11/21/2025 - Google's Gemini 3 Takes the Clear AI Lead After Three Years of Competition

Description

Chapters

Transcript

0:06 [music]
0:12 [music]
0:16 I feel it.
0:41 in a westerly direction.
0:47 This car is my train.
0:51 I've been driving. I've been wondering
0:56 what it is I'm running from again.
1:01 Feel like a gy [singing] old man.
1:05 Holding on to 29.
1:09 I just realized my little microphone was
1:11 not plugged in. Got this fancy ass
1:14 microphone. You got to plug it in if you
1:16 want it to work. Hello. Testing,
1:18 testing, testing, testing.
1:23 Up ahead of trucks carrying a wide load.
1:27 Preab house cut in half.
1:32 Cute [music] little front door and two
1:33 windows. My law
1:37 should laugh.
1:45 >> Sorry to interrupt the concert.
1:47 >> Yes. If you want to the microphone to
1:50 actually turn on after you plug it in,
1:51 you have to change your audio settings
1:53 inside of StreamYard or else it won't
1:55 actually pick up the fact that you
1:57 turned on your new microphone.
1:58 >> Yeah,
2:00 let's see.
2:03 >> Thank you. Microphone. Boom.
2:06 >> There it is. Hello. Testing. Testing. Is
2:09 that good? Hello. Test. Test. Test.
2:12 >> You're good.
2:13 >> All right. Beautiful. Beautiful.
2:15 Beautiful. Thank you. Thank you for
2:16 that. Um, okay. All right. Fantastic.
2:20 Fantastic. There's Cam Catkin from
2:22 Cleveland in the house. We got
2:24 Cleveland. Well represented.
2:27 Fantastic.
2:30 All right. There's that.
2:31 >> You know, the amount of time I spent
2:33 thinking she actually lived in
2:34 Cleveland. Like, I was trying to network
2:36 with her [laughter] and you're like, I'm
2:39 not actually from Cleveland. I don't
2:40 know why he says that. [laughter]
2:43 All right. From the
2:46 >> [laughter]
2:48 >> I say it because it's funny. It's funny.
2:50 It's comedy.
2:52 [laughter]
2:58 [music]
3:07 >> [music]
3:12 [music]
3:17 [music]
3:26 >> Woohoo!
3:39 Standing between
3:42 you and the places in
3:47 standing near
3:50 you and the fire makes it clear
3:56 your trouble to me.
4:01 >> [music]
4:02 >> Rich trouble caves you'll see.
4:06 [music]
4:07 Leading in close.
4:10 Smell of your perfume scares me most.
4:16 Leading away. Hey
4:19 feel stronger every day.
4:25 You're trouble to me.
4:30 Red can you see?
5:00 Oh, good evening, good people. Friday
5:01 night, date night. Hello. Hello. Hello.
5:04 Hello.
5:07 Oh, we're going to go play tonight. What
5:09 are we going to do? We're going to play
5:10 with I don't know. There's There's a lot
5:11 to play with. I want to go back to Nano
5:13 Banana. Dig into that a little bit. Um,
5:15 super fun sale. set a live goal and
5:19 invite viewers to enjoy 50% off their
5:22 first month purchase of Madagascar
5:24 vanilla beans.
5:27 Fantastic. We've got the fantastic nug
5:30 like the Sonic ice makers, the nugget
5:33 ice makers.
5:35 I someone someone at Tik Tok's going to
5:37 get fired because these prices are just
5:40 too low. [laughter]
5:46 You know what I mean? There's so just
5:48 click the box. You're not going to get
5:50 prices like this until like an hour from
5:53 now. [laughter]
5:59 What am I doing with my life? What am I
6:02 doing with my life? It's really good. We
6:06 got electric bikes. We bought four of
6:08 them. We can't get rid of them.
6:11 It's going to be good.
6:20 >> [laughter]
6:27 [music]
6:35 [music]
6:46 [music]
6:53 [music]
7:01 [music]
7:07 [music]
7:18 >> Heat. Heat.
7:34 Hey.
7:40 [music]
7:46 [music]
7:52 [music]
8:13 >> [music]
8:18 [music]
8:24 [music]
8:33 >> I swear, Kyle, I'm going to start
8:35 collecting your characters. [laughter]
8:38 Link compilation clips. That would
8:40 actually be a fun idea cuz I swear to
8:42 God, I don't know what my characters
8:44 are. Like the one I just did, the ice
8:45 maker one, was just like the depressed
8:48 e-commerce guy. [laughter]
8:54 >> [laughter]
8:57 [snorts]
9:00 >> the overcompensating depressed
9:02 e-commerce guy. [laughter]
9:06 Oh man.
9:08 All right. Welcome. Good day. Good day.
9:10 Good day. Good day. Good day. Good day.
9:13 Good day. Good day. Good day. Good day.
9:15 Good day. Good day. Good day. Good day.
9:17 Good day. Good day. Good day. Good day.
9:19 Good day. Good day. Good day gay day.
9:22 Good day.
9:24 Depressed e-commerce guy would be a good
9:27 Friday night g date night host.
9:30 [laughter]
9:33 I'm starting to name them as you go.
9:35 Well, that's good. I you know I could I
9:38 could uh I could audition for Saturday
9:40 Night Live as the first guy over 60 that
9:42 they hire. [laughter]
9:48 [clears throat]
9:50 Oh my god. Um, what a week. Y'all have a
9:54 Y'all have a week. It There was a lot.
9:57 Um, when it comes to the AI stuff, it
9:59 was a lot. For me personally, it was a
10:00 lot. Just a lot of a lot of
10:03 shape-shifting I had to do this week.
10:06 Context switching, different different
10:09 projects, different mindsets, different
10:12 frameworks.
10:14 Um,
10:16 which is it's a lot. Um, M
10:21 what do y'all think? What are you
10:23 excited about? What are you not excited
10:25 about? What do you want to talk about?
10:26 What do you want to play with?
10:32 And the lot is in 100 flies. Oh, cool.
10:39 How?
10:44 I can't. There's no way for me to copy
10:46 this.
10:49 Ban user and delete their comments. That
10:51 would be rude to Vicki.
10:59 What were your highlights? Your week
11:01 highlights
11:03 this week. Um, that's a good question,
11:06 Cam.
11:07 Um,
11:14 just go to 100flies onthewall.com. Okay,
11:16 cool.
11:21 Did you see the doodle art I did, Vicki,
11:23 on LinkedIn? 100 flies on the wall.
11:30 Um,
11:32 let me go
11:36 the week that felt like a year. I guess
11:38 I've been saying that all day.
11:40 [laughter]
11:42 Let me share this tab.
11:46 All right, we're sharing.
11:50 I did. I made one, too, but not on the
11:52 first try.
11:54 Yeah, mine probably wasn't on the first
11:56 try either [laughter] to be quite
11:58 honest.
11:59 A lot of times I'll present stuff as if
12:01 it just happened, but it's it's in a
12:03 longer conversation and I don't feel
12:05 like talking about all the context.
12:08 The week that felt like a year.
12:11 The mood, dizzy, delighted, low-key
12:13 terrified.
12:15 The globe demo that ruined everyone's
12:17 productivity. Kyle kicked things off by
12:18 casually dropping a small bomb. You
12:21 build an interactive spinning globe app
12:23 in 15 minutes using Gemini 3 and AI
12:25 Studio. Not a mockup, not a prototype, a
12:27 working app that when you drop a pin
12:30 anywhere on [laughter] Earth, Nano
12:31 Banana generates a historical moment and
12:34 a hero image for it. Yeah. Yeah. Pretty
12:36 crazy.
12:38 [sighs and gasps]
12:40 Yeah. Nano Banana is not an image model.
12:42 It's something we don't have a
12:44 vocabulary for. Well, the the the uh the
12:47 Frontier model companies have have a
12:50 name for it. It's multimodal models. Um
12:54 but I feel like the multimodal modalness
12:56 of the models has been fairly hobbled
13:00 up until um Gemini 3 and Nano Banana.
13:07 [clears throat]
13:08 Wow, that's so cool. These are so cool.
13:13 These are amazing, Vicki. Thank you for
13:15 doing these. So, if you go to 100 flies
13:17 on the wall,
13:18 what those are are those are those are
13:21 nice little snippet recaps of the office
13:24 hours. So, we had office hours today and
13:27 you know, people come in. Sometimes I
13:28 show stuff. Today I showed a bunch of
13:30 stuff because there was a bunch of stuff
13:31 to show, but a lot of times office hours
13:33 are just people talking. So, there's
13:34 always really good wisdom in these. So,
13:37 if you want to just cruise around and do
13:39 a little osmosis learning, go to a 100
13:41 flies on the wall and uh yeah, it's
13:44 gorgeous. It's gorgeous, I tell you. Um
13:48 really nice. Uh Mr. K made a Tik Tok
13:51 saying he's quitting teaching to consult
13:53 on the district. Yeah, Teton Todd.
