3/24/2026 - Why OpenAI Just Killed Sora and What It Signals About Their Next Big Move

Description

Chapters

Transcript

0:06 Uh, you know what?
0:12 Donald Trump sounds like he's what?
0:34 Hm.
0:36 Yo, what's happening good people?
0:39 How is everybody?
0:41 Coming at you live from Orlando. Here,
0:43 let's do this.
0:45 We will go.
0:48 And we will go live
0:52 from
0:53 Mickey's house.
0:57 Orlando.
1:00 FL.
1:03 Save. Boom. Look at that.
1:07 Look at that. Look at me all in here
1:09 with my my soft
1:11 pastelly kind of blues.
1:14 Very 2015.
1:20 What's happening good people?
1:22 What is going down?
1:25 I'll tell you what's going down. Sora is
1:27 going down.
1:30 Sora is going down. Down, down, down,
1:34 down.
1:35 Um, let's see. Spin B3's in the house.
1:39 What's happening?
1:46 Can I do something about my shirt? Um.
1:51 I could change it.
1:54 Let me see. Hang on.
2:06 Hey Kyle.
2:07 Yeah.
2:08 Just so you know, you are still on
2:10 camera on your Tik Tok.
2:13 How? Oh, cuz of my mirror?
2:15 Yeah.
2:17 No No good a free show tonight.
2:26 Free show is for everyone.
2:32 Kyle, you can get banned on uh
2:35 On all the platforms.
2:38 All right.
2:41 Okay, let's see.
2:44 It's been pretty
2:46 thought Apple Watch was the shock
2:48 collar.
2:50 Oh, that's funny.
2:55 What's happening, good people?
2:58 What is happening?
3:00 Um
3:02 What are we going to talk about tonight?
3:04 So,
3:06 wild news
3:08 is that um
3:12 OpenAI is killing
3:14 Dang you, Brandon.
3:18 Oh, they They want to see me strip.
3:25 It's getting saucy in this hotel room.
3:26 You have no idea.
3:28 You have no idea.
3:29 It is It is debauchery central.
3:33 Um
3:37 I deleted my Tik Tok. Well, this is why
3:39 we're doing We're doing both the Tik Tok
3:42 and the um
3:44 and the
3:47 YouTube. Hang on. It looks like one of
3:49 my cameras have got schmutz on them.
3:51 Let me de-schmutzify. on. Hold Hold,
3:54 please.
3:56 Everybody calm down.
4:00 Everybody calm down. Hey, you want to
4:01 see something funny?
4:03 Sure, Kyle.
4:05 What do you find hilarious tonight? So,
4:07 I get into my hotel or so So, I'm at a
4:10 I'm at some
4:12 big-ass resort in Orlando.
4:18 And uh
4:20 What's happening?
4:21 Um
4:27 Okay, I'm just cracking up now. So, I
4:29 get into my room, and you know what's in
4:30 hotel rooms, right? It's usually a
4:32 Bible.
4:33 So, I walk into into my hotel room this
4:35 time, and I got this book,
4:38 Stop Overthinking.
4:41 I got Practicing Mindfulness.
4:46 Apparently, they got wind of the AI
4:48 Salon Mastermind Practice Lab,
4:51 and preceded my little my little
4:54 office away from home, and then I got
4:57 Atomic Habits.
5:01 Yeah, exactly. And he sent the advanced
5:03 team.
5:04 I don't know how all these I don't know
5:06 how all these books got in my office in
5:07 my in my hotel room, but they did.
5:10 Let's see.
5:14 Oh, for your enjoyment. If you wish wish
5:16 to purchase, $10 a book. I'll tell you
5:19 what.
5:21 That's some That's some actually clever
5:23 marketing. You know what? This is This
5:24 is a marketing ploy.
5:27 I thought it was just I actually thought
5:28 someone probably just left them here.
5:30 It's a marketing ploy. You can read
5:32 them,
5:33 and you can buy them. Yeah,
5:35 who knew?
5:37 Wild.
5:39 Big Brother is always listening. Do you
5:41 have AI glasses yet? Yes, these these
5:44 are uh
5:45 my eyeballs. It goes into this large
5:48 language model up here.
5:50 Quickly view books. Atomic Habits is
5:52 interesting. I listened to it on
5:53 Audible.
5:55 Not going to lie, kind of bummed about
5:56 Sora. I know, same same Daniel. So,
5:59 couple of things about Sora. So, here's
6:01 here's what I know so far.
