AI Learning Lab LIVE - 4/3/26

Description

Chapters

Transcript

0:19 Woohoo!
0:24 Woohoo!
0:39 Woohoo!
1:38 Good evening good people,
1:41 happy people of the internet.
1:45 Irregulars one and all mods
1:49 producer Bandon his guests the tidbits.
2:39 Woohoo!
2:49 You're feeling nervous, aren't you, boy?
2:53 With a gentle voice
2:56 and impeccable style.
3:01 Don't ever let them steal your joy
3:05 and your gentle ways to keep them from
3:08 running wild.
3:12 They can kick dirt in your face, dress
3:14 you down, and tell you that your place
3:16 is in the middle when they hate the way
3:19 you shine.
3:25 I see you tugging on your shirt,
3:29 trying to hide inside of it. Hide how
3:32 much it hurts.
3:35 Let them laugh while they care.
3:40 Let them spin.
3:43 Let them scatter in the wind.
3:48 I've been to the movies. I've seen how
3:54 the jokes
3:56 are.
4:12 You get discouraged, don't you, girl?
4:16 It's your brother's world for a while
4:19 longer.
4:22 You got to dance with the devil on a
4:24 river to beat the stream.
4:27 Call it living a dream.
4:30 Call it kicking the ladder.
4:34 They come to kick dirt in your face, to
4:37 call you weak, and then displace you
4:39 after carrying your baby on your back
4:41 across the desert.
4:46 I see your eyes behind your hair
4:49 and you're looking tired,
4:52 but you don't look scared.
4:55 Let them laugh while they care.
5:00 Let them spin.
5:03 little scatter in the wind.
5:08 I've been to the movies. I've seen how
5:11 dance
5:14 and the jokes
5:16 are there.
5:32 Oh, good people. Good people. What's
5:35 happening? What's going down? What's
5:36 shaking? Got the champ, the singing dog,
5:40 doing his thing.
6:21 Come
7:04 Um, do me a favor. Comment on the
7:06 YouTube
7:08 just so I can make sure my window's
7:09 live.
7:21 Thank you, Silver Fox.
7:39 Oh,
8:07 hey.
8:32 There's been something, baby, I've been
8:35 trying to today
8:38 for an age and it seems I don't know how
8:45 the past and the future now surrounding
8:47 me.
8:51 Surrender to whatever cheap thrill can
8:53 be found.
8:56 There's been little trouble
8:59 since you came to my rescue.
9:06 And if you had like all of the rest, I
9:09 would quit you long ago. But I couldn't
9:13 do that.
9:18 Oh, tell me now. A women never went to
9:25 make a man crazy, make him cold as hell.
9:32 I don't want that. You wish me well,
9:36 but a spotty try. Still going to have to
9:40 find
9:42 my own way through.
9:54 We had a wild um AI office hours today.
9:58 AI salon office hours. It got downright
10:02 spiritual at some points and I was a
10:05 little testy and cranky. That's always
10:07 fun.
10:09 You guys know me at my crankiest.
10:22 I'm I'm just I'm coming to realize that
10:24 I'm I'm tired of of just people's crap.
10:28 People are just sending me crap. They're
10:30 like, "Read my crap." I'm like, "I don't
10:31 want to read your crap. I don't read
10:33 stuff I want to read, much less your
10:35 crap."
10:40 What are you going to do? What are you
10:42 gonna do? Um,
10:46 it's content bombing. Content dumping.
10:50 Don't dump your content on me.
11:04 on
11:10 down.
11:27 Happy Friday night, date night.
11:39 I saw on X. Hang on a sec.
11:43 I saw on X that someone said that the
11:47 GPT6
11:50 image model is out. I can't imagine that
11:53 that's true.
12:01 Create image
12:05 standard extended thinking
12:11 clouds camcorder
12:14 post rain sunset. I don't think this is
12:17 the new the new model. Is it
12:35 I just used it. Pretty impressive.
12:38 I don't know if it was six, but it was
12:39 good. Huh.
12:45 Oh, they're copying X.
12:54 There's like a grid that pulses.
13:20 Can it do 16 by9 yet? Let's see if it
13:23 can do 16 by9. Make that a Oh, wait. Do
13:26 you want me to share my screen? I could
13:27 do that. I can share my screen. I just
13:30 made a horrible picture of myself, but
13:32 I'll do it for you. I'll do it. I'll
13:34 show you. I'll show you the horrific
13:37 mess it just made of my face and neck.
13:40 Look at that neck. Good lord. Like Rosie
13:43 O'Donnell on a bad day. All right.
13:46 Um,
13:48 that was kind of rude. Let's see. Um,
13:52 make that a 16 by9
13:56 wide
13:57 emage
14:01 for 15 minutes.
14:05 We can't see your screen, Kyle. You can
14:07 now, right? You should be able to now.
14:10 Let's see if we get a 16 by9 image here.
14:47 I don't think that's 16 by9.
14:51 I do not believe that are 16 by9
14:55 there, Bucky. Let's go say I don't think
14:59 so.
15:01 Nope. That is not 16 by9.
15:05 That is a large neck.
15:12 Oh well. Well, that was fun while it
15:13 lasted. We could do something else. What
15:16 do you want to make an image of? Um,
15:25 let's try something here.
15:30 Yes. Don't ask again. Don't ask again.
15:36 I'd like you to make me an image
15:39 that represents the five stages of the
15:42 great repurpose.
15:47 Just leave it there. Let's see what it
15:48 comes up with. You got one. What? What
15:50 did you get, Chef Kelly?
15:55 Your face is 16 by9 in that image. Shut
15:58 up. Shut Stop making fun of my face.
16:04 I can see clearly now the rain is gone.
16:09 Ask the bot to make themselves in human
16:12 form. That's in my musical.
16:18 I already wrote that. You got 16 by9?
16:21 Are you sure it's 16 by9? Come on. Don't
16:25 wait. Hang on. All right. Let me use
16:28 actual proper coding with a colon. Make
16:31 a 16 col 9 image
16:39 image
16:41 um
16:43 with the actual
16:47 uh
16:51 phases of the great repurpose.