13:54 That's he it's basically his dream job.
13:57 He was hoping to get a a job above
14:00 above, you know, school
14:03 um at the district level and he got it.
14:05 And so he's basically advising on how
14:07 they can not be idiots with AI. So
14:10 [laughter]
14:12 good on him. We, you know, another minor
14:15 victory for uh for one of our uh one of
14:18 our community members. Actually, it's a
14:20 major victory. It's really cool.
14:26 That's [clears throat] a big decision
14:27 and very important. Mr. K. Uh yeah. Uh
14:29 Mr. Caranakus,
14:32 I forget how to say his last name. Uh
14:34 Matt. Matt K, Mr. K. Um, he was the one
14:39 that that co-led the uh the education
14:41 guild when it was a guild with with
14:43 Stacy Le. Um,
14:47 Kubernauts,
14:50 huh?
14:53 Oh, Matt Kubernauts.
14:57 Oh,
14:59 that's how you remember it. Oh, okay.
15:01 Cool.
15:03 Carabinos. Matt Carabinos.
15:06 I knew it was Greek. That's Greek,
15:08 right? Carabinos. I think that's Greek.
15:11 [laughter]
15:12 Producer Brandon gave me a phonetic
15:14 spelling of it that was completely
15:16 wrong. [laughter]
15:20 That's awesome. [laughter]
15:22 It's not Cubernaut. It's Carabinos.
15:28 Oh, lordy. That's hilarious.
15:31 Um, what were my highlights of the week?
15:33 was the question from Cam Katkin. Um,
15:43 [sighs]
15:44 [snorts]
15:53 there's a weird
15:56 This has been coming,
15:59 but but I've seen more posts on this
16:01 this week than than any other time.
16:06 which is
16:09 I think Google's
16:11 Gemini 3 and Nano Banana
16:15 joint launch this week
16:18 um is the first time since November
16:22 30th, 2022
16:24 that Open AAI hasn't been in the clear
16:26 lead.
16:27 Like I know we've always had like
16:29 benchmarks have gotten close to OpenAI,
16:31 slightly above it, slightly below it,
16:33 but it's always in the neighborhood.
16:35 Um, Gemini 3 is
16:40 clearly different than anything else out
16:42 there. Nano Banana 2 or Pro is clearly
16:48 different than anything else out there.
16:50 So I think my biggest highlight from the
16:52 week is like I would say this is the
16:55 week, what is the week? The 21st. So
16:57 essentially close to 3 years to the day
17:00 that chat GPT launched,
17:03 Google went from being effectively a
17:06 laughingtock with AI to dominating, you
17:11 know, at least right now in the moment,
17:13 right? That that that I feel like
17:15 they've taken the lead for the first
17:16 time in three years. And their first
17:18 year and a half was was an
17:20 embarrassment. They were they were
17:21 really bad at this AI stuff. And unlike
17:24 Apple
17:26 um who just basically did nothing for
17:29 those three years, Google was at least
17:32 trying and then you know at some point
17:35 in those three years they put their best
17:36 people on it and all of a sudden their
17:39 AI investments are are starting to pay
17:42 off and they put they put some product
17:45 people in there um like Logan Kilpatrick
17:48 and people that are actually
17:51 thinking about who the target audience
17:53 is for their products and how you find
17:54 them, things like that. So anyway,
17:58 National Paid Appreciation Week. Yeah,
18:00 exactly.
18:01 Exactly. I don't know if Pate Mod's in
18:03 the house tonight, but
18:06 I think I think this is the first week
18:09 this is the first week that if I were
18:11 Open AI, I'd be going like, "Fuck,
18:14 [ __ ]
18:16 [ __ ]
18:19 We didn't think they'd come this fast.
18:21 God dish.
18:24 How's GPT6 coming? Buler, anyone?
18:29 Anyone?
18:33 That Sundor Pai is really getting under
18:35 my skin now.
18:39 Demis Sabas, you jerk. You
18:46 We've got Sora. You can make cameos.
18:50 You can make little clips of your
18:52 friends.
18:55 Could someone could someone get me a wet
18:58 cloth?
19:00 [laughter]
19:02 That's That's my Sam Alman impression
19:04 this week. [laughter]
19:10 Yan Lun quitting meta.
19:13 Poor Sam Walman like uh uh how did they
19:16 catch up? Not enough vocal fry.
19:22 I've got way too much energy for Sam. I
19:25 got I really got to get a Sam
19:26 impression. I really do.
19:29 Yeah. [groaning]
19:32 Sam. Oh man.
19:39 Dude, we hard to stretch that
19:40 hypothetical money as far as Google's
19:42 real money. Well, that was someone I I
19:46 forget who it was. one of the guys I
19:47 respect today. Oh, what's his name? Uh
19:50 Kyle,
19:51 the English dude who's also named Kyle
19:53 that does AI stuff. He um he said today
19:56 I thought I thought I I missed this and
19:59 but it makes perfect sense. It actually
20:01 made me think of Pate M that um
20:06 the thing that is doubly impressive
20:08 about Gemini 3 being as good as it is is
20:11 that it was trained
20:13 from what I can understand solely on
20:16 TPUs.
20:18 So those are Google's chips, Google's
20:21 tensor processing units. Pate works on
20:24 optimizing TPUs. Um that's that's the
20:28 work he does. And
20:31 That means that they're not using Nvidia
20:34 hardware. They're not using Nvidia GPUs.
20:36 So, you've now got um you've got a model
20:40 out there that's better than anything
20:41 else out there that wasn't trained on
20:43 Nvidia GPUs. That that I find pretty
20:45 amazing. Um it's really a big Silicon
20:49 Valley soap opera. It really is.
20:53 Side hustle. Mima had two procedures
20:55 done today and I've never been happier
20:58 for anesthesia. Oh, side hustle me. I
21:01 hope you're feeling okay or healing
21:03 well. Rest this weekend. Please rest,
21:05 rest, rest. And you know, sleep tonight.
21:07 Just use my voice as a as a sleeping
21:10 aid. Take take your painkillers and just
21:15 check on out. [laughter]
21:18 Oh man. I'm not going to be doing
21:21 anything too interesting tonight. I
21:24 don't think. You never know.
21:26 um
21:32 had an interesting conversation.
21:35 Oh, art gallery opening. Yeah, I can
21:37 show you some art gallery stuff.
21:40 [clears throat] So, I just got back.
21:41 Actually, Gabby Gabby called in six sick
21:44 to her own art opening. [laughter]
21:48 So, she's not feeling great.
21:52 [clears throat]
21:53 So I went in herstead.
21:55 Is that what you say? In herstead.
21:57 That's something you would say on
21:58 Bridgetton, right? Yes. Yes. Mr. Shannon
22:02 went to the opening in Miss
22:03 Shannon'sstead.
22:06 He rode the steed in herstead.
22:10 [laughter]
22:14 Uhhuh.
22:17 Uh, let's see.
22:22 Where's Oh, here we go. So, here's
22:26 here's some of her work. [clears throat]
22:28 So, this is the show. This is the stuff
22:30 that I hung,
22:32 we say, instead instead of her. Um, so
22:35 there's that. That's a 36x 36 piece in
22:39 the middle. And those are 12 x 12s on
22:41 the outside. And they're they're pretty
22:43 cool. like they some of them have these
22:45 these cool little dots
22:48 dots dots and stuff and they're really
22:52 beautiful in person. It's kind of hard
22:53 to see them here but they're they have
22:55 metallics and pearls in them and things
22:58 like that.
23:02 And then the one on the top and the
23:04 bottom that's that's her co-art that
23:06 she's hanging with. Kelly Austin Rolo is
23:10 the top and the bottom piece and that's
23:11 Gabby's piece in the middle.
23:15 And then there's more of them looking
23:17 all pretty. The medium is acrylic on
23:20 canvas. Uh some of these are acrylic on
23:22 board, but it's acrylic paints. They
23:24 don't look like acrylic paints because
23:27 she does some magic [ __ ] with them to
23:29 get them to flow.
23:32 These are big. These are 36 by 36s on
23:35 the end and then 48 by 36 in the middle.
