asked, you know, to
57:08 do spoken word and it did a really good
57:11 job.
57:12 >> Oh wow.
57:18 >> Enterprise AI is at a breaking point.
57:23 Over 70% of AI initiatives fail to scale
57:27 beyond pilot millions are being spent
57:29 with little to show in operational
57:32 impact.
57:35 Not because [music] the technology
57:36 doesn't work,
57:38 but because the approach doesn't.
57:41 AI has been deployed top down.
57:45 The [music] seauite defines the vision.
57:48 It manages the systems.
57:51 But the people who actually do the work,
57:53 they're not part of the solution design.
57:57 The future of AI isn't top down. It's
58:00 employed driven, governed by IT, [music]
58:03 and aligned to the business. At Upskill
58:06 USA, we don't replace your workforce,
58:11 we upgrade it.
58:13 Start with a free audit.
58:17 Let your employees take the assessment,
58:20 then map every role to three outcomes:
58:23 automate, augment, author.
58:26 Because the companies moving fastest
58:28 aren't removing humans, they're
58:30 amplifying them. [music]
58:33 Employees enter the emulation station
58:36 where they break down their work,
58:37 identify inefficiencies, and build AI
58:40 powered workflows tailored to what they
58:42 actually do. Every action is structured,
58:46 every output is measurable, and it
58:49 maintains full governance and control.
58:53 This isn't a future [music] state
58:56 conversation.
58:58 Your employees are already solving
59:00 problems every day
59:03 without AI.
59:05 The question is, how much longer can
59:08 [music] you afford that?
59:10 Upskill your people. Unlock what already
59:13 exists inside [music] your organization
59:17 and turn AI from experimentation
59:23 [music] into execution.
59:32 Beautiful.
59:34 There you go. Thank you. Yeah. Um, the
59:38 uh the first half of that, by the way,
59:39 all the B-roll was Google Vids. We just
59:42 gave the uh VO, we just gave the script
59:45 to Google Vids and it came up with all
59:47 of those uh visuals of the people
59:49 walking around an office and stuff like
59:51 that.
59:53 >> Wow, that's cool.
59:55 >> Cool. So, what's the plan with it now
59:57 that it's built? [clears throat]
59:59 Uh we're Rick's talking about taking it
1:00:02 down to some VCs down in the Miami area
1:00:05 and um obviously we got to fine-tune and
1:00:08 figure out how to pay wall and things
1:00:10 like that, but you know, it's pretty far
1:00:12 along for a 9-h hour app. [laughter]
1:00:16 >> So amazing.
1:00:18 >> It's amazing. Have you had anyone play
1:00:20 with it yet? Kick the tires?
1:00:23 >> Uh not yet because it's only been out in
1:00:25 the world for about 36 hours. Um, but
1:00:29 >> we're uh we're open to uh beta testers
1:00:32 if you know anybody in the salon.
1:00:36 >> Uh I wrote an article about our
1:00:38 experience over the weekend in the in
1:00:40 the community feed today. So
1:00:42 >> Okay, cool. And there's a link to it in
1:00:43 there, right?
1:00:44 >> Yep.
1:00:45 >> Beautiful. Beauty, beauty, beauty.
1:00:48 Well,
1:00:48 >> congratulations.
1:00:50 >> Thank you.
1:00:52 Uh see you on the other side.
1:00:54 >> See you out there.
1:00:56 All right.
1:00:57 I'll kick the tires for you. There you
1:00:59 go. Source Camp. Source Camp going to go
1:01:02 uh figure out what she can do with her
1:01:04 skills. She's [laughter] g She's going
1:01:06 to upskill.
1:01:09 I don't know how you upskill from from
1:01:10 where you are. [laughter] You're pretty
1:01:12 much doing it. Um
1:01:16 yeah, crazy times.
1:01:18 Crazy times. What other thoughts? What
1:01:20 other thoughts? Why am I not more
1:01:22 productive? Because I'm spending four
1:01:23 hours a day doing busy work. Salesforce
1:01:26 is its own oxymoron.