6:05 It was the Sora team. It's great
6:07 marketing, isn't it? It's really smart
6:08 marketing. And just a little bookmark,
6:10 the little bookmark is the only thing
6:13 that sort of clues you in. So, if you're
6:15 illiterate like me, and I'm not really a
6:17 reader. I I and to say not really a
6:20 reader is doing a disservice to
6:22 non-readers. I don't read. I've read one
6:25 book in the last
6:27 I don't know, 5 years. I shouldn't admit
6:29 that.
6:30 And it was one that was assigned to me.
6:33 It was a good book.
6:34 I'm a professional now because of it.
6:37 But anyway,
6:39 um
6:41 That was probably dangerous. You know,
6:42 an eyeball cut, a paper cut on your
6:44 eyeball? Probably not pleasant.
6:48 Anyway, that's the only piece of
6:49 marketing here.
6:50 Okay, Sora.
6:52 So, if you don't know, Sora was OpenAI's
6:55 um video model. And um
7:00 when it was
7:03 when it was announced, it was by far
7:07 the best video model in the world. And
7:09 then they released Sora 1, which was
7:11 like a hobbled version of what they
7:13 showed.
7:15 And then it took them, I don't know, 2
7:16 and 1/2 years to get to Sora 2, or a
7:19 year and a half to get to Sora 2.
7:21 And then they launched it to great
7:23 fanfare, and they launched it with this
7:25 app, the Sora app,
7:27 which is like a little social network
7:29 where you can make fake videos of like
7:31 you and Sam Altman. I made a bunch of
7:33 those.
7:34 And
7:35 um
7:37 And so, they announced today that it's
7:39 going away.
7:41 So,
7:42 an X post that I saw this evening, which
7:45 if it's on X, you know it's true, right?
7:47 Nothing on there's
7:49 um false. So, cuz it's it's
7:52 maximally seeking truth. That's what
7:54 Elon Musk told us, so that must be true.
7:57 So,
7:58 um wait a second, you're early. I am an
8:00 hour early. I decided to go an hour
8:02 early. I'm on the East Coast. I figured
8:04 for the East Coasters, I'll give you a
8:06 little gift. I'll be a little less
8:08 ridiculous for me.
8:10 You know, like it's always ridiculous
8:11 for us. I know. I know.
8:14 For those of you on the East Coast,
8:16 I'm sorry. I'm a dick.
8:18 What can I say?
8:20 When I decided to go live at 8:00 at
8:22 night, uh
8:24 I didn't have a sense that I was in a
8:25 country with
8:28 other people in it.
8:31 So, a news report about Sora shutting
8:33 down, yeah. They want to concentrate on
8:35 real world things. Well,
8:37 so a couple of things.
8:40 Apparently, it was costing OpenAI 10 to
8:44 15 million dollars a day to run Sora,
8:48 which is in the billions for a year. I
8:50 don't know, someone can do the math on
8:51 that, but that that can't be pleasant.
8:54 Well, three 3.6 billion
8:56 if it's 10 10 million a day.
9:01 Jesus.
9:03 The [ __ ] numbers with AI are
9:04 staggering.
9:06 Um so, that was their little side
9:08 experiment. Sora was costing them 3.6
9:10 billion dollars a year or whatever.
9:13 Um so, they're shutting it down. They
9:15 need the the uh
9:18 the GPUs for training and for inference.
9:21 Apparently,
9:22 they have trained a new model that is
9:25 supposed to be quite good. Yeah, and I'm
9:27 going to talk about that, too, Brandon.
9:29 They've trained a new model that's going
9:30 to be quite good.
9:32 Um
9:33 as as we would assume. Now,
9:36 to be clear,
9:37 OpenAI completely lost the plot with
9:40 GPT-5. I think the GPT-5 series sucks.
9:43 Like it's probably fine for coding. I
9:45 think GPT 5.4 with Codex
9:49 um engineers seem to be really pleased
9:51 with it. Anyone that I know that likes
9:53 words
9:55 hates it.
9:57 So if you're a fan of the words,
10:01 the GPT-5 series was an abject disaster.
10:05 Maybe GPT-6 will be better. Who knows?
10:09 Apparently what they're going to focus
10:11 on is
10:13 creating an Uber
10:16 GPT experience that includes chat,
10:20 Codex, the programming tool,
10:23 um and a browser, the Atlas browser. So
10:26 so they're apparently creating some sort
10:28 of Uber kind of tool.
10:31 And then they also have Peter
10:32 Steinberger working for them. He's the
10:34 one that created Open Claw, which Open
10:36 Claw,
10:38 if you don't know,
10:39 that's the little happy agents that do
10:41 all the work for you when you're not
10:43 when you're sleeping,
10:44 unless
10:46 you're me,
10:47 and then
10:49 I'm not going to talk about Adam.