23:39 Can you please hang a black bar? Shoot
23:41 it.
23:42 Oh, by the way, Brandon told me before
23:45 we came on today, producer Brandon has
23:47 been producing the show for a year now.
23:49 He started last November. I find that
23:52 absolutely remarkable.
23:57 I sent some of Gabby's work to my BFF,
24:00 uh, which is a watercolor. So, nice.
24:02 Wow. I didn't think it was acrylic.
24:03 Yeah, they're acrylic. It's so hard to
24:05 get work into beautiful galleries. Yeah.
24:09 I'm so impressed with Gabby's work.
24:11 Yeah, her work's really really gorgeous.
24:14 And in person, it's better. It's hard.
24:15 It's hard to look at this stuff online
24:17 because it's uh it just flattens
24:20 everything out. What What the hell
24:22 happened there? Oh, I see. There we go.
24:26 Mr. Shannon, well hung. [laughter]
24:32 For the first time in my life, I've
24:34 heard that [laughter]
24:43 it's comedy. [laughter]
24:49 And then we had some people there
24:52 tonight hanging out,
24:55 eating snacks. It was called Tides and
24:57 Typography.
25:02 I've heard it said that people come into
25:05 your life for a reason.
25:10 Todds and typography.
25:13 Oh, yeah.
25:16 Cool, huh? All right. I just opened a
25:18 YouTube on my computer to magnify the
25:20 paintings. Beautiful. Oh, that's so
25:22 nice. Here, Cam. I'll since you're uh
25:24 since you just got it up on YouTube, let
25:25 me go back to the these are the ones
25:28 that are
25:30 there's Gabby holding one of the pieces.
25:32 You can see a little of the metallic
25:34 there.
25:41 See how it's got like like these little
25:43 shimmers and things in it? And what's
25:46 cool is when you when you like walk past
25:48 them because it's metallics and pearls,
25:50 the the the
25:52 color of the paintings change.
25:57 She's like Wilson from Home Improvement.
26:00 [laughter]
26:04 This one's pretty gorgeous.
26:09 Very cosmos.
26:14 Yeah.
26:15 Nice. She's got the talent. She's got
26:18 talents and skills.
26:20 Talents and skills.
26:35 That is some magic. I can't believe she
26:37 can do this with acrylic. Yeah. No, it's
26:39 it's uh it's taken her years. I it like
26:44 figuring out all of the like the order
26:47 that you have to pour them in and how
26:49 you have to babysit them and how you get
26:52 certain colors to rise up from
26:54 underneath and other colors to go down
26:56 from the top. And it's crazy. She uses
26:59 all sorts of potions. It is. It's a lot
27:01 of magic and it's a lot of babysitting
27:04 [laughter]
27:05 cuz when when she first started doing
27:08 this
27:09 and she hadn't quite figured it out, she
27:12 would um
27:14 she would have uh a beautiful painting
27:17 and then I'd go back in like an hour
27:19 later. I'm like, "Where's that
27:20 painting?" She's like, "It's on the
27:21 floor. It would just go away." So, it
27:24 took her a while to figure that [ __ ]
27:26 out,
27:28 but she got it dialed in now, boy. I'll
27:31 tell you what. I'll tell you what. I'll
27:34 tell you what.
27:37 [sighs] So, that was nice. Love the
27:39 colors. Yeah, they're they're really
27:40 quite beautiful. They really are.
27:43 They're gorgeous. They're gorgeous. I
27:46 got in a nice AI tussle today with with
27:49 a a a lovely young man, the son of the
27:52 other artist in the show, Owen.
27:56 And uh he had written a paper on how AI
27:59 was going to be the death of intellect
28:01 five years ago. And I'm like, "Yeah, no,
28:03 it's not." [laughter]
28:09 We had a good tussle. It was good. It
28:12 was good. Your shirt is cool, Kyle.
28:14 Thank you. Yeah, this is a this is a
28:16 cool photography. This is a Robert
28:18 Graham.
28:19 Um, how was the gallery? Why do you have
28:21 a [laughter] black eye? Yeah, exactly.
28:26 He he wasn't he wasn't aggressive. He
28:29 was he was he was just he was a little
28:30 sort of, you know, low-key, but he was
28:33 definitely a low-key hater. And then his
28:35 girlfriend come up came up and said,
28:37 "They use too much water." I'm like,
28:40 "You're right. They do." [laughter] She
28:43 didn't know what to do with that.
28:47 It's It's funny when people have
28:49 arguments and they're expecting you to
28:50 fight back and you don't. They don't
28:52 know what to do. They're like, "Uh,
28:54 [laughter]
28:56 where do I go from here?" They t they
28:58 tend to just walk away. [laughter]
29:02 He talked to the wrong person. Well, I
29:04 hope he talked to the right person. And
29:05 I, you know, what I was telling him was,
29:08 you know, for every trope and argument
29:10 that he had, I'm like, like, have you
29:13 used it? Like, let me tell you some
29:15 stories. Like, that's the thing is like,
29:18 I don't have to argue about AI,
29:21 right? I don't have to argue that AI is
29:22 good or bad. Like, AI is neither. AI is
29:25 just a thing.
29:28 But what I get to argue is I get to
29:31 witness people in in this community
29:34 changing their lives. And I've got very
29:36 specific stories. And when people hear
29:38 that, they're like, "Oh, I huh yeah. Oh,
29:42 I mean, of course it could be good like
29:44 that. Well, if you treat it like that,
29:46 it'll be good, but all the other people
29:48 it's going to be horrible for." And then
29:50 at one point he goes he he goes, "I
29:52 agree with what you're how you're saying
29:55 people should interact with AI, but how
29:57 do you train up people to do that?" And
29:59 I said, "That's why I started this
30:01 community." [laughter]
30:08 He was like, "A well, good luck with
30:10 that [laughter]
30:12 little whipper snapper." It was good.
30:16 [laughter]
30:16 Why? Why is it that I'm [ __ ] 60 and
30:19 the 21year-old is like,
30:22 "Technology is bad. Get off my lawn, you
30:24 you old fuck."
30:27 Why am I the one going woohoo AI good?
30:32 [sighs]
30:34 We're Everything is [ __ ] weird.
30:36 Everything is twisted and upside down.
30:39 We are in an upside down world.
30:42 The young people should be [ __ ]
30:44 everything up. If ever there was a
30:46 technology that was the [ __ ] things up
30:48 technology
30:50 that is what the youth is supposed to
30:52 do, it's AI.
30:54 Why is it us?
30:57 Just cuz we're bitter and we've been
30:58 hanging out for three or four decades
31:00 waiting for this moment. All right,
31:03 we'll take it. We'll step in.
31:07 But we're tired, man.
31:10 We should be on the porch going, "Get
31:12 off my lawn, you flipping kids with your
31:14 stupid AI. Robots are going to kill us."
31:19 It's them saying that.
31:21 What happened?
31:24 Where are the rebels?
31:27 [laughter]
31:29 Joy Party, you missed Mrs. Shannon's
31:31 art. I'll put it back up. I'll put it
31:33 back up for you, Joy Party.
31:49 >> [sighs]
31:51 >> Wow, I can't get wicked songs out of my
31:55 head. It's driving me crazy. And about
31:58 the time I forget it, there's Cynthia or
32:03 Cynthia Arivo or Ariana Grande on
32:06 another talk show singing or crying.
32:10 I've heard it said,
32:15 "Make it stop. [laughter] Please make it
32:18 stop. Please make it stop. Please."
32:22 [laughter]
32:26 My wife [ __ ] down the aisle to that
32:28 song. [laughter]
32:35 Our skull closets are bursting with
32:37 ideas. Kyle Alpha Shannon. Exactly.
32:42 [laughter]
32:48 Oh my god.
32:50 And it's just that song. It's for good.
32:53 I and and it's just the first two lines.
32:57 Gabby said the other night, she goes,
32:59 "You have to learn more lines to
33:01 [laughter] that."
33:06 I'm like, "I know." Then I would
33:08 [laughter] say, "I've heard it said."
33:20 [laughter]
33:22 Oh, good lord. All right, here. Let's
33:25 edit this. This is driving me crazy how
33:27 crooked this is. Let's Let's get a
33:30 little OCD together, shall we?
33:33 [laughter]
33:44 D
34:02 I have been changed [singing]
34:04 for good.
34:12 There you go.
34:14 That's those pitches.
34:16 Aren't they Aren't they pretty? Aren't
34:19 they pretty? Joy Pury.