1:01:28 Yeah. It's it's just
1:01:31 I mean what's so impressive about
1:01:33 Salesforce
1:01:35 is also its greatest weakness which is
1:01:38 once you put all your [ __ ] in there the
1:01:41 the the switching cost is essentially
1:01:44 gargantuan. So like no one uninstalls
1:01:47 it.
1:01:49 And so just you've got decade customers,
1:01:52 you know, decadesl long customers, no
1:01:54 matter how bad it is, they're just going
1:01:56 to keep using it because the thought of
1:01:58 changing it is uh Lori K, that's
1:02:01 awesome, Brandon. I don't honestly don't
1:02:03 know anyone who uses their Salesforce
1:02:06 account. We do at Story Vine, but not
1:02:08 because of me, because of other people.
1:02:11 [laughter]
1:02:21 >> [clears throat]
1:02:24 >> Hang on one sec.
1:02:45 Details.
1:03:08 All right. Looks like there was a
1:03:11 a compromise of a third-party developer
1:03:14 library
1:03:17 of Axios.
1:03:23 The security team has rotated code
1:03:25 signing material for all OpenAI MacOss
1:03:28 apps.
1:03:30 Please update any MACOS apps before May
1:03:32 8th.
1:03:35 All right.
1:03:38 >> [clears throat]
1:03:38 >> Kyle, two things. One, you're not
1:03:39 sharing your screen yet. Uh, and two, I
1:03:42 didn't know if that was a share screenw
1:03:44 worthy moment, but also um,
1:03:46 [clears throat] Lovable had a non leak
1:03:49 that was reported as a leak today. So,
1:03:51 you might see that over the next couple
1:03:53 days if it gains any traction.
1:03:55 >> What it was was initially
1:03:58 [clears throat] when they first launched
1:03:59 Lovable,
1:04:01 when you published your project
1:04:03 publicly,
1:04:04 >> Yeah. It published your chat history
1:04:08 with the project just like chat GBT when
1:04:11 you share a chat
1:04:13 with the philosophy of build and public.
1:04:15 What lovable didn't account for was that
1:04:18 vibe coders didn't understand security
1:04:20 and were putting their API keys directly
1:04:22 into the chats instead of the security.
1:04:26 >> So they stopped
1:04:27 >> doing that six months ago. They they
1:04:29 stopped sharing the chats because they
1:04:32 didn't want to expose people. But some
1:04:34 journalist found, you know, history of
1:04:37 this and and wrote a hit piece on them
1:04:39 and now everyone's running with Loveable
1:04:42 got hacked and they're like, "No, we
1:04:43 didn't." So
1:04:45 >> yeah. Yeah. That's wild. That's wild.
1:04:49 There's a new version of Goggout for for
1:04:51 uh uh for OpenClaw,
1:04:56 which if you're not doing OpenClaw, you
1:04:59 won't care. I'll try to scare share my
1:05:02 screen for a bit. I'm just going to look
1:05:03 at at um some X stuff to just see if
1:05:06 there's any cuz it looks like this is
1:05:08 going to be a big week. I would assume I
1:05:10 think when does OpenAI do their their
1:05:12 launches? I think they do them on
1:05:14 Thursdays, right? So maybe tomorrow
1:05:16 we'll hear from them that uh
1:05:24 Mr. Strawberry, you're not a emotionally
1:05:27 prepared for this model. He's been
1:05:28 really hyping the [ __ ] out of this new
1:05:30 open AI thing.
1:05:33 Um, I know a lot of people who have it
1:05:35 but they don't use it.
1:05:39 Oh, Salesforce. Yeah. Fastest AI side
1:05:43 side hustle to 500k a year if you can
1:05:45 actually execute
1:05:47 millions of small businesses, almost
1:05:49 none of them willing to embrace AI right
1:05:51 now. Yeah. Yeah. Yeah.
1:05:55 Did you see open clause founders TED
1:05:57 talk? I have not.
1:06:00 I saw that there was one, but I didn't
1:06:02 see it.
1:06:08 Oh, right. It's 420.
1:06:11 [laughter] Look at the dudes getting
1:06:12 stoned in their little helmets.