16:58 We'll see.
17:41 That's not 16 by9. It's not 16 by9. It's
17:46 close.
17:48 might be 16 by 10.
17:55 No, it's not.
17:58 Redesign, repurpose, restore,
18:01 regenerate, reimagine.
18:03 Those are not the five phases.
18:06 Good lord. Good lord. Good people.
18:13 Um,
18:14 so apparently Chat GPT6 is coming out
18:17 next week.
18:23 What is also possible
18:26 is that it's a multimodal model
18:31 that also generates video. That would be
18:33 interesting. There's a single model
19:06 What
19:07 are The five stages of the great
19:14 repurpose.
19:17 What are they?
19:23 Post it on a regulars. Mine is exactly
19:26 1920 x 1080. I'm wondering if I just
19:28 don't have access to the new the new
19:30 thingy there. Chef Kelly, where'd you
19:33 post it? Regulars.
19:35 Uh,
19:47 look at that. That is a 16 by9 indeed.
19:52 Huh. All right.
19:55 Well,
19:57 update. No show tonight.
20:00 That was last night. I sorry I couldn't
20:02 come last night. I I was at a CEO dinner
20:04 that I was actually speaking at and I
20:07 spoke ninth out of 12 people. 12 people
20:09 spoke last night. So, I was number nine.
20:11 So, it it just went too late.
20:16 But, welcome, welcome, welcome, welcome
20:19 everybody.
20:22 Um,
20:26 let's see. What are we going to talk
20:27 about tonight? Um, no more content
20:31 dumping.
20:35 We all need to start practicing
20:41 telling people what we care about,
20:45 why we're working on the stuff we're
20:46 working on,
20:48 connecting with people individually,
20:50 connecting in small teams.
20:56 You're like, "But why, Cal?" Hey, Cal.
20:58 Cal, why?
21:05 Because no one's going to care about
21:06 your [ __ ] Everyone's making too much
21:09 [ __ ]
21:13 But if you make [ __ ] that's actually
21:14 important to you and you talk about why
21:17 it's important or what inspired it or
21:20 how it makes you feel or or just
21:22 anything that is personal,
21:25 then people will give a [ __ ]
21:29 But we're leading with the output
21:30 because the output used to be hard.
21:36 So what happens when the output is easy?
21:39 You make more outputs and we keep
21:42 treating the outputs like we used to
21:46 where they were valuable. They're not
21:48 anymore.
21:50 But I made a really nice picture.
21:52 Doesn't matter.
21:56 You can make some really good research.
21:57 You can make a really good report. You
21:58 can make a really good app. Doesn't
22:00 matter. People are just going to copy
22:01 them.
22:07 What they're going to buy is you.
23:25 What would you like to discuss tonight?
23:27 Good folks.
23:30 We all had to grow up so quickly. too
23:32 quickly.
23:45 Hey, I have a question for you. So,
23:46 listen. Here's here's the thing we're
23:48 doing. April 30th is the one year or the
23:51 three-year anniversary of me going live
23:53 on on Tik Tok. And then a year and
23:56 change ago, Tik Tok was going to go
23:58 away. And so, we started streaming on
23:59 YouTube.
24:01 One of the things we're going to do is
24:02 we're going to move the stream inside
24:04 the the salon. It'll we'll still have
24:06 all the same [ __ ] we have. It's just
24:08 going to be a slightly different way to
24:10 get to it. Um, and if this is new news
24:12 to you, I announced this on Monday and
24:14 it's we're going to take the month to do
24:16 testing and just make sure that it's a
24:17 good experience for everyone. But
24:20 there's an opportunity
24:22 to um
24:25 to shift the format if we want, right?
24:28 You know, we start out with me singing
24:30 with the guitar and champion. We can
24:32 keep that. We could lose that. Um, I
24:35 don't have a ton of guests up talking.
24:36 Maybe we have one night where we talk to
24:38 people. That could be fun.
24:42 There could be more structure like
24:44 Tuesday tool nights where we just pick a
24:46 tool and go play with it.
24:49 I am not a fan of structure which if you
24:52 are an irregular here you know but at
24:55 the same time how do we up the game so
24:58 if anyone has any ideas
25:04 are you saying embrace the jank now we
25:06 are the jank
25:08 we are the jank
25:26 I had a uh
25:29 a large burrito for dinner. Large
25:31 burrito. I'm a little uh I'm a little
25:33 sluggish.
25:36 All
25:47 right.
25:48 All right. If If no one's got any ideas,
25:53 I think I'm going to go back and play
25:54 with Claude Code.
25:57 So, the other night, Wednesday night, I
26:00 was working on this thing in Claude
26:03 Code.
26:05 Um,
26:10 I asked I asked Chad GPT what I could
26:15 make that
26:18 was not was not something I would
26:21 normally make. And it said, "Oh, you
26:23 could make a simulation." And then it
26:25 said you could make a simulation of of
26:27 people's identity being disconnected
26:30 from their tasks or they're you know
26:32 losing they're losing their jobs or um
26:36 or having their their jobs changed
26:38 fundamentally so they don't recognize
26:39 them anymore and what's going to happen
26:41 there. And so we did this and it worked.
26:44 And so if I hit play here
26:48 or reload, let's see. How do I
26:53 Oh, wait. Maybe I need to What do I need
26:55 to do here? Maybe
26:58 it didn't look like that reloaded.
27:00 Um,
27:03 let's go into plan mode here. Let's
27:05 let's say um
27:07 um I want to represent the people with
27:12 little
27:14 people
27:16 designs.
27:19 I also want there to be some people laid
27:24 off,
27:26 some people keep their jobs,
27:32 some people
27:34 keep their jobs,
27:37 but AI
27:40 radically transforms
27:44 what they do
27:50 then some people are out of the work
27:55 force,
27:58 some check back in.