34:21 Oh, I was talking about Joy Pretty
34:23 tonight with this this young little get
34:26 off my lawn 21-year-old. AI is evil. AI
34:29 is evil. I'm like, tell that to flipping
34:32 woman that's transforming her life right
34:34 now.
34:36 She's had stories trapped in her head
34:38 her whole [ __ ] life. Now she's able
34:40 to get them out. You tell me that's a
34:42 bad thing there, you little punk.
34:46 [laughter]
34:50 I've never heard that song in my life
34:52 except when Kyle sings it. I am so sorry
34:55 that you didn't hear Cynthia Revo sing
34:58 it first. Well, I I guess the good news,
35:00 Vicki, is it only gets better from here.
35:03 [laughter]
35:06 I'm a little punchy tonight. I really
35:08 am. Um,
35:11 all right. What are we going to do? We
35:12 should go do something, shouldn't we?
35:15 Uhhuh. [laughter]
35:22 Hey, people.
35:28 Okay.
35:29 [clears throat]
35:35 Fractious Friday. Um
35:42 trying to think what's going to be
35:43 interesting.
35:57 Let's do Tik Tok pin. Kyle, now you can
36:00 make a nulling of your wife's work and
36:03 make a shirt from it. Oh, that'd be kind
36:05 of cool. Kyle, why do you think Gemini
36:07 Pro is in
36:10 in the lead this week? Um,
36:22 other than semi-trolling
36:25 Elon Musk,
36:28 Sam Alman tends not to give props to
36:31 other Frontier model companies.
36:34 And he did this week. He was basically
36:36 like, "Nice model, Google. Um,
36:41 I heard a couple of rumors today that
36:46 internally at OpenAI, um,
36:51 they're not in a panic, but they're like
36:53 they're clearly on their heels.
36:56 Um,
36:59 you got people like David Shapiro who I
37:02 respect a lot talking about the fact
37:05 that he feels like OpenAI has lost their
37:07 way at about the same time where Google
37:10 has found their focus, right? Google for
37:14 for a year and a half was was just
37:18 it it was like they were spinning up
37:20 projects and they were announcing a
37:22 bunch of stuff but not shipping it. And
37:24 so they they got like a really bad
37:26 reputation for bad models.
37:29 Um confusing where the models were, like
37:32 you couldn't find them. And then even
37:34 worse than that was they would announce
37:35 [ __ ] that they would just never ship. Um
37:38 and in the past year and a half, like in
37:40 the year and a half since then, they've
37:42 started shipping [ __ ] And it's like all
37:44 of the seeds that they planted in that
37:46 first year and a half are now starting
37:48 to starting to sprout and turn into to
37:50 really interesting things. So, I think
37:52 for me it's a combination of a couple of
37:54 things. The um
37:57 I don't know if we can find that chart.
37:58 Maybe you can find that chart um Brandon
38:01 with the with Google so far ahead.
38:03 Gemini so far ahead of the other models,
38:05 but but Gemini 3 is just it's really
38:07 something different um just in terms of
38:10 its quality of of uh
38:15 of what it can do and its reasoning is
38:17 really good. And then on top of that,
38:19 the nano banana piece of it, the image
38:22 generation piece of it combined with the
38:24 reasoning turns it into a visual
38:28 reasoning engine, which we've we've
38:30 never had. Um, and that to me feels
38:33 different. Um, look at this one chart.
38:35 And I don't know if you'll be able to
38:37 see this cam, so I'll try to zoom in on
38:38 it. Um, [clears throat]
38:45 let me share this.
38:52 Why
38:54 can I not share this?
38:59 See, I can't get it out of my [ __ ]
39:01 head.
39:05 All right, this chart.
39:08 So
39:10 this is cost per task and then on the
39:14 on the left hand side is quality, right?
39:17 So cost and quality.
39:20 Um
39:22 so you see all these clusters down here
39:26 like 03 mini high, remember when 03 mini
39:28 high came out? Like it's that little
39:31 blue triangle right there. It's like
39:33 buried. Um, GPT5 high is there. So, it's
39:39 above these other things, but it's
39:42 there. This is Claude Sonnet 4.5, which
39:45 everyone was talking about as amazing.
39:47 Here's Gro 4. Everyone was talking about
39:49 as amazing. Here's GPT5 Pro. Everyone
39:53 was talking about amazing. Um,
39:58 and then if you look
40:01 way out here is Gemini 3 Pro
40:05 and then and then way way out here is
40:10 Gemini 3 Deepthink preview which I think
40:14 you can get to in the uh in the
40:17 development side on the API side.
40:19 So it's just clear these are clearly in
40:23 a different neighborhood of than all the
40:25 other models, right? And and chat GPT
40:29 was pretty much has always been in the
40:31 lead, right? It's always been in the
40:32 lead and the other ones kind of catch up
40:36 and you know like they they leaprog each
40:38 other a bit, but it was never
40:40 significantly different and Gemini 3 is
40:43 significantly different.
40:47 So
40:51 unless chat GPT responds with something
40:53 in this neighborhood, which I I'm not
40:56 hearing buzz that they will. I heard
40:57 that they might drop GPT6 before the end
41:00 of the year, but I [snorts] don't know
41:02 if it's going to be this much better. Um
41:09 I don't know. I don't know. We'll see.
41:11 We'll see. I mean, with all these
41:14 things, again, it doesn't it doesn't
41:16 really matter that like this is just
41:19 going to keep going and climbing and
41:20 climbing and making predictions about
41:24 who's going to be on top five years from
41:26 now is is kind of pointless, but it it
41:28 will consolidate down to three big names
41:30 like it always does. Tik Tok question is
41:33 visual region reasoning like judging a
41:37 book by its cover. Um, kind of. And
41:41 visual reasoning is things like take a
41:43 bunch of torn up pieces of paper and
41:46 have the system rearrange them,
41:49 right? Like we can do that. Like our
41:51 brains work like that. Computers have
41:54 historically had a tough time doing
41:55 that.
41:59 Or the thing that I showed last night.
42:02 Oh, I can show you a new version of
42:03 that. Pop pop pop. Uh
42:06 where did I do that? I did that in
42:08 Gemini, right?
42:10 >> [snorts]
42:20 >> Don't down.
42:39 So, I don't know if you saw this last
42:41 night. Oh, I'm not sharing this.
42:44 Um, this is I this is an updated version
42:49 of a graphic that I showed last night.
42:52 Um, where [cough and clears throat]
42:57 I uploaded this script of my musical.
43:01 Why is that? Oh, I know. I see what's
43:03 going on. I uploaded this script of my
43:05 musical and I had Gemini
43:08 create an infographic of
43:13 of the plot
43:15 and it nailed it, right? It starts out
43:19 at the beginning. It goes to
43:22 the tech team trying to get the safety
43:25 testing done. It goes to the reporter
43:27 asking for the story. Goes through the
43:30 launch of Ping Chat. There's this really
43:33 cool well there's a cool moment
43:35 structurally where the story splits into
43:37 two and it's kind of this back and forth
43:40 between looking at Tara the the
43:43 reporter's wife and Sydney and his
43:46 relationship with both of them. It's
43:48 this parallel thing. It figured out that
43:50 this was a parallel structure,
43:52 [laughter]
43:53 right? So, it's not just making pretty
43:56 pictures. Like, it read an entire play
44:00 and it created an infographic complete
44:02 with song titles. Like, these are the
44:04 actual song titles from the musical in
44:07 the correct places. And it it nailed act
44:11 two as well.
44:14 So
44:16 when I say visual reasoning th this is
44:22 when you put a prompt in it's not
44:25 generating an image. When you put a
44:28 prompt in what it's doing is it's
44:29 running a large language model reasoning
44:32 pass right? You know how when you watch
44:35 reasoning engines and they they're like
44:37 well the user asked for this and
44:38 therefore I'm going to do this and
44:39 therefore I'm going to do this and
44:40 therefore I'm going to do that. It's
44:42 doing that
44:44 all before it gives anything to the to
44:46 the image engine.
44:48 So I assume what's happening here is
44:50 it's doing all that reasoning. It then
44:52 writes up this massive
44:55 detailed prompt about exactly what needs
44:57 to be in the image model and then the
45:00 image model just makes it.
45:04 So
45:05 crazy, right? My organic chemist friend
45:08 saw this in the infographic. Sent her
45:12 imagination. Spinning for molecule. Oh,
45:15 I saw a cool one. Let's Let's go do one.
45:17 Let's go do one. I I saw a cool one from
45:20 I think it was Nate Nate B. Jones did
45:23 one that I thought was brilliant.