1:06:19 [laughter]
1:06:27 Oh, mind your Tik Tok censorship.
1:06:30 Yeah, I'm sure we'll be we'll be uh
1:06:32 given warnings here in a moment. Amazon
1:06:34 announces they will invest up to
1:06:38 $25 billion in anthropic. Fascinating.
1:06:43 Peter Diamandis, we're not building
1:06:45 tools anymore. We're witnessing the
1:06:47 emergence of something between software
1:06:48 and life, between servant and peer.
1:07:10 Yan Lun left Meta because Zucker
1:07:14 Zuckerberg wouldn't bet the company on
1:07:16 Jeppa
1:07:18 joint enabled predictive architecture.
1:07:21 Oh wait, joint embedding
1:07:24 predictive architecture.
1:07:27 Last week his group dropped the first
1:07:29 Jeepa that actually trains end to end
1:07:31 from raw pixels. 15 million parameters
1:07:34 single GPU a few hours. The timing is
1:07:38 not coincidence.
1:07:41 For four years, Meta has been able to
1:07:44 wait. Meta has been
1:07:48 the house that Jeepa built. Lacun
1:07:50 published the original paper in 2022.
1:07:55 The architecture was supposed to be the
1:07:57 escape hatch from LLMs, the path to
1:08:00 robots that actually learned physics
1:08:01 instead of hallucinating about it. Every
1:08:04 version shipped fragile, stopped
1:08:06 gradients, blah blah blah. Meta kept
1:08:08 funding LLMs. Llama shipped. Llama
1:08:11 scaled. Llama got beat by Quinn and Deep
1:08:13 Seek. Zuck spent $14 billion dollars to
1:08:16 buy Scale AI and install Alexander Wang.
1:08:19 That was like the 23-year-old exec that
1:08:22 he put in above Yan Lun. That was
1:08:24 basically like [ __ ] you. Lun's research
1:08:27 kept winning papers and losing product
1:08:29 roadmap. He left. He started AMI labs
1:08:32 and said publicly that LLMs were a dead
1:08:34 end. Now the paper la world model one
1:08:38 regular riser replaces the entire pile
1:08:41 of furistics. I don't know what that
1:08:43 means. Oops. Hang on. I got to plug in
1:08:47 my machine. My machine's about to die.
1:08:52 Got to plug in my machine.
1:09:04 All right, there we go.
1:09:13 Project the latent embeddings in onto
1:09:16 random directions. Run a normality test.
1:09:18 Penalize deviation from gauian. The
1:09:20 model cannot collapse because collapsed
1:09:22 embeddings fail the test by
1:09:25 construction. Hyperparameter search went
1:09:28 from 0 n to the 6 polomial to 0 log n
1:09:32 logarithmic. Six tunable knobs became
1:09:35 one. Well, duh.
1:09:38 Everybody knows that
1:09:42 one regularizer replaces an entire pile
1:09:45 of huristics.
1:09:47 You know like we talk about here we talk
1:09:50 about normalizers and regularizers and
1:09:53 huristics with the latent embeddings
1:09:55 projected into the random directions
1:09:58 which once you run a normality test
1:10:00 against the deviation from Gausian
1:10:03 [laughter] I mean I mean who wouldn't
1:10:05 want to deviate from Gausian right am I
1:10:08 right
1:10:14 all right whatever Yan Lun's up to
1:10:16 something. Apparently, [laughter]
1:10:21 we need paint. [laughter] Cam Ken, what?
1:10:27 Oh, Kyle, that was so five days ago.
1:10:30 [laughter]
1:10:32 Pate has been summoned. Oh my god. What
1:10:35 if everything you see, feel, and
1:10:36 experience is part of an high highly
1:10:38 advanced computer simulation? I'd be
1:10:40 [ __ ] impressed.
1:10:42 That's what I'd be.
1:10:53 Elon Musk. Actually, AI and robotics
1:10:56 will mean that everyone can have a
1:10:58 penthouse if they want.
1:11:01 No,
1:11:04 there's only one penthouse in a
1:11:06 building.
1:11:09 Not everyone can live there.