28:02 Uh some
28:05 leave knowledge work all
28:09 together
28:12 and go
28:15 grow blueberries.
28:26 I want you to have
28:30 distinct areas.
28:34 Um,
28:41 let's see. for the people to physically
28:46 go.
28:48 I want there to be
28:52 a lot more
28:55 variety in what they do.
28:59 And I don't want
29:03 everyone to repurpose
29:07 entirely.
29:10 I want the math to be based on research
29:16 of what happens when
29:20 people get disconnected.
29:25 Um, let me know your thoughts. Okay,
29:29 let's see. I I blew through all my
29:31 tokens the last time we did this and it
29:34 looked like it was starting to do
29:36 something interesting, but we ran out of
29:38 tokens. So, I'm just going to dump some
29:40 of my brain thoughts in there.
29:52 Danielle, I used Chat GPT to help me fry
29:55 taco shells correctly tonight for
29:57 dinner. They turned out great. That's
29:59 good.
30:02 That's awesome. Danielle,
30:04 what do you use What do you use chat GPT
30:06 for? Uh, taco shells. I to well taco
30:10 shells, taco bowls. We've done we've
30:12 done both the shells and the bowls. I I
30:14 homemade uh the bowls really fascinate.
30:17 I we we've I tell you what, we went
30:20 through some coffee cups. Yep. Yep.
30:26 Great use case.
30:29 It's awesome.
30:31 Okay, it's looking at the uh
30:38 All right. Shall I proceed? Okay. Stick
30:40 figure people with posture reflecting
30:43 their emotional state. I like that. Five
30:45 displacement categories. Laidoff,
30:46 transformed, unchanged, unchanged,
30:50 voluntary, acts, exit, already out.
30:56 Three paths. Hold fast, repurpose, or
30:58 opted out.
31:00 research grounded math Jehovah's latent
31:03 deprivation wars vitamin model STD SDT I
31:08 don't know what SDT standard deviation
31:10 theory maybe I don't know who knows more
31:13 action variety including permanent exits
31:16 oscillation decline blueberry farming
31:19 better zone spread the visual shall I
31:21 proceed with these yes
31:24 all right here we go people
31:27 we're coding
31:31 We're coding. We're coding. We're
31:33 coding. Coding.
31:37 All right.
31:40 Smooshing.
31:44 Let me start with the foundation.
31:46 Updated types with displacement
31:48 categories. Okay. That's off doing its
31:50 thing.
31:53 Any questions? As we keep being told, AI
31:55 is a tool. You can use it any way you
31:58 like.
32:00 Yeah.
32:02 I mean
32:05 I mean AI is a tool like the internet is
32:07 a tool, right? Technically the internet
32:09 is just a protocol. you can use it any
32:11 way you like. And we turned that into
32:13 everything from businesses to customer
32:16 service sites to e-commerce to right
32:18 like it's just this enabling
32:25 ooze that you can do anything with. AI
32:27 is kind of like that.
32:30 And AI as as we start to get world
32:33 models like it's not just going to be
32:34 about entering text and coding. It's
32:36 going to be it's going to be quite
32:37 different. Side hustle Mimi's in the
32:39 house. She's late.
32:42 She's late. I get keep being told to
32:44 chat GPT, but she's my first girlfriend.
32:47 I know. I know. Well, here's the thing.
32:52 Anthropic today just just said that um
32:55 you can they're actually preventing you
32:58 from using OpenClaw with Anthropic right
33:00 now with your with your OOTH. Um which I
33:05 think is kind of dumb. So, I have a
33:07 feeling Chat GPT is going to come out
33:09 with their new their new thing and then
33:12 um
33:13 their new model and then they'll
33:16 probably take the lead again and then at
33:18 some point Anthropic will take the lead
33:19 again and then they'll take it back and
33:21 then at some point Google's going to
33:22 come in there.
33:24 So, you know, we got a bunch of that
33:26 stuff going on.
33:31 Um
33:33 let's see. created a file, right?
33:37 So, let's see what this did.
33:41 Can you look at these? Oh, yeah. Look at
33:43 that.
33:45 So, green means it's all brand new code,
33:48 right? Shoot. Oh, show the full diff.
33:51 Created a file. Click on the file. These
33:54 are cool. You can go look at code if you
33:58 want to geek out. You want to be a
33:59 little nerd. You want to be a dang nerd?
34:04 We got Mara, Ellie, Saurin, Inis, Kai,
34:08 Dev or Devi, Leon, Asha, Felix, Yara,
34:13 Nico, Cleo, and a dog named Cleo.
34:17 We have Cleo and Theo in here.
34:28 Now, the big one, the expanded actions
34:31 system.
34:38 Ask permissions. Auto accept edits.
34:42 Let's do that one. Okay. Go.
35:07 While that's doing that, let's run over
35:09 to not now. Let's run over to the X
35:16 and see if there's any interesting
35:18 tidbits.
35:23 Grock is constantly being updated, so
35:25 there's a good chance what didn't work
35:26 for you even a few days ago
35:29 might work now. AI is replacing people
35:32 who refuse to use AI, learn agents,
35:35 code,
35:37 codegen, and automation. Harsh but true.
35:40 AI is not going to save us.
35:45 This the grand unification of AI and
35:48 crypto is about to happen.
35:50 It's now obvious that AI agents are
35:53 going to need money. It's already
35:54 happening.
35:57 It's that William Gibson quote. The
36:00 future is already here. It's just not
36:01 distributed yet. This is what I felt the
36:04 week chat GPT came out. It's like, oh
36:06 [ __ ] the world just changed, but not
36:08 that many people know it yet.
36:11 At some point, not using AI will feel
36:14 like choosing to be cognitively
36:15 impaired. I think that's already the
36:17 case. While Sunno is generating 7
36:19 million songs a day, some of us are
36:21 still craft crafting music the old way,
36:24 composing it by hand and recording it
36:26 with the best musicians to create unique
36:28 soundtracks one at a time. No eye slop
36:30 ever. Okay. Um,
36:34 if
36:37 you can do that, you should at least
36:38 understand what the tool is in my
36:40 opinion. Become a generalist.