45:27 And so we'll go do it, too. We'll go
45:29 steal
45:31 We'll go steal Nate's idea. So, what I'm
45:33 going to say is I'm going to say,
45:36 um,
45:38 I want
45:42 an infographic
45:46 [snorts]
45:47 in
45:49 isometric
45:53 Lego style
45:58 that describes
46:06 the process of
46:11 quantum entanglement.
46:17 And I'm going to go create images. Wait,
46:20 what are those new things? What did I
46:22 just see?
46:25 Visual layout.
46:29 Is that new?
46:34 Okay, that wasn't there last night.
46:38 There's deep think. So deep think,
46:41 that's the one that's like way off to
46:43 the [ __ ] races. All right, so anyway,
46:46 so we're just going to make a banana.
46:48 We're going to make a banana image. So
46:50 we're in thinking mode, thinking pro.
46:53 And so we're going to do an isometric
46:55 Lego description of quantum
46:58 entanglement.
47:02 Um
47:04 be detailed
47:07 in the process,
47:10 explanations,
47:12 explanations,
47:14 visual metaphors,
47:19 etc.
47:27 So now if you think about this
47:33 you need to understand quantum physics.
47:36 You you need to understand what quantum
47:39 entanglement is. You need to be able to
47:41 describe that. You need to be able to
47:43 figure out what are the components of
47:45 that that if you were going to put it
47:46 into a single image would be the most
47:49 important things to put in that image in
47:51 what order.
47:54 You need to write that copy.
47:56 You need to come up with visualizations
47:58 and metaphors for all of that stuff.
48:02 Then you need to be good enough with
48:03 isometric illustration or with building
48:06 Legos [laughter]
48:08 that you could build that thing.
48:12 Oh, this this this is quantum
48:14 entanglement was too easy, I think.
48:19 And this doesn't quite look like Legos.
48:21 It sort of looks like Legos. Creation
48:24 and superp position. Particles created
48:26 together in a shared undefined state
48:28 like two spinning coins. Neither heads
48:32 nor tails yet. Quantum source
48:35 superposition state. The non-local link
48:40 can be separated by vast distances. is
48:42 the entanglement bridge connects their
48:44 fates instantaneously
48:46 exceeding light speed
48:50 limits for influence. Right? So that
48:52 bridge between
48:55 superp position
48:57 coin one and the one on the other side
49:02 spin down and spin up
49:07 measurement. The measurement on once
49:10 measuring one particle state instantly
49:12 collapses its superp position into a
49:15 definite outcome. Spin down or spin up.
49:18 Instant correlation. Bob.
49:22 The distant partner state is immediately
49:25 determined to be perfectly correlated.
49:28 Eg opposite spin regardless of the
49:30 distance.
49:33 So that's not super Legoy. It's got a
49:35 little Lego thing there. Um, so let's
49:39 ask it rather than us trying to figure
49:40 this out. Let's go. Um, what would be a
49:45 more complicated
49:47 physics
49:50 concept where
49:54 you could build
49:57 a really complicated
50:02 uh, Lego
50:05 uh,
50:08 structure [snorts]
50:09 to explain saying it um this current
50:14 example
50:18 isn't
50:20 very Lego like
50:25 [clears throat]
50:33 irregulars at your convenience. Oh boy.
50:37 Uh,
50:41 [sighs]
50:48 [whistles]
50:52 [snorts]
50:55 what's that? The [ __ ] is that?
50:59 [laughter]
51:00 The [ __ ] did it just make?
51:04 Um,
51:07 I'm not sure. Let's see. I'm not sure
51:11 what you made. And this is photographic.
51:18 I wanted an
51:21 isometric
51:24 illustration of a Lego
51:28 construction
51:30 with labels.
51:34 Let's see what that does.
51:50 A quantum disentanglement machine.
52:10 The standard model and Higs field,
52:18 the lepton field, the Bzon field, the
52:21 Higs field,
52:25 the quark field.
52:28 The Higsfield mechanism, the mass
52:30 generator.
52:33 Oops.
52:40 The Higs Bzon detector.
52:45 Wow.
52:47 Um, I have an idea.
52:50 Do tell Kyle. We're going to go AI
52:52 Studio.
52:55 google.com
52:58 and I'm going to go over here and I'm
52:59 going to go to build. Am I in build
53:01 right now? Build. We're going to go to
53:03 build
53:05 [cough and clears throat]
53:08 and I'm going to say um
53:11 I want to be able to upload a photograph
53:20 and have
53:24 the system
53:28 convert that into A Lego
53:38 like a Lego what sculpture? A Lego
53:42 assembly
53:45 that looks
53:48 like the core subject.
53:57 Good lord. subject
54:06 of the
54:09 image.
54:13 I then want
54:16 to be able
54:19 um
54:21 to build the
54:25 model
54:27 with stepbystep
54:31 instructions.
54:33 I think I saw one someone build this.
54:35 So, basically what we're going to do is
54:37 we're going to try to get this to create
54:39 a 3D model thing that you upload a
54:43 photo, it analyzes the photo, figures
54:47 out the math how to turn that into a
54:49 Lego construction, and then let you
54:50 build it.
54:53 All right,
54:55 I'm a lay person. Quantum physics will
54:58 turn our world upside down.
55:04 I want to take a timeline of our family
55:07 life over a hundred years as a
55:09 wallpaper. Oh, I'll show you something
55:10 cool I made with Nano Banana today that
55:13 you could do for every [ __ ] meeting
55:15 you're in. This is really cool. All
55:16 right, let's start this thing building.
55:19 All right, so this is going to go build
55:20 us a Lego maker thing. Um, we'll come
55:24 back to that. So, I'm going to go back
55:26 to Gemini now. Gemini.google.com.
55:40 Can't get this [ __ ] song out of my
55:42 head.
55:46 It's a [ __ ] torture. [laughter]
55:53 Have you ever had the ADHD? Have you
55:56 ever had it? No. It's more It's good.
55:59 It's good. It's exciting for us. We like
56:02 it. Except the with the song thing. The
56:06 song thing. It's uh what do they call
56:09 that? Um
56:12 uh there's a word for it. It's coming to
56:15 me. Uh insanity
56:23 doodle. Uh must be in the other account.
56:27 It's in the other account. You big dum
56:30 dum dum dum dum dum dum. It's in the
56:34 other account. Yeah.
56:39 Oh, this was a cool thing I did. I think
56:42 I showed that though. Is this the one?
56:44 Yeah. Okay. So,
56:49 I went
56:53 So, today we had a Google or we had a
56:56 content evolution collab meeting.
57:00 And
57:04 um
57:06 what were we talking about? Well, we
57:08 were talking about nano banana. We were
57:09 talking about it as a visual reasoning
57:11 engine. We were talking about what the
57:12 [ __ ] that even means. I don't know if
57:15 that's the correct term, by the way.
57:17 Don't Well, you can quote me on that.
57:20 I'm calling it a visual reasoning
57:22 engine. Um I think Google might just
57:25 call it a multimodal model.
57:27 Um, but it but people are talking about
57:30 it like it's a pretty picture maker.
57:32 It's it's not it's way more than that.
57:34 So anyway,
57:36 so we had this meeting and you know, you
57:38 get back your transcripts for the
57:39 meeting and Google's got them all
57:41 snapshotted there, but you still got to
57:42 [ __ ] read them,
57:45 right? And if you got ADD and you have a
57:47 song from Wicked stuck in your head,
57:48 you're not reading [ __ ]
57:51 You're just seeing Cynthia Revo in your
57:53 head going [laughter]
57:56 over and over and over and over. So, you
57:58 can't read. Have you tried to read with
58:00 her in your head singing that two lines
58:03 of that song? No, you haven't.
58:07 Lucky.
58:09 So, I took the transcript, the full raw
58:12 transcript of the meeting. You really
58:15 need to see someone. Oh, I really need
58:17 to see a regular. Oh, no. Okay, I'll go
58:19 over. Um, so I took the the the raw
58:22 transcript of that meeting and then I I
58:25 put Here's my prompt.
58:28 Do I have the prompt? Oh, no, I lost my
58:30 prompt.
58:39 Here's its response. Um, I'm aiming for
58:42 a visual note takingaking style like a
58:44 doodle artist to create an infographic.
58:47 I don't know if you've ever been in a
58:49 high-end so a lot of times in in like if
58:52 you work with Deote or you know one of
58:55 the big consulting companies and you do
58:57 a high-end brainstorming SE session you
59:00 know the ones that cost like 200 grand
59:03 for a two-day session or whatever they
59:05 usually bring in a doodle artist and
59:07 it's someone who will on a giant board
59:11 doodle what people are talking about. So
59:13 as concepts are coming up they're
59:15 drawing things. So, I did a prompt that
59:17 said, "I want you to act as a doodle
59:20 artist that was in this meeting."