1:11:11 >> [laughter]
1:11:11 >> Unless you just make really tall stilts
1:11:14 and put a single floor really high in
1:11:16 the air. [laughter]
1:11:18 And I suppose with robots and infinite
1:11:21 abundance you could do that, but no. No.
1:11:26 Okay, whatever. Where my Gen Xers at?
1:11:31 45 to 60.
1:11:36 Zero tolerance for stupidity. There you
1:11:38 go. Speak sarcasm fluently.
1:11:42 That's really funny.
1:11:45 Let's see. Reporting
1:11:47 for duty.
1:11:51 Every company
1:11:54 needs a cranky
1:11:58 gen Xer in the corner office
1:12:07 to tell
1:12:09 the youngsters.
1:12:14 Why?
1:12:16 What
1:12:20 they built
1:12:26 is a pile of And then we're going to
1:12:30 make a a poo
1:12:33 [laughter]
1:12:34 a poo emoji.
1:12:41 >> [laughter]
1:12:44 [gasps]
1:12:45 >> There you go. All right. Beautiful.
1:12:50 [clears throat]
1:12:54 I just deviated into Gausian. I want a
1:12:57 floating [laughter] island. Yeah,
1:12:59 exactly. Everybody with AI and robots,
1:13:01 everybody could have a flo floating
1:13:03 island. [laughter]
1:13:05 Hey, how come there's no oceans? Oh,
1:13:08 everybody wanted an island? Really? And
1:13:11 they all got one? Yeah, it's massive
1:13:14 abundance. It's hyper abundance.
1:13:17 So, is there any beachfront left? No, we
1:13:20 got rid of that. Everyone wanted an
1:13:21 island. So, isn't that just land? Yeah,
1:13:24 pretty much. Where'd the [clears throat]
1:13:26 water go? We buried it. [laughter]
1:13:30 Humans suck.
1:13:33 Charles, cool app. Fairy Tale
1:13:34 Revolution.
1:13:36 It didn't give me posts for X and Blue
1:13:38 Sky, but did for Instagram only when I
1:13:40 was choosing. I chose X and Blue Sky,
1:13:42 not Instagram.
1:13:44 Maybe just user error. I don't know.
1:13:46 Claire. Okay, cool. Claire's given
1:13:48 feedback.
1:14:06 plugins.
1:14:09 So, I don't see a lot of buzz about new
1:14:12 models coming other than Mr. Strawberry.
1:14:17 And generally, we'll get you you'll get
1:14:20 some noise.
1:14:38 Tell me.
1:15:01 Oh, you want to hop up and talk about
1:15:03 that, Brandon?
1:15:07 >> Hello again.
1:15:10 >> Uh, yeah. So, next month for mastermind,
1:15:13 I'm putting together a four-week
1:15:15 prototyping sprint for Lovable. So,
1:15:18 we're going to go over in the four
1:15:20 weeks. Basically, if you've never been
1:15:23 down the rabbit hole of vibe coding,
1:15:25 what vibe coding is and what Lovable is,
1:15:28 and then how to even think about how to
1:15:31 come up with an idea and how to use
1:15:33 Lovable to build it and launch it,
1:15:36 stress test it, and it's going to be a
1:15:38 really cool four-week program, and I'm
1:15:41 excited about it.
1:15:42 >> That's super cool. When does it start?
1:15:44 >> Uh May uh right May 6.
1:15:48 >> Okay. So, if you're in the master if
1:15:50 you're not in the mastermind, go to
1:15:51 community.thesalon.ai.
1:15:53 Look for the mastermind. There's lots of
1:15:55 information in there about it. Join it
1:15:57 if you want to be part of that. Um, are
1:15:59 you going to are you going to go into
1:16:00 one of the things that flumxes me a bit
1:16:03 about vibe coding because I just haven't
1:16:05 done it enough is once I'm past the
1:16:08 initial build, like going back and
1:16:10 fixing stuff and tweaking it, you seem
1:16:12 to be pretty good at that. How do you do
1:16:14 that without screwing things up? Is that
1:16:16 what are you gonna cover that session?