10:51 But but but apparently other people
10:54 digging Open Claw.
10:55 Um
10:57 so I would think that Codex and Open
10:59 Claw, like this new product, probably
11:03 has agents, web browser, a programming
11:06 interface, and then all the chat GPT
11:08 stuff we know and love.
11:11 Um
11:13 they're also Sam Altman also on April
11:17 I forget what date it is
11:20 is making an announcement about
11:23 um
11:24 world.org. So if you don't know
11:26 world.org, they're the ones that created
11:29 world.id, which is the human verified
11:32 blockchain-based thing that's got a
11:34 crypto coin associated with it.
11:37 Um
11:39 if you're if you're using LinkedIn or if
11:41 you're using X right now, I called this
11:43 one when
11:45 when did I get my world ID? I got it in
11:47 2023, I think, in August of 2023.
11:50 It might have been 24, I don't know. A
11:52 while ago.
11:53 I got it a while ago and what I said
11:55 was, I'm I'm going to get this just to
11:57 get it, just to understand the
11:59 experience and then most people yelled
12:01 at me, "Don't give them your
12:04 Don't give them your biometric data,
12:05 Kyle. That's really not very wise."
12:09 Um
12:13 It was before you written the paper cut
12:15 tonight.
12:18 Um
12:20 So, I got it a while ago and what I said
12:21 was, there's going to be a time
12:25 where
12:26 the AI robots are going to get so
12:29 prolific that
12:32 social networks are going to be
12:34 unusable.
12:35 And then there's going to be a demand, a
12:37 cry for human verified stuff.
12:41 So, a possibility
12:44 is that OpenAI launches a humans-only
12:47 social network, which would be kind of
12:50 smart.
12:53 You have a receipt. I do have a receipt,
12:55 as a matter of fact. Hang on. For those
12:57 of you on TikTok, I'm going to go find
12:58 out my world ID uh
13:01 time.
13:14 July 2023.
13:16 That was a while ago. Damn it.
13:19 I must be
13:22 psychic. July 2023.
13:25 July 2023 is when we uh
13:29 when we got ye old world ID. Um so
13:31 apparently there's a world ID thing
13:33 coming.
13:34 There's a new model from Open AI coming.
13:37 There's some sort of unified something
13:39 coming.
13:42 They now employ the guy that wrote Open
13:44 Claw. So, it it looks like what they're
13:47 doing with Open Claw is they're just
13:49 building a bunch of hooks into Open
13:51 Claw. Open Claw's remaining open source.
13:54 And then the guy that wrote it is
13:56 working with all the hook developers
13:57 within
13:59 within Open AI to connect [ __ ] So,
14:03 So, a couple of things.
14:07 I think this is the year
14:09 where
14:11 social media's going to become [ __ ]
14:13 useless.
14:14 Like X X at this point is useless. Like
14:18 it's just useless. It's
14:21 And LinkedIn?
14:23 Everyone's commenting with bots. Wait,
14:27 like there's no
14:28 you know, TikTok pin.
14:30 Pika has another new app called pika.me
14:32 like virtual you. It keeps iMessaging.
14:36 That's interesting. Oh, keeps iMessaging
14:39 you?
14:41 I would be scared of an AI me. I'm
14:42 pretty sure that's how Skynet happens.
14:45 Other news,
14:47 Hachette cancels book Shy Girl
14:50 because it was rumored to have been AI
14:53 written.
15:03 I'm just [ __ ] exhausted.
15:07 Did you see all the people marching to
15:09 to Open AI headquarters? Well, wasn't
15:10 that many of them. It was like
15:13 fif- 100 people.
15:15 150 people, I don't know. Some amount of
15:17 people. They had all their signs, "Stop
15:18 AI. Slow down AI."
15:21 AI's not slowing down.
15:24 AI's not slowing down.
15:26 Writers are going to use AI.
15:31 And you know what's going to happen?
15:33 Lazy writers
15:36 are going to use AI lazily
15:39 and they're going to create crap.
15:41 And you know who's going to read the
15:42 crap?
15:44 No one. No, that's not true. Agents.
15:46 Agents are going to read the crap. No
15:48 one's going to read the crap.
15:53 Artists
15:57 writers, conceivers
16:01 creators, filmmakers
16:03 poets
16:06 They're going to find ways to use these
16:08 tools that are novel and interesting and
16:10 exciting.
16:15 And everyone, every single person that I
16:18 know that's using AI to any
16:26 quality degree
16:30 is not just pushing a button and out
16:31 squirts a novel.
16:37 Yeah, Windrider.
16:40 I personally don't care who writes a
16:41 book so long as it's good.