59:24 A graphic recorder. Is that what they're
59:25 called? Okay, cool.
59:28 But anyway, here's what it made.
59:32 I mean, this is what those things
59:34 [ __ ] look like. And this is It's
59:36 right. It's right. It started out with
59:38 Cindy [ __ ] talking about the Tony
59:40 Robbins thing. Felt like it had grifter
59:42 energy, which I thought was [ __ ]
59:44 hilarious. Um, she talked about Sabrina
59:47 Sabrina Romangh being the exception that
59:49 did a really good, you know, session in
59:51 the midst of a bunch of infomercial
59:53 [ __ ] We talked about a branding
59:56 update for a product that we're making.
59:58 We talked about habits versus practices.
1:00:00 Um, we talked about world models. We
1:00:02 talked about an AI economic bubble. We
1:00:05 talked about philosophy and Bnee Brown.
1:00:09 Um, and then I started demoing [ __ ]
1:00:11 chat GPT group chat the Gemini vibe
1:00:14 coding thing where I vibe coded the
1:00:16 globe and then we started talking about
1:00:18 nano banana and I showed three examples
1:00:20 so here's like accurate fly tying steps
1:00:23 like that was the first thing I showed
1:00:25 and then a 1977 classroom with TRS 80
1:00:28 computers on the desks and then the
1:00:30 musical script that got turned into an
1:00:33 accurate plot journey map
1:00:35 and then look at this in the bottom
1:00:37 corner happy Thanksgiving see you in
1:00:39 December that's exactly what one of
1:00:41 these artists would do.
1:00:45 I mean,
1:00:47 that's insane.
1:00:50 This is not just making pretty pictures.
1:00:53 This is reasoning visually.
1:00:57 [laughter]
1:00:58 It's a visual reasoning engine, right?
1:01:02 This is like the videos that have the
1:01:04 person doing the Sharpie illustrations.
1:01:06 Yeah.
1:01:08 And I mean, you could turn this into an
1:01:09 animation where you had one of those
1:01:10 hands drawing it. You need to get to
1:01:12 this first, though. You need to start
1:01:14 with the drawing. Well, now you can just
1:01:16 do this.
1:01:18 This is a single prompt.
1:01:21 You could do this for every meeting. Can
1:01:23 you imagine if your job is to is to send
1:01:26 notes from the meeting where you not
1:01:28 only send the notes, but you send one of
1:01:31 these.
1:01:34 And let's say no one knows who you are
1:01:35 in your company.
1:01:37 And your job is literally you're like
1:01:39 the scribe. They're like, "Don't talk in
1:01:42 our meetings. Only the important people
1:01:44 are talking." And you're like, "Okay."
1:01:48 And I don't know why you talk like that.
1:01:49 You shouldn't. You should have more
1:01:51 confidence. You're like, "All right, I
1:01:54 won't say anything. I'll just take
1:01:56 notes,
1:01:58 right?"
1:02:00 And then imagine you take notes. You
1:02:03 secretly throw them into Chad GBT to
1:02:05 make sense of them.
1:02:06 >> [laughter]
1:02:08 >> And then you go to Nano Banana and you
1:02:10 make one of these. And you send this to
1:02:12 the big [ __ ] boss. You know what the
1:02:14 big [ __ ] boss is going to do? How'd
1:02:16 you do this? Oh, I just know more than
1:02:20 you. I knew which tool to use. Well,
1:02:24 this is really remarkable. Can you do
1:02:25 more of these? I could if I was allowed
1:02:28 to talk in meetings, for example.
1:02:31 [laughter]
1:02:32 And then the boss will be like, "All
1:02:34 right. Yeah, you you should probably
1:02:36 come to more meetings. You're going to
1:02:38 get [ __ ] promoted.
1:02:42 Crib your notes,
1:02:44 right?
1:02:46 Crazy. And then look at this. Look at
1:02:49 this. What else I did?
1:02:55 Oh, that was a cool one, too.
1:02:57 >> [sighs]
1:02:58 >> So, in one of my talks I gave,
1:03:01 I talk about this classroom in 1977, my
1:03:05 my advanced math class in seventh grade
1:03:08 in 1977. And I walked in
1:03:12 and on every single desk was a
1:03:14 RadioShack TRS80 computer. And so I
1:03:17 asked Nano Banana to make that and it
1:03:19 did, but it it used the wrong model of
1:03:22 TRS80. So then I went and I got this
1:03:25 picture
1:03:27 That was the model that was on all the
1:03:29 desks. Look how that look how [ __ ]
1:03:31 sexy that was. In 1977,
1:03:35 I walked into a classroom full of those
1:03:38 bad boys. They were all glowing. All
1:03:40 their screens were on.
1:03:44 And so one prompt later,
1:03:48 I have the classroom.
1:03:51 Look at this. These are these are that
1:03:54 computer and Aski art on every screen
1:03:57 says welcome to the future. And on this
1:03:59 one on on the front computer I said show
1:04:01 the basic programming language that has
1:04:05 the code for the ASI art.
1:04:08 Yeah. The the the dual the dual 5 and
1:04:10 a/4 inch floppies built into the case.
1:04:14 Look at this. Look at the [ __ ] for
1:04:17 micica desks that we had in 1977. And
1:04:21 the and that [ __ ] that [ __ ] 1960s
1:04:24 Fallout shelter
1:04:26 plastic ceramic tile floor [ __ ] That
1:04:30 was gross. And the plastic chairs.
1:04:36 Then I And then I said and I I did one
1:04:38 more prompt. I said, "Make it a wide
1:04:40 image." [laughter] So So there's the
1:04:42 wide image.
1:04:44 Look at this.
1:04:51 That's visual reasoning.
1:04:58 The keyboard must feel fantastic to type
1:05:00 with. It was a good keyboard.
1:05:03 It was a really good keyboard. That was
1:05:05 a cool machine.
1:05:07 The Radio Shack TRS80
1:05:10 dual five and a quarter floppies all in
1:05:12 a single exposure in enclosure.
1:05:19 just bonkers.
1:05:21 All right, let's go see if we have a an
1:05:23 app.
1:05:27 Where?
1:05:30 Where is it?
1:05:41 Ah, here.
1:05:45 Brickify.ai.
1:05:50 We're going to go full screen. Okay.
1:05:51 Upload.
1:05:53 Upload any photo. Our AI architect will
1:05:56 analyze it and generate step-by-step
1:05:58 assembly instructions
1:06:01 for a custom microscale set. I don't
1:06:04 know what a microscale set is.
1:06:38 All right, let's upload that image.
1:06:43 Analyzing structure.
1:06:50 Oh, this says built with Gemini 2.5
1:06:52 flash. So, I don't think I had the right
1:06:56 I didn't have 3.0.
1:06:58 We're going to go back. If let's see if
1:07:01 this works. I'm going to assume it's not
1:07:03 going to work.
1:07:08 >> [laughter]
1:07:17 >> 4x gray 1x1 bricks 2x brown. [snorts]
1:07:21 Oh, parts needed for this step.
1:07:25 The voodoo wanderer.
1:07:29 Start over.
1:07:31 Okay, that was bad.
1:07:38 back to start.
1:07:40 This did say Gemini 3 Pro preview,
1:07:43 didn't it? Yeah, it did.
1:07:48 Um, I want
1:08:02 Oh, yours did that, too. It labeled it
1:08:04 wrong. Okay.
1:08:07 [clears throat] Um,
1:08:10 >> no, it didn't label it wrong, Kyle. It
1:08:12 it it is sneaky. Uh, I think I don't
1:08:14 know if it's throttling or if they just
1:08:16 haven't updated it, but unless you
1:08:18 prompt it to use Gemini 3.0 Pro Preview,
1:08:21 it will use gener 3.0 preview to build
1:08:25 the app, but it'll code the app to use
1:08:28 2.5. Like, it doesn't know to tell the
1:08:31 app to use 3.0 in production.
1:08:37 >> I know that those words formed a
1:08:40 sentence. [laughter]
1:08:43 You have to tell it in your prompt that
1:08:46 you want it to use 3.0. Otherwise, it'll
1:08:49 default to using 2.5 in the application
1:08:51 itself.