1:16:19 >> Yeah. Yeah, we'll we'll definitely talk
1:16:21 about that. I mean it it's more of an
1:16:23 art than a science, but
1:16:26 >> uh yes.
1:16:27 >> Okay. All right. I mean like even that's
1:16:29 good to hear. That's really cool.
1:16:34 >> Yeah.
1:16:34 >> All right, Groovy.
1:16:36 >> That's exciting. Okay, cool. And and
1:16:38 information I assume is in the uh in the
1:16:42 in the mastermind.
1:16:43 >> Yes. Andy just posted it a little bit
1:16:45 ago which was reminded me to tell you
1:16:47 about it.
1:16:48 >> Oh, cool. Oh, good. Awesome. Beautiful.
1:16:54 My first sun song was waiting on
1:16:56 strawberry ah memories. Yeah. Remember
1:16:58 when Mr. Strawberry had all the all the
1:17:01 the [clears throat] rumors about what
1:17:02 was coming and then he sort of lost his
1:17:04 credibility
1:17:07 and now he still says all this stuff and
1:17:09 I'm like if only I believed you.
1:17:10 [laughter]
1:17:12 I just don't.
1:17:29 All right, hang on. [clears throat]
1:17:32 This is Greg Brockman talking about
1:17:35 the new OpenAI model.
1:17:45 So I think of spud as a new base as a
1:17:48 new pre-train and that we have had this
1:17:52 uh I'd say it's like we have maybe two
1:17:54 years worth of research that is coming
1:17:57 to fruition in this model. It's going to
1:17:59 be very exciting and I think that the
1:18:02 way that the world will experience it is
1:18:04 just improved capabilities and that for
1:18:07 me it's never about any one release
1:18:09 because as soon as we have this one
1:18:11 release it'll be an early version of
1:18:13 what we have coming. We'll do much more
1:18:15 of each of these steps of the
1:18:16 improvement process. And so I think that
1:18:19 where we're going is almost just we have
1:18:21 this engine of progress that just moved
1:18:23 faster and faster and that spud is just
1:18:26 one step. So, I think of Spud.
1:18:28 >> All right. So, Spud's their new model.
1:18:29 Horrible name.
1:18:33 All right.
1:18:39 Deep Cam Live. You can swap your face
1:18:41 live on camera in real time. Okay.
1:18:45 Wait. Some of you don't think. Okay.
1:18:47 That's nothing. That's nothing. Elon
1:18:50 Musk.
1:18:52 Sam Alman.
1:19:01 Kimi 2.6 Six
1:19:06 video hero sections. Cinematic aesthetic
1:19:09 autocomposited
1:19:21 [music]
1:19:26 [music]
1:19:34 >> [music]
1:19:40 >> I don't know what that is. I haven't
1:19:42 played with it. Has anyone else played
1:19:43 with that? Kimi
1:19:51 [whistles]
1:19:57 >> [gasps]
1:20:01 [clears throat]
1:20:02 >> UBI works.
1:20:14 >> [laughter]
1:20:24 >> Wait,
1:20:25 look at this. They have I didn't know
1:20:28 someone got
1:20:34 someone [laughter] someone got into my
1:20:36 my uh my CrossFit training. Somebody
1:20:40 recorded me in my CrossFit training. Who
1:20:43 did that? Come on, you guys.
1:20:51 [laughter]
1:20:57 I'm 37 years old. What's the best life
1:21:00 advice you can offer me? Go sit on a
1:21:03 bench.
1:21:04 [laughter]
1:21:05 Just wait it out.
1:21:09 It It'll show up eventually.
1:21:12 >> [laughter]
1:21:17 >> Be honest. Do you talk to yourself? Be
1:21:19 honest. Do you not?
1:21:22 [clears throat]
1:21:24 Wait. [laughter]
1:21:33 Oh, man. All right. It's 9:30. Listen, I
1:21:36 have to go because I have to get up
1:21:38 bright and early. I have a 7:30 flight
1:21:40 tomorrow morning.
1:21:42 I'm mostly packed. [gasps]
1:21:45 I'm a little sleepy,
1:21:48 so I'll eventually just take a nap
1:21:50 tonight.