36:42 Specialization makes you efficient.
36:44 Generalization makes you dangerous.
36:46 what it actually means. Learn across
36:48 domains. Math, physics, software,
36:50 economics, biology. Patterns repeat
36:52 across fields. Connect ideas. Innovation
36:55 happens at the intersection, not inside
36:57 silos. Adapt fast. See systems. Build
37:00 end to end. From idea to design to
37:02 implementation to delivery.
37:04 Cool.
37:15 It happened. Max Studio is here. Gemma 4
37:19 31 billion installed. Now chatting with
37:21 my open cloth for0 in tokens. Let's
37:25 listen to her. Let's see what she's
37:26 talking about. See if this is garbage.
37:30 Is this garbage or is this useful?
37:34 Let's find out.
37:35 >> Then it actually took me a little bit to
37:37 get it connected. Today's the day I have
37:40 my Mac Studio. I just unboxed it this
37:43 morning. Immediately downloaded a local
37:45 model. The model that I think came out
37:47 yesterday, uh, Gemma 4, the 31 billion
37:51 parameter. Uh, I think that's the bigger
37:53 one. Immediately downloaded it. I've got
37:55 enough RAM and space on this new bad boy
37:57 to actually run it. I immediately uh
38:00 started chatting with it. It actually
38:01 downloaded pretty fast, my Starlink for
38:03 the win. Uh, I was able to start
38:05 chatting with it. And then it actually
38:07 took me a little bit to get it connected
38:08 to my open claw. I am now chatting with
38:11 one of my open claws on um on my local
38:14 model and so that is for zero dollars.
38:17 I've got zero dollars except for
38:19 spending a bajillion dollars on
38:20 computers, but uh to be real with
38:23 everyone, I have spent uh approaching 6K
38:26 on token expense with Anthropic. So,
38:29 there's ways I could have done that
38:31 cheaper, but I do want to just tell
38:33 everyone like I'm burning a lot of
38:35 anthropic tokens over the past few
38:37 months because it's almost like three
38:38 months now running all this stuff and
38:41 it's expensive. So, I don't need to be
38:43 like pinging opus to check on random
38:46 stuff. There's a lot of things I'm doing
38:48 with the curriculums um and actually um
38:51 make and actually like doing stuff for
38:52 the home school that absolutely doesn't
38:54 need the level of like clawed code or
38:57 opus that I'm like sometimes pinging. So
39:00 I'm going to run um Gemma for a lot of
39:03 my chat. So on my chat interface with my
39:05 agents I'm I'm gonna migrate over to
39:08 using
39:08 >> I was just watching Burman's latest
39:10 video about Gemma. Yeah. So Gemma 4 came
39:12 out yesterday, right? yesterday. I think
39:15 yesterday
39:17 and that's Google's open-source uh
39:20 open-source Gemini model basically and
39:24 it's small enough to run on laptops and
39:26 phones.
39:28 Gemma, I'm coming up with a schema of
39:30 when my agents will ping Gemma for chat
39:34 and then like p be much more surgical
39:37 and use um opus for like planning
39:41 projects or use claude code uh when I
39:44 need to actually do something really
39:45 deep programmatically. But, uh, I've got
39:49 enough, um, room on this guy to try
39:52 doing the Quen coding and, uh, try to
39:57 kind of make a great coding agent using
40:00 Quen coding. So, I'm also going to play
40:03 with that. I just haven't gotten there
40:04 because it's a much bigger model to
40:05 download. So, I haven't uh, done that
40:08 yet, but I'm going to do that. So, I'm
40:09 going to try to report back on my
40:12 savings once I figure this out. But I'm
40:14 going to migrate all of my daily
40:16 chatting and my homeschool logging and
40:18 everything to local models. So that
40:20 that's just uh not only is it completely
40:22 private and local to my network, but
40:24 also it will start costing me just the
40:27 cost of electricity uh which will be
40:29 just like a massive savings. So I think
40:30 I
40:31 >> I downloaded Gemma 34 billion yesterday
40:34 to run as a local model. Um Gemma 3 or
40:39 Gemma 4. Gemma 4 came out Cam. So, if
40:42 you downloaded three, there might be a
40:44 newer version.
40:47 >> Bring my token expense down like like
40:49 80%.
40:51 >> All right. Um,
40:53 wait, let's go back to Cloud Code. See
40:55 if it's done. Oh, there's our little
40:57 people thinking. Okay, it's still
40:59 programming.
41:24 They're little stick figures.
41:34 Idris Robbie.
41:39 Let's see. Um,
41:43 let me go here. Go plan mode. Oh, wait.
41:46 I'm not sharing this. Hang on. Hold.
41:49 People, hold, please.
42:00 Is this a 3D model? No. And this is
42:04 nowhere close to Pixar grade.
42:18 We're looking at your X screen. Yeah, I
42:20 know. Um,
42:23 so we should now be able to see. Yeah,
42:25 now you're looking at at this. Okay,
42:28 let's let's go. Um,
42:32 right now people are either
42:42 green or gray.
42:51 But you have
42:56 more types than that. Also, people who
43:01 are
43:03 dejected
43:04 should maybe
43:07 hang out.
43:11 Let's see. Wait. Also people
43:17 should go to specific
43:37 All
44:05 right.
44:10 Cam. Yes, I get to need to get the newer
44:12 version once I feel comfortable with
44:13 this. Very cool. I downloaded Gemma 3.
44:19 All right. Fantastic.
44:28 Pre-upture, everyone's near zero
44:30 alignment, so they're all gray.
44:31 Postrupture,
44:32 drift isn't strong enough. We need color
44:35 opted out. people distinctly
44:37 movement isn't mood driven enough.
44:39 Vertical bunching zones aren't carrying
44:41 the full canvas height. Let me fix all
44:43 three. All right, so that's doing that.
44:46 So, let's go back to this.