1:08:52 >> Weird. Okay. I want you to use
1:08:55 Gemini
1:08:59 3 Pro Preview to create
1:09:05 a
1:09:08 Lego
1:09:12 um
1:09:14 object designer
1:09:22 where I upload a photo,
1:09:28 then the app shows me
1:09:33 a 3D Lego
1:09:37 model of the object. object.
1:09:44 We'll put an S behind it
1:09:47 from that photo.
1:09:53 And then I can
1:10:00 use a slider
1:10:03 to
1:10:05 deconstruct
1:10:09 the object
1:10:12 and
1:10:14 get step
1:10:17 by step instructions.
1:10:21 That's not how you spell upload or
1:10:22 deconstruct.
1:10:25 I no longer
1:10:27 accept
1:10:28 that typos are relevant. I They've
1:10:32 always not been relevant, but they it's
1:10:35 a pet peeve of mine, but I don't believe
1:10:38 that anymore
1:10:40 cuz it'll figure it out. And for me to
1:10:43 go back and fix those three words for
1:10:45 you OCD people, weirdos,
1:10:49 [laughter]
1:10:50 we're not weird. We're irregular. I
1:10:53 know.
1:10:54 Doesn't mean you're not a weirdo. Okay.
1:10:58 Both. And it's the genius of and. You
1:11:02 can be a regular and weird.
1:11:05 Anyway, that's not the point.
1:11:08 I'm not fixing the typos.
1:11:11 >> [laughter]
1:11:14 >> But it means a lot to us. [laughter]
1:11:20 Y'all are cute and weird. Exactly.
1:11:24 Exactly.
1:11:26 All right.
1:11:30 It's quiet over there on Tik Tok. Are
1:11:32 you Are you all on Tik Tok? You're all
1:11:34 right. Irregulars while this builds.
1:11:36 Yes. Let's go to irregulars.
1:11:39 regulars.
1:11:47 [laughter]
1:11:49 I've heard it said
1:11:55 see I would have made a good alphaba
1:11:58 [laughter]
1:12:00 or Glenda. Galinda, the gull, the gull
1:12:04 is silent.
1:12:06 [laughter]
1:12:09 Ah, that's pretty good. That's solid.
1:12:12 That's solid. We'll be one hour late. I
1:12:14 hope y'all got that.
1:12:18 [laughter]
1:12:21 All right.
1:12:35 >> [whistles]
1:12:44 [whistles]
1:12:49 >> Oh, I'm not sharing this. Here's your
1:12:51 tab.
1:13:01 >> [whistles]
1:13:07 [whistles]
1:13:10 >> The Lego viewer. Here's our Lego viewer
1:13:13 code.
1:13:15 Stud radius, stud height, unit size,
1:13:19 brick height, plate height.
1:13:22 Amazing.
1:13:28 Yeah, I guess I guess Legos are
1:13:31 so consistent and they've been around
1:13:33 for so long, this is probably relatively
1:13:35 trivial for it to build.
1:13:38 So when I was 1 hour late, I was
1:13:40 actually on time. Yes, correct, Cam
1:13:42 Katkin. I was an hour late. If you were
1:13:45 an hour late, you and I got here right
1:13:47 around the same time.
1:13:52 >> [clears throat]
1:13:54 >> Add lightning fast real-time responses
1:13:57 to your app using 2.5 flashlight
1:14:00 fixes in Typescript error where the key
1:14:04 prop was being res rejected on the brick
1:14:07 3D component and fixes a react hook
1:14:10 rules violation. You know, one of the
1:14:12 things that I hate are React rules
1:14:15 violations. All right, let's go full
1:14:19 screen.
1:14:21 Oh, this is cool. Upload photo. Great.
1:14:30 [whistles]
1:14:37 [whistles]
1:14:40 We could do Champ. Champ's head.
1:14:46 Let's try Champ's head. We'll see what
1:14:47 that does.
1:14:50 Analyzing.
1:14:53 Turn reality into bricks.
1:14:56 Upload a photo to see the magic happen.
1:14:59 Well, they got to display the photo.
1:15:04 What is brickbuilder.ai?
1:15:07 This is the app that I'm building.
1:15:10 I just I saw someone say they built one
1:15:13 of these things and it was cool. So, I'm
1:15:15 trying to build I'm trying to vibe code
1:15:17 it. You upload a photo and it turns it
1:15:19 into a Lego object, a Lego sculpture,
1:15:23 and then shows you how to build it.
1:15:26 Gemini is analyzing your object. I don't
1:15:28 see my photo. So, this giving me no
1:15:31 visual feedback other than a spinning
1:15:33 [ __ ] wheel.
1:15:36 I don't like it, people. I don't like it
1:15:38 one bit.
1:15:50 Oh, what happened?
1:15:54 There's sort of a dog face. Look,
1:15:57 there's Champy. [laughter]
1:16:01 It made Champy. It worked.
1:16:07 That's very cute. [laughter]
1:16:11 It's Champy's face on a on a big globe
1:16:16 and it has steps. Step one. So, wait, if
1:16:20 I go back Oh, yeah. Look at that.
1:16:24 Unbelievable.
1:16:28 Holy [ __ ]
1:16:30 When Gemini got stuck on my last app, I
1:16:32 asked it to show me what it was doing,
1:16:34 and it did. Blown away.
1:16:38 This is amazing.
1:16:47 That's crazy.
1:16:51 78 bricks.
1:16:55 [laughter]
1:16:57 That's so good. Okay. So, wait. Let's go
1:16:59 back and then we'll say um Okay. So
1:17:02 we'll say this is okay
1:17:06 but you
1:17:09 never
1:17:10 showed
1:17:12 showed
1:17:14 the uploaded photo.
1:17:18 You also
1:17:20 have
1:17:22 a really
1:17:24 boring
1:17:28 progress indicator
1:17:35 that is a circle.
1:17:40 But imagine
1:17:42 a progress
1:17:46 animation that is Legos being built.
1:17:54 Um
1:18:06 well, let's just do those two things.
1:18:08 All right. And the the interface is for
1:18:11 horseshit.
1:18:13 Do you need to give
1:18:26 Yeah.
1:18:29 Yeah. It should also have a a Lego brick
1:18:32 inventory, shouldn't it?
1:18:34 Cuz it knows. I I mean, but honest to
1:18:37 God, not for nothing people,
1:18:41 that thing just took an image that was
1:18:44 really not a perfect image for this kind
1:18:47 of thing. Like it it was a disembodied
1:18:49 dog head on a on a disco globe
1:18:55 and it turned it into a 3D Lego object
1:18:58 that I could spin around and it had a
1:19:00 slider where I could deconstruct it.
1:19:02 That's pretty crazy. [laughter]
1:19:06 That's pretty [ __ ] crazy.
1:19:10 Looks like a lemur does.
1:19:16 I will add state to track the uploaded
1:19:18 image URL. Pass it to the controls
1:19:20 component. I replace the generic loading
1:19:23 spinner with a Lego loader component.
1:19:35 analyzing one error.
1:19:40 Are we having fun tonight? Hey, I was
1:19:43 wondering. Yeah. Hey, listen. I was
1:19:46 wondering. Are you having fun tonight?
1:19:50 Are you having fun? You having fun?
1:19:54 No. Yeah. No. May maybe
1:19:58 your favorite part of the night was the
1:20:00 nachos.
1:20:02 Okay. No, that's that's totally fair. I
1:20:04 think that's apparently you make very
1:20:06 excellent nachos. [laughter]
1:20:10 I don't feel less than. I don't feel
1:20:11 less than. Why would I feel less than?
1:20:13 That's not My nachos got cold.
1:20:16 [laughter]
1:20:20 Okay, we're going to upload a photo
1:20:22 here. Let's see.
1:20:25 All right, we'll do her. See, I don't
1:20:27 have any good objects, do I?
1:20:31 Oh, we can do mini figs. Maybe this will
1:20:33 figure out mini figs.
1:20:36 Oh, look. Look. We have a much more
1:20:38 creative
1:20:39 progress bar now, don't we? And it says
1:20:42 constructing. And there's our picture
1:20:44 preview.
1:20:47 I opted for sushi. That's pretty cool.
1:20:50 Oh, are you building along? This is
1:20:51 cool. I want nachos now. I know. I I
1:20:55 always talk about nachos and hot
1:20:56 pockets. I have never had nachos or hot
1:20:59 pockets here on the live.
1:21:07 Like like this is crazy, man. I just
1:21:10 talked this thing into existence. I
1:21:12 still don't
1:21:15 I don't know, man.
1:21:22 I hope there's someone in here goes,
1:21:24 "Oh, that's just a relatively uh
1:21:26 straightforward uh NodeJS 3D uh object.