1:21:52 Um,
1:21:53 so listen, you're on your own this week,
1:21:56 so a couple of things. Um,
1:22:09 don't take any of this [ __ ] too
1:22:11 seriously right now. We don't know
1:22:13 what's coming. We don't know what's
1:22:15 coming. We don't know what's here.
1:22:17 I would say just play.
1:22:20 Figure out the [ __ ] that that brings you
1:22:22 joy this week and just go do that.
1:22:25 You want to vibe code? Go vibe code. You
1:22:28 want to go play in the dirt, go play in
1:22:29 the dirt.
1:22:31 I think that's the that's the thing
1:22:33 until until something comes out where
1:22:35 it's clear that's like, oh, we got to
1:22:36 deal with this. I don't think we're
1:22:38 there yet. I think we're
1:22:41 I think we're in a bit of a holding
1:22:42 pattern.
1:22:45 And if and when something like Mythos
1:22:47 comes out that really does change
1:22:49 things, then we can start to talk about
1:22:50 what's it mean. But like, I don't even
1:22:52 know what it [ __ ] means to have these
1:22:54 autonomous AI agents yet.
1:22:57 It's not like I don't get them. I get
1:22:59 what they are. What I don't get is
1:23:03 why should we give a [ __ ]
1:23:07 and I know it's in there. I just don't
1:23:09 know what it is yet.
1:23:12 And I'll figure it out. Or I'll figure
1:23:15 out that it wasn't in there.
1:23:21 I don't think so, Brandon. I
1:23:22 [clears throat] I looked at scheduling.
1:23:24 I don't think that's that's actually
1:23:26 going to be the case.
1:23:28 Um,
1:23:30 so what I will do is this. If I have an
1:23:32 opportunity during the day to go live
1:23:35 when I'm in between [ __ ] I'm doing, um,
1:23:40 I'll just do that. The thing that would
1:23:41 precipitate that if is if some some big
1:23:44 model comes out that's actually really
1:23:46 interesting and good, I'll probably go
1:23:48 live and talk about it if I've got the
1:23:50 time. But I'm not going to have a ton of
1:23:52 time this week and probably not next
1:23:55 week. So, so the next two weeks are
1:23:57 going to be a little thin, but I'll uh
1:23:59 I'll put some videos out and I may go
1:24:01 live sporadically here and there just on
1:24:04 TikTok. Okay.
1:24:06 We'll be running a muck without you.
1:24:08 Have a good week, Kyle. Thank you very
1:24:09 much much. Denver is the worst. I agree.
1:24:13 [laughter]
1:24:15 I mean, it's lovely and all. Um but
1:24:19 yeah. All right. Cool. Peace out
1:24:22 everybody. Um, have yourself a fantastic
1:24:25 week and I will be in touch. Um, jump
1:24:28 into the AI salon. So, oh,
1:24:31 important announcement. You may know
1:24:33 this, but we've got some folks here. Um,
1:24:38 when I come back from this twoe gauntlet
1:24:40 of of speeches and meetings,
1:24:44 the AI salon is moving inside or the AI
1:24:47 learning lab is moving inside the AI
1:24:48 salon. So, that's going to happen the
1:24:51 first week of May when we come back. So,
1:24:53 I'm really excited about that. So, get
1:24:55 ready. Bring all your friends in. Let's
1:24:57 get some activity going in there. I want
1:24:59 to get the I want to make the inside of
1:25:00 the salon a vibrant um you know,
1:25:03 bustling place again. We've been
1:25:05 bustling out here on the edges. Time to
1:25:07 bring it all inside. Time to concentrate
1:25:09 this energy um and get people really
1:25:12 excited about it. So, I'm super excited
1:25:13 about that. So, that'll happen upon my
1:25:16 return. Lots of lots of fun stuff
1:25:17 coming. All right. Cool. [snorts] Peace
1:25:20 out. Have a beautiful evening and rest
1:25:23 of the week. And I'll uh I'll post in
1:25:25 the salon and let you know what's
1:25:26 happening, what's going down. All right.
1:25:28 Bye, everybody.