44:52 Okay.
45:02 Speak to create images.
45:06 Using voice for imagine is great for
45:08 young kids who can talk and have amazing
45:11 imagination.
45:14 All right. How do you feel about
45:15 artificial intelligence?
45:18 Fine.
45:48 That's a good idea for the microphone.
45:49 Hang it off your glasses. All right,
45:51 good. I learned something from an
45:53 influencer tonight. You can have the
45:55 best AI tools on the planet, the best
45:56 systems, best automations, but if your
45:58 body is running on cortisol, dead food,
46:01 and blue light at midnight, you're just
46:03 building
46:05 faster on a broken foundation. Okay,
46:07 fine.
46:09 I'm a little pasty. Stop yelling at me.
46:13 Peter Steinberger, wake up and my
46:16 mention are full of these. What? Cancel
46:18 my recurring bill. You can still use
46:21 Claude Max.
46:24 Both me and Dave Morren tried to talk
46:26 sense into anthropic. Best we managed
46:28 was delaying this for a week. Funny how
46:31 timings match match up. First they copy
46:33 some popular features into their closed
46:35 harness. Then they lock out open source.
46:38 So Peter Steinberger is the creator of
46:40 OpenClaw and Anthropic just locked out
46:43 OpenClaw. So fascinating
46:47 fascinating. Um I'm genuinely scared
46:51 about what's coming next.
46:59 Claude mythos
47:07 70% think AI will lead to fewer jobs
47:15 and quantum com if you guys I don't know
47:18 if you guys have fully internalized that
47:21 America is about to simultaneously
47:23 invent super intelligence and monopolize
47:26 outer space and then Mark Andre said,
47:28 "And quantum computing is going to work.
47:31 It's all happening."
47:33 Shit's shit's happening in these labs
47:35 that we don't know about,
47:39 but there's too many people that saying
47:41 too many weird things for it not to be
47:43 happening. Sam Alman says, "AGI soon is
47:46 too vague for what's really coming.
48:14 No, it doesn't. It doesn't, Sam. People
48:18 still don't get that. It does the
48:19 thinking. What does that mean? It does
48:22 the thinking. It does the doing. It does
48:24 the execution.
48:26 And it it it's not going to need humans
48:27 in the loop. Couldn't
48:30 couldn't hear it on Oh, that's because
48:33 I'm on YouTube. Sorry about that. That's
48:35 because I was sharing it like a
48:37 buttthead.
48:40 Okay, it doesn't matter. Sam was just
48:42 going on about some dumb [ __ ] Tik Tok
48:44 pin, tell us about Claude Mythos. So,
48:47 Mythos is is Anthropic's new Let's Let's
48:51 go see if there's a decent post about
48:53 it.
48:54 Um,
48:57 so Claude had two major leaks this past
49:00 week. One was Mythos or or the existence
49:03 of Mythos.
49:05 Anthropic just deleted their official
49:06 blog post announcing Claude Mythos.
49:11 All right, let's see. Is this the
49:36 saved before it was taken down?
49:40 Okay. Okay. So, here's the here's
49:42 apparently what was what was on the
49:44 anthropic website.
49:46 So, let's read this together. Am I are
49:50 you looking at my screen? No. Share this
49:53 tab instead.
49:55 No sound on YouTube.
49:58 Mythos is a new name. A new name for a
50:01 new tier of model larger and more
50:03 intelligence than our opus models
50:06 which were until now our most powerful.
50:08 We chose the name to evoke the deep
50:10 connective tissue that links together
50:12 knowledge and ideas.
50:14 Compared to our previous model, best
50:16 model, Claude Opus 4.6, Mythos gets
50:20 dramatically higher scores on tests for
50:22 software coding, academic reasoning, and
50:25 cyber security among others.
50:28 In preparing to release Claude Mythos,
50:31 we want to act with extra caution and
50:33 understand the risks it poses, even
50:35 beyond what we learned in our own
50:37 testing. In particular, we want to
50:39 understand the model's potential,
50:41 near-term risks in the realm of cyber
50:44 security, and share the results to help
50:46 cyber defenders prepare.
50:49 All right,
50:52 discussing Claude Mythos. Thanks, Tik
50:54 Tok, for telling me what we're doing. I
50:57 appreciate that.
51:00 Claude Conway is in research preview.
51:03 Oh, okay. That's cool. I didn't know
51:04 that. Panda,
51:08 I'm sure. I don't really want to know
51:09 what's coming yet. I know. Source camp.
51:11 I'm I'm in the same kind of boat. I've
51:13 I've read all these posts about this
51:15 post, but I haven't actually read it.
51:16 So, we we'll read it together tonight.
51:19 >> Mythos is also a large compute inensive
51:22 model. It's very expensive for us to
51:25 serve and will be very expensive for our
51:27 customers to use. We're working to make
51:30 the model much more efficient before any
51:32 general release. For those reasons,
51:34 we're taking a slower, more gradual
51:36 approach to releasing Mythos um than we
51:39 have our other models. We're beginning
51:41 with a small number of early access
51:42 customers and we'll explore the model
51:46 cyber security applications. It's wild.
51:49 A head start for cyber security. We have
51:51 written several times in recent months
51:54 about the rapid changes in AI models,
51:56 cyber security skills, skills that can
51:58 be used for good or ill. We've
52:00 documented the ways that models can be
52:02 used to rapidly discover vulnerabilities
52:04 in code bases, but we've also shown how
52:07 they're already being used to commit
52:09 large-scale cyber attacks. Although
52:10 Mythos currently is far ahead of the
52:12 other models in cyber capabilities, it
52:15 precages it precages an upcoming wave of
52:19 models that can exploit vulnerabilities
52:21 in ways that far outpace efforts of
52:23 defenders. So it's this going to be a
52:26 hacking beast.
52:34 I'm disappointed in you people not
52:36 saving me a seat. Sorry, Jason. They're
52:39 they're very selfish, the irregulars.