16:43 Or or more accurately, I don't care what
16:46 tool was used to write the book so long
16:48 as it's good.
16:50 A piece of work either moves you or it
16:52 doesn't. The person that created it is
16:54 either willing to put their name on it
16:55 or they're not.
16:58 They used the tool
17:00 who gives a [ __ ] Yeah, it's not really
17:02 your writing.
17:09 I'm exhausted.
17:11 Okay, anyway, back to Sora.
17:14 Um
17:19 Jensen Huang
17:22 yesterday. Thank you, Wolf Van Clinton,
17:25 on TikTok, for the gifts.
17:28 Um
17:30 casually said in an interview, I think
17:32 we've hit AGI.
17:34 Dario Amodei from Anthropic, in the past
17:36 month, has been weird. He's already
17:38 weird, but he's been weirder.
17:42 And he's saying things like, "You know
17:44 that article I wrote on the age of
17:46 spiritual machines? We're pretty close
17:48 to that." That That was him describing
17:50 AGI.
17:52 So, I think Anthropic's probably got it.
17:55 Um
18:01 I don't know if OpenAI's got it. Their
18:03 Their GPT-5 models suck, but they're
18:06 powerful.
18:08 Um
18:10 So, I would think that they've got it.
18:12 Elon Musk
18:15 claims to have it or claims to be close
18:18 to it.
18:20 So,
18:22 you know,
18:26 it looks like that stuff's coming. And
18:28 And when you look at something like
18:30 OpenAI's going to take this thing that
18:32 they
18:33 publicly,
18:35 for 2 years, said this Sora video
18:37 thing's coming. We're going to launch a
18:39 social app. We're going to
18:42 take over Hollywood. We're going to do
18:43 all this stuff.
18:45 They cut a deal with Disney 3 months
18:47 ago, 2 months ago, 3 months ago. I don't
18:50 know, some months ago.
18:53 Where Disney
18:55 This is This is remarkable. This
18:56 happened in the early days of the
18:57 internet, too.
18:59 Normally, when a company like Disney
19:01 that's got valuable properties cuts a
19:03 deal with someone like OpenAI,
19:06 they charge OpenAI for the right
19:10 to use their characters.
19:13 Disney said, "Hey Sam,
19:16 we'll pay you a billion dollars if you
19:19 can stick our little characters in your
19:21 little movie-making thing there."
19:23 And Sam said, "Okay, give me that
19:25 check."
19:27 Well, Disney's backing out of that deal
19:30 cuz Sora is going away.
19:33 So,
19:35 if you've got a billion dollars in
19:37 investment from Disney
19:40 for this tool
19:42 that, you know,
19:43 everyone could have made their own
19:45 Disney film,
19:46 you know, in however many a year, year
19:48 and a half,
19:50 and you flush that down the toilet, that
19:53 says to me that something big's coming.
19:58 Um
20:00 the other thing that's going on with
20:02 OpenAI, it looks like they're they're
20:05 Sam said he's going to his focus is
20:07 going to be on fundraising.
20:09 So, it looks like he's out seeking
20:10 another 10 billion dollars at a 750
20:14 billion-dollar valuation.
20:18 So, they're going to be a
20:19 trillion-dollar company
20:21 here here shortly.
20:25 Just just [ __ ] mind-boggling. So,
20:28 apparently Sam's going to be focusing on
20:30 um
20:31 on fundraising and
20:35 securing
20:36 uh data centers energy, energy for data
20:39 centers.
20:43 So,
20:45 what's that mean?
20:57 I'm a little punchy.
20:58 I don't know if I didn't sleep enough or
21:00 what it was.
21:05 What do you think is their interface to
21:07 each AGI more texting?
21:09 No. Well, so you know what's funny Mike
21:11 Diamond's the the um
21:17 one of the one of the tweets that I saw
21:19 said that the Sora team is going to be
21:21 focused on world models. So um
21:28 Runway ML is is is working on world
21:31 models.
21:33 Midjourney is working on world models.
21:36 I would be dramatically surprised if
21:40 what OpenAI learned with Sora isn't
21:43 being flipped into a world model. So so
21:46 that that kind of
21:48 rings true that that they would be going
21:50 there.
21:51 um
21:55 No, I I I don't think text is going to
21:57 be the primary interface. I think it's
21:59 going to be world models. I think it's
22:00 going to be
22:07 We're going to kind of if if you're old
22:09 if if you're on this channel you're
22:10 probably old. And if you're old you
22:12 probably remember Prodigy. Remember
22:14 Prodigy?
22:16 Prodigy was Sears's AOL.
22:20 Was it Sears? I think it was Sears.