1:21:30 There's there's nothing really
1:21:31 interesting going on here technically.
1:21:33 Uh you know, if if you know what I'm
1:21:35 saying, you know,
1:21:40 how do I
1:21:42 There we go. [laughter]
1:21:48 Well, it's got a little disembodied
1:21:49 banana there,
1:21:51 [laughter]
1:21:53 but still.
1:21:56 Oh, you can't you can't pan the photo.
1:21:58 So, doesn't display the whole photo, but
1:22:01 like there's there's this cyber glasses
1:22:03 and it's 13 steps,
1:22:10 you know? So, there's there's some work
1:22:11 to do there. You got to tell it it's got
1:22:13 to have some some basis in reality.
1:22:16 Like, you know, floating objects,
1:22:19 objects floating in space are not okay.
1:22:22 [laughter]
1:22:29 Look at this [ __ ] VR goggles. That's
1:22:32 pretty cool. [laughter]
1:22:36 [clears throat]
1:22:41 Oh, you know what else I should be able
1:22:42 to do? I should be able to add a a
1:22:44 prompt to this where I can fix it. Like
1:22:48 I could say, you know, turn these other
1:22:50 people into mini figs and have him on
1:22:53 like a banana foundation,
1:22:56 a banana pad,
1:22:59 right?
1:23:03 All right, there you have it, people.
1:23:07 That's it. That's it. I give up. I think
1:23:11 that was pretty good. I mean, for two
1:23:14 prompts,
1:23:15 having a functioning thing that
1:23:17 interprets a photo,
1:23:20 designs and builds a 3D Lego dude that
1:23:25 you can then deconstruct. And then Andy
1:23:28 and And what's his name? Brandon,
1:23:30 producer Brandon, his idea, put in like
1:23:33 an inventory list.
1:23:37 Crazy.
1:23:39 Crazy. The other thing you could do with
1:23:41 this is you could Oh, you know what
1:23:42 would be cool to add to this is a is a
1:23:45 complexity slider,
1:23:47 right? Like this one is a relatively
1:23:49 simple one. It's only using 29 bricks,
1:23:52 but you could say use up to 500 bricks
1:23:55 and it would make a really detailed Lego
1:23:57 model. That would be pretty cool.
1:24:01 Okay. Um,
1:24:04 if you haven't done it, go to
1:24:05 aifestivous.com.
1:24:09 AI festivist feign
1:24:17 up. Sign up. and then and block out
1:24:21 December 26th and de December 27th. Tell
1:24:23 your family they're either going to hang
1:24:25 out with you and learn about AI or
1:24:27 you're going to go do a work thing
1:24:28 without them. Okay? You're going to come
1:24:31 hang with hang out with us.
1:24:33 So that's then also
1:24:37 if you have not joined the AI salon
1:24:39 mastermind, what the [ __ ] are you doing?
1:24:41 Come on, get with it. [laughter]
1:24:44 I know I should be zen because we're
1:24:46 talking about designing a daily AI
1:24:48 practice, a practice around AI.
1:24:50 Um,
1:24:53 it's a really powerful process so far.
1:24:55 Like, it's already changing how I'm
1:24:57 working. Um, and you should join it. And
1:25:00 so, from now through the end of the
1:25:02 year, it's just 20 bucks a month.
1:25:04 January 1st, that goes up to 47 bucks a
1:25:06 month. So, now's a good time to do it.
1:25:09 Every Thursday at noon Eastern is the
1:25:13 mastermind practice lab where we get
1:25:14 together and work on designing our
1:25:16 practices. Okay. Um weekend homework. If
1:25:20 you have not vibe coded something, I
1:25:22 want you to vibe code something in
1:25:24 either Lovable
1:25:26 or in Gemini. And how you get to the
1:25:28 Gemini vibecoder thing is you go to um a
1:25:33 studio.google.com/app
1:25:35 google.comappsapps
1:25:41 and that will take you to the vibe
1:25:43 coding builder where I just built this
1:25:44 thing.
1:25:47 I want you to build an app this weekend
1:25:49 and then I want you playing with nano
1:25:51 banana.
1:25:53 I want and I want you throwing
1:25:55 complicated [ __ ] at it. Like upload
1:25:58 upload a spreadsheet
1:26:03 and
1:26:04 tell it to make you an infographic about
1:26:06 the data in that spreadsheet. Or upload
1:26:09 a video and tell it to analyze the
1:26:12 frames of the video. not the dialogue,
1:26:14 not the transcript, but the frames of
1:26:16 the video and have it make you an
1:26:20 infographic
1:26:22 that synopsizes the visual components of
1:26:25 the video. Something wild and weird. Do
1:26:28 something weird with image generation
1:26:32 that should not be possible
1:26:37 because it is now. But we don't, none of
1:26:39 us know what that means to be able to
1:26:42 have a thing that can think about
1:26:47 the image it's about to construct
1:26:50 and then construct it.
1:26:53 It's weird.
1:26:55 All right, so that's your weekend
1:26:56 homework.
1:26:58 Gemini
1:27:01 vibe coding in Gemini
1:27:04 making images in nano banana. And I'll
1:27:06 give you an option.
1:27:09 You could also just go to Notebook LM.
1:27:12 They just added um infographics and
1:27:15 slidemakers to notebook LM using this
1:27:18 new nano banana thing.
1:27:22 All right. Beautiful Tik Tok pin. Hey
1:27:26 Kyle, you were right. My Armenian poet
1:27:28 keeps sending me poems for Sunno. Yeah,
1:27:31 I think you [laughter] I think I think
1:27:34 your Armenian poet digs it.
1:27:36 Okay, Dr. Jay, here's the next here's
1:27:39 the next uh
1:27:41 here's the next evolution of that
1:27:43 project.
1:27:45 Turn that into um
1:27:49 like a Spotify album
1:27:53 and and and like get some press about
1:27:56 it, get some visibility for it. And then
1:27:58 and then maybe what you even do is do
1:28:00 like a you know what would be fun to do,
1:28:02 Dr. J is a uh a video contest where
1:28:07 where people
1:28:09 um make videos of those songs and you do
1:28:12 it as a contest so you get a bunch in a
1:28:14 row and then basically because you know
1:28:17 what you could vibe code up. This would
1:28:18 be great. You get all the songs on
1:28:20 Spotify and then you do a video contest
1:28:23 where people can go adopt a song, adopt
1:28:26 a poem, adopt a song, right? And if I
1:28:30 adopt the thing, then I'm then I've got
1:28:32 to go make a video of that thing. So I I
1:28:34 basically check it out of the library.
1:28:36 No one else can make a video until I
1:28:38 upload my video to it. I like lock it
1:28:40 and you give people three days to make
1:28:42 it or something like that. That'd be
1:28:44 kind of cool. Can you give me the URL
1:28:46 for the Vibe Coding with Gemini? Yes. Um
1:28:49 aistudio.google.com
1:28:52 google.com
1:28:56 and then you can kick click on the build
1:28:57 tab or you can just do slashapps aps
1:29:02 aistudio.google.comapps.
1:29:07 Also you can just do ai.dev. That might
1:29:11 be the easier way.
1:29:13 Just type in ai.dev. ai.dev dev
1:29:19 takes you to the the builder.
1:29:24 Much easier. AI.dev. Thanks, Vicki.
1:29:31 Much easier.
1:29:33 But that's your homework. Go build some
1:29:34 [ __ ] Go experience what the difference
1:29:38 in these models is if you vibe coded
1:29:40 before.
1:29:42 like just the level of sophistication of
1:29:45 what's being built with Gemini 3 versus
1:29:48 what what has been built historically
1:29:50 with anything I've done. It's just of a
1:29:52 different level.
1:29:54 Um, we should all experience that. All
1:29:56 right.
1:29:59 Beautiful, beautiful, beautiful,
1:30:01 beautiful, beautiful, beautiful people.
1:30:03 Beautiful. Have a fantastic weekend.
1:30:06 Have a fantastic weekend everybody.
1:30:11 You know, there's two ways you could do
1:30:14 this weekend thing. You could have an
1:30:16 all right weekend.
1:30:19 You could have a mildly depressing
1:30:20 weekend.
1:30:22 You could have a weekend where, I don't
1:30:24 know, didn't do anything. Or you could
1:30:26 have a fantastic weekend.
1:30:30 That's a choice you get to make. Okay.
1:30:33 All right. Have a fantastic weekend.
1:30:37 I don't know what that character was.
1:30:39 All right. later. Bye.