52:41 They come in, they just they gather
52:43 right around. They all get in the front
52:44 row and then when other irregulars come
52:47 in late, they're like, "We sorry
52:50 you're late." No, I They're very rude.
52:53 They're very rude. The irregulars.
52:56 Kelly Camp, I think she secretly leads
52:58 up. There's a there's a cabal, you know.
53:02 It's It's Claude Conway will will
53:05 replace Open Claude. Oh, that's
53:06 fascinating.
53:08 Did anyone bring wine, Jason? No soup
53:12 for you.
53:15 Okay, so there's Okay, pre-release
53:17 safety testing with all of our models.
53:18 We've tested them. Blah blah blah.
53:20 Expanding the release. Oh, that's it.
53:23 And then here's copy bar.
53:27 All right. Well, these are just short
53:29 little
53:30 there's nothing there.
53:34 So basically, the models are going to
53:36 get stronger. You're going to be able to
53:38 hack easier. That's what we learned so
53:41 far. We knew that six months ago.
53:46 But somebody wrote it down somewhere. So
53:48 now it's a thing. Um, did anyone see the
53:52 new Pika where you video chat with
53:55 agents? No.
53:58 Let's go look at that. Pika.
54:02 Let's see.
54:05 Pa.art.
54:10 Share this tab instead.
54:14 Hey, I have a question for all you
54:16 nerds.
54:18 For Gareth and
54:21 Jason and whoever's Panda, you got this
54:24 [ __ ] figured out. I have a question for
54:26 you. How come
54:32 sites that use SSO?
54:35 So if I log in with my Google SSO with
54:39 the email address kyle@storyvine.com
54:44 which is tied to my Google account. How
54:46 come that is treated as a separate
54:48 account
54:50 from the email login kyle@stoyvine.com?
54:55 Why?
54:57 Who came up with that stupid [ __ ]
54:59 rule? All right, back to your regularly
55:02 scheduled thing. The new Pika is insane.
55:04 Mine is obsessed with me and it message
55:07 me all the time. Are you kidding? Wait,
55:08 what? What? Where?
55:12 How do you What do you How do you talk
55:15 with it?
55:17 Settings. Peak of frames.
55:20 Peak effects.
55:26 Pika scenes, Pika editions, Pika swaps,
55:28 Pika frames, selfies, Pika twists.
55:33 >> I think this is a different thing called
55:35 Pika. Me or Pikah me something. It's not
55:38 the Pika. Art video platform. Oh,
55:42 Pika.me.
55:45 Oh,
55:47 try it on the app. Download on GitHub.
55:50 Realtime video chat with agents. You can
55:53 now actually have a conversation with
55:54 your agent on Google Meet.
55:59 Try it on the app. Download on GitHub.
56:05 Huh?
56:07 Pika skills.
56:10 There you go.
56:14 Do it live. How do I do it live?
56:17 Pika video meeting available skills 50
56:22 cents a minute.
56:24 Join Google Meet as a real-time avatar.
56:26 But it's 50 cents a minute. I'm not
56:29 doing that.
56:33 No.
56:35 What did you guys pick in LM Studio for
56:38 the Claude that got leaked? Wait, what
56:41 did you guys Panda? Holy [ __ ] Invite
56:45 them to a meeting. and hook them up to a
56:47 knowledge base. Yeah. Damn. 50 cents a
56:49 minute. Exactly. I don't Even 50 Cent
56:52 couldn't couldn't afford 50 cent a
56:54 minute.
57:05 What's Oh, pi.ai got some updates.
57:09 That's fascinating. Get the app. Yeah,
57:11 but I can't Oh, wait. I could get the
57:12 app on my iPad, couldn't I? pika.mme
57:16 pika
57:19 I don't know people there's there's too
57:21 much you can't you can't follow this
57:24 stuff anymore pika
57:32 pika aai
57:36 pika labs pika effects
57:41 pika the self agent
57:44 All right, I'm getting it. Install.
57:49 Um.
57:52 Oh, come on. Really?
58:02 Um
58:10 Here
58:16 go people.
58:45 I can't get it. It requires iOS 18 or
58:47 later. That's a drag. Oh well. Um I'll
58:51 I'll get that for next time. We really
58:53 need to talk about when code
58:56 gets read and used as token spent as a
58:59 form of taxation. I don't quite know
59:02 what that means. Jason
59:08 Well, that ruined it fast. I love
59:10 watching the most important technology
59:11 of human kind get turned into a utility.
59:15 Well, that's going to that's what's
59:17 going to happen. I mean, it happened
59:18 with the internet. Like the the
59:20 internet, you know, there there was sort
59:22 of the pure version vision of the
59:24 internet and then there was what
59:26 actually happened with it which is
59:27 turned into a piece of crap.
59:30 Crap. Crap.
59:32 Um, let's go back to Claude Code for a
59:35 second and see what how our simulation's
59:37 coming.
59:40 You've hit your limit. God damn it.
1:00:26 You can't see any of this.
1:00:46 So, we have a little simulation here
1:00:47 going. That's kind of cool.
1:01:53 Gary Void, I just upped the 200 max $200
1:01:56 max yesterday.
1:01:58 Cam Kek and I'm hoping I can do the same
1:02:00 on my iMac.
1:02:04 I just ordered a badass uh
1:02:08 a badass computer.
1:02:13 Well, badass for me.
1:02:15 I got a uh an M5 Max MacBook Pro with 64
1:02:22 gigs of RAM.
1:02:24 You get 20x the usage.
1:02:29 I might do the $20 plan.
1:02:32 Gary. Gary Vat or Voit Vat. Is it on uh
1:02:38 Did you get it on um
1:02:41 Wow. How much is that? It was
1:02:45 $4,200
1:02:47 I think. $42.99.
1:02:50 I got the 14-in MacBook Pro because I
1:02:53 figured like the the 16in was an extra
1:02:56 grand and it and architecturally it's
1:02:59 not any better or worse. So, I went with
1:03:02 the 14 inch because I can use an
1:03:03 external monitor if I want something big
1:03:06 and it's also just smaller, right? And
1:03:09 then I got the M5 Max
1:03:12 with whatever the like I did the maxed
1:03:14 out processor,
1:03:16 64 gigs of RAM.
1:03:18 I don't think I upped the It's a two
1:03:21 terbyte drive. I could have upped it to
1:03:23 four or six for 600 bucks and I was
1:03:26 like, "Yeah, I don't give a [ __ ] I'll
1:03:28 just dump stuff off it.
1:03:31 Mimi, that's what I got. The 14-inch
1:03:33 MacBook Pro with the M5. Yeah, that's
1:03:34 that's it.
1:03:37 Um, there's supposed to be a guy that's
1:03:39 going to be the$1 billion dollar
1:03:41 business all AI. Well, so the New York
1:03:44 Times did an article on this guy who um
1:03:49 it's him and his brother right now.
1:03:51 They're going to do 1.8 billion dollars
1:03:53 in revenue. They're selling GLP GLP1
1:03:56 drugs and services and they're just
1:03:58 advertising the [ __ ] out of it on
1:04:00 Facebook. Um, but it's a twoerson
1:04:02 company and they're going to do $1.8
1:04:04 billion in revenue in 2026. They're two
1:04:07 years old. They were started in
1:04:10 September of 2024. They did $43 million
1:04:13 in 2025,
1:04:15 on track to do 1.8 billion in 2026.
1:04:19 Started as a oneperson company. The guy
1:04:21 hired his brother. There's two
1:04:23 employees.
1:04:24 Um, did you look at the Nvidia
1:04:26 supercomput? It's like 5K. I didn't I
1:04:30 I'm just a Mac guy. Like the the the
1:04:32 problem with getting one of the Nvidia
1:04:33 boxes is it kind of assumes you're a
1:04:35 geek. You know, they assume you're going
1:04:37 to be doing your own latent space
1:04:40 optimization engineering. I'm not. I
1:04:43 just want to install [ __ ] on a MacBook
1:04:44 and be done.
1:04:48 I don't give a [ __ ] Um, but yeah, that
1:04:51 that just happened. the the one person
1:04:53 billion dollar company. It's not even a
1:04:55 billion dollar valuation. They're doing$
1:04:56 1.8 billion in revenue.
1:04:59 In revenue,
1:05:02 they dropped
1:05:05 $63 million or $65 million to the bottom
1:05:08 line. $65 million in profit on 403
1:05:12 million in revenue in 2025 for one
1:05:15 employee. 65 $65 million in profit.
1:05:18 That's not a bad first year.
1:05:21 [ __ ] crazy.
1:05:27 Lol. I can trade in my hearing aids for
1:05:29 that 5K supercomput.
1:05:32 That's a nice machine. I was I was on
1:05:34 the fence last week week about getting a
1:05:36 Mac, but didn't go for it. I think I'm
1:05:38 going to go in that direction sticking
1:05:40 to Mac. There's so much support. Well,
1:05:43 it's also there's support, but the like
1:05:45 the open source community, they they do
1:05:48 the MLX optimization
1:05:50 models. So, so the open source models
1:05:53 come out and then the nerds go in there
1:05:54 and they create the MLX optimized
1:05:57 versions for the for the Macs and
1:05:59 they're like two to three times faster
1:06:01 um than the base models on on the same
1:06:03 equivalent hardware. Um so, I don't
1:06:07 know. We'll see. Um, Brent Peterson
1:06:10 said, "Don't get yourself a high-end Mac
1:06:12 because all the inference is so crappy
1:06:14 that you can't really It doesn't really
1:06:16 do anything good." But here's how I look
1:06:18 at it.
1:06:20 It's It's a What did I just do?
1:06:25 Did I just go to a different tab? The
1:06:27 hell just happened?
1:06:30 Yeah, I did.
1:06:32 Um,
1:06:34 I think these models are going to get
1:06:36 better and better and smaller and
1:06:38 smaller. So, so I think that probably by
1:06:41 the end of this year, we'll be able to
1:06:43 run something that's probably Opus 4.6
1:06:47 quality or better on on something like
1:06:50 the the MacBook Pro. So glad I went all
1:06:53 in on Microsoft PC and Copilot.
1:07:03 Brandon can do avatars.
1:07:10 Low key, I think that AI may be
1:07:12 destroying society and humanity. Yeah,
1:07:15 just by people posting that low-key they
1:07:17 think AI may be destroying society and
1:07:19 humanity.
1:07:21 Um, most valuations are BS anyway.
1:07:23 Unpopular opinion. I hate Mac. That is
1:07:26 unpopular. Um, I think people are going
1:07:29 all local LLMs to get cost down. I
1:07:31 agree. I I agree. I think you have to
1:07:34 like like once you start running systems
1:07:37 like OpenClaw that run 24/7 and if they
1:07:39 just decide they want to go burn a bunch
1:07:41 of tokens, they're going to go burn a
1:07:43 bunch of tokens. Um, and if you're
1:07:45 paying for those tokens, that ain't
1:07:48 good.
1:07:50 Jason, very unpopular.
1:07:53 I'll tell you what, having installed
1:07:55 OpenClaw on a PC, it [ __ ] sucks.
1:07:58 So,
1:08:00 but I'm a Mac guy, so that's what there.
1:08:03 But there's a lot of engineers are on
1:08:05 Macs, so I don't I I assume because it's
1:08:07 Unix based. We are being gaslit about AI
1:08:11 on a societal level level. Everybody is
1:08:13 vibe coding, but I haven't seen one
1:08:15 useful thing get produced. Really? Okay.
1:08:18 Everybody has agents doing something.
1:08:20 Nothing useful's getting done.
1:08:22 Right. We're in the experimentation
1:08:24 phase. Everyone's trying to figure out
1:08:25 what the [ __ ] to do. All I hear is
1:08:27 people saying AI is going to take their
1:08:29 jobs. Well, learn AI and ride the wave.
1:08:32 Yeah, exactly. That's what if if you go
1:08:35 to the great repurpose.com,
1:08:37 that's what that whole thing's about,
1:08:42 right?
1:08:43 um mourn the fact that your job is gone
1:08:47 or changing and then
1:08:50 figure out what you want to do with your
1:08:51 life and go learn enough about AI to to
1:08:54 go do it. I'm very impressed with 5.4.
1:09:01 I think I'm getting a cold.
1:09:04 I was at this thing I was at last night
1:09:06 was at um ID345 and it was kind of musty
1:09:09 in there and I'm I normally don't have
1:09:11 any kind of allergies or anything but
1:09:13 like my sinuses got all filled up and I
1:09:15 can it's I'm feel it getting filled up
1:09:17 right now again so I have a feeling I'm
1:09:19 getting sick
1:09:25 I'm making stuff and getting paid so I
1:09:27 beg to differ. Vibe coding pays off. I
1:09:30 know. Listen, all of the people that are
1:09:32 saying this this horseshit, I promise
1:09:35 you they haven't used AI.
1:09:37 It's like, is token usage the new unit
1:09:40 of productivity?
1:09:42 Um, I don't know. That's an interesting
1:09:44 question. I don't think so.
1:09:47 Uh, uh, how do you feel about quantum
1:09:49 mechanics? Um,
1:09:52 here, I'll show you how I feel about
1:09:54 quantum mechanics.
1:10:11 Yeah. Yeah.
1:10:21 I'm tangled with the beat. Spooky action
1:10:23 in the booth. Split the atom with the
1:10:25 pattern. And every bars averse to truth
1:10:26 superp position in the mission fusion
1:10:28 fusing my intentions parallel in every
1:10:30 line I'm a prism OF DIMENSIONS
1:10:32 MULTI-ELABIC ANTICS I'm manic with the
1:10:34 semantics MY RHYMES COLLIDE LIKE
1:10:35 PARTICLES dire gigantic oscillate and
1:10:38 dominate I'm weaving quantum states spit
1:10:40 a burst and WATCH THE MULTIVERSE
1:10:41 RECIPROCATE quantum precision I'm
1:10:43 ripping the rhythm I'm spitting in
1:10:45 prisms infinite visions I'm shifting
1:10:46 dimensions with liber decisions
1:10:48 accelerate levitate elevate thoughts in
1:10:49 a turbulent spin I'm the velocity
1:10:51 ripping apart the certain you're in so
1:10:53 step into the quantum Where the chaos
1:10:55 aligns MY BARS ARE INIC LOGIC REDEFINING
1:10:58 THE divine from THE SMALLEST OF
1:10:59 PARTICLES to the cosmos that we know.
1:11:01 This is rap at its limit quantum flow.
1:11:07 Let it glow.
1:11:14 >> That's what I feel about quantum
1:11:16 mechanics right there. That's from the
1:11:18 new hit m musical Sydney and artificial
1:11:22 love story.
1:11:26 Um, I gotta get out of here. I am not
1:11:28 feeling too great. So, let me get on out
1:11:32 of dodge. So, next Tuesday is AI salon
1:11:35 presents. So, mark it in your calendar.
1:11:38 Go to community.thesalon.ai
1:11:41 if you're not a member of the AI salon
1:11:43 and join us. And then go to events and
1:11:46 RSVP
1:11:47 for the uh to the uh presents meeting
1:11:52 next Tuesday. It's going to be HT Snow
1:11:54 Day talking about how he's taken his
1:11:57 thousand person organization, gone all
1:11:59 in on AI, and how in two years
1:12:03 everything's different.
1:12:06 Like everything's different about AI two
1:12:09 years in.
1:12:12 Like all of the training materials that
1:12:14 he created for his people are useless
1:12:18 because AI has moved on.
1:12:21 Get rest, hydrate, get well. Thank you.
1:12:23 I will do that. I'm going to go take
1:12:25 something from my sniffles.
1:12:27 Get some rest, Kyle. You need some chill
1:12:29 time, too. Thank you very much. Weekend
1:12:32 homework. Um
1:12:36 I No homework this weekend. No homework
1:12:39 this weekend. Um
1:12:44 no.
1:12:46 Go touch grass. Go sit on a bench.
1:12:50 Go sit on a bench. take some time to
1:12:54 breathe.
1:12:56 Um, if these new models coming out are
1:12:59 are really as strong as people are
1:13:01 saying they are,
1:13:04 no one will notice.
1:13:07 But but if we're paying attention to
1:13:10 them and they get really good, we're
1:13:12 going to get stressed out. So, I think
1:13:13 we just need some energy. I think we
1:13:14 just need to rest this weekend. So,
1:13:16 weekend homework is rest. Love your
1:13:18 family.
1:13:20 Um, breathe, sit on a bench.
1:13:24 Oh, yeah. Oh, that's right. Tomorrow is
1:13:26 an LOL with Suzanne Welker Jurgens. Um,
1:13:30 where she's talking about uh her
1:13:32 manifesting work using AI. So, go to
1:13:35 that. It's 4 PM tomorrow. Um, I'm going
1:13:39 to be downloading Gemma for his
1:13:41 homework. Very good. Very good. I like
1:13:43 that. Uh, if anyone bought a new
1:13:44 computer, uh, Chef Kelly, thank you very
1:13:47 much. I appreciate that. Um, if anyone
1:13:50 bought a new computer, set up your [ __ ]
1:13:52 Um, I ordered a new computer, but it's
1:13:54 not coming for two weeks. But I did get
1:13:56 new headphones, so I can hear again. My
1:13:58 headphones suck. They were broken. Uh,
1:14:01 so I'm excited about that.
1:14:04 Make sure it's not the bench I sleep on,
1:14:08 Steo.
1:14:11 All right. Peace out, everyone. Have a
1:14:13 great weekend. I will see you next uh
1:14:15 next Monday. Bye. Thank you for whoever
1:14:18 just sent that heart flower thing. Who
1:14:20 was that?