22:22 I think Sears owned Prodigy, right? Am I
22:25 Am I right? Am I or am I not remembering
22:27 that right? I might I might have that
22:28 wrong. But Prodigy
22:30 was this really janky ass
22:33 um dial-up
22:36 you know, chat
22:38 tool.
22:39 Was it Sears?
22:41 Co-owner, but yes, it was Sears. Okay,
22:43 good. I I'm like I'm like
22:47 am I that old that I that I'm am I
22:49 remembering that wrong? But the the
22:51 co-owner of Prodigy was Sears. And if
22:54 you're not old you're like, "What's
22:56 Sears?"
22:57 Sears was a store
22:59 where people that looked like me sold
23:01 you carpets and clothing.
23:07 I remember Prodigy. And Prodigy had this
23:10 really janky ass it was like bad
23:14 it it was like the font was really
23:16 jagged and Tik Tok build their world
23:19 model. Yeah, they are.
23:23 They are.
23:24 Um
23:26 and it was just it was it was just a
23:28 horrible horrible service. It was slow,
23:30 it was buggy,
23:32 it was janky.
23:34 And then AOL came along and then the web
23:36 came along and everything sort of
23:38 evolved out of the out of that
23:39 proprietary world. CompuServe was
23:41 another one that was really janky.
23:45 This era, these last 3 years, this era
23:48 of chat GPT with text as the primary
23:50 interface is going to feel like that.
23:52 We're going to look back on this time
23:54 and go, "Hey, remember when you used to
23:56 have to type in?"
23:57 And they were all wondering if it was
23:59 AGI.
24:01 But nobody knew and they kept going, "Do
24:03 we have AGI yet?" I don't know. And then
24:05 they kept trying to figure out what's
24:06 AGI anyway.
24:08 And then Katie Couric and Bryant Gumbel
24:10 were on TV going like, "What is AI?
24:14 What is internet? That A with the circle
24:17 around it."
24:22 We're We're living through history
24:24 again.
24:27 Um but I think this year is the year
24:30 that we see some credible
24:33 decent
24:34 world models.
24:36 Probably toward the end of the year.
24:39 And then I think you're going to start
24:40 to see
24:42 I think you're listing
24:45 all the ISPs of my late father's
24:46 friends.
24:51 That was That was what we did back then.
24:54 Government commercial break. The US DOL
24:56 is launching a new AI literacy course
24:58 designed to ensure that every American
25:00 worker can benefit from the
25:02 opportunities that AI economy presents.
25:05 Hmm.
25:08 To get started, simply text ready to
25:10 20202.
25:12 Is that a joke or is that real?
25:18 Very real.
25:22 We should
25:24 We should call Wait, what is it? We text
25:27 ready to 20202? I'm going to text it.
25:30 Tell TikTok I love them, but I'm I'm
25:32 going to text something.
25:34 Text ready. Okay, two
25:36 20202
25:40 ready.
25:42 I wonder if they know
25:44 which AI they're talking. Thanks for
25:46 enrolling in AI 101. Look out for your
25:48 first lesson soon. Okay. Wait, let me go
25:51 back to TikTok. Sorry, TikTok. I had to
25:54 uh
25:54 I had to text the government that I'm
25:57 ready to learn AI.
25:59 Hey Hey, Source Camp, you don't need to
26:02 teach anyone anymore. The government's
26:04 got it handled.
26:11 Oh, here you are. Couldn't figure out
26:12 who you were talking to.
26:15 Hello.
26:19 Oh my god. Oh my god. Oh my god. It's
26:23 just so crazy. It's so crazy. You know,
26:27 it's crazy. It's just crazy.
26:30 It's crazy. Hey, TikTok. It's crazy. Oh,
26:34 wait. I got a text. Hold TikTok, please.
26:37 Okay.
26:39 Let's see.
26:41 AI make
26:47 Okay, we we got to download this
26:49 graphic. Save.
26:53 Everybody's got to see this graphic.
27:08 Listen,
27:09 I should not I should not be making fun
27:12 of our government. I really shouldn't,
27:15 but I'm going to.
27:20 Cuz someone made this graphic and I feel
27:23 bad for them. I'm about to mock them.
27:28 Oh, it's so bad. Especially on TikTok. I
27:31 got to go back to TikTok, sorry.
27:33 Okay, I'm back. Calm down. Just
27:35 Calm down. I changed my shirt for you,
27:38 okay?
27:39 So, if you if you get some pauses
27:41 tonight while I go get my AI 101 classes
27:45 from the government.
27:52 Okay.
27:53 Am I sharing? I think I'm sharing,
27:55 right? There we go. There we go. Make
27:57 America AI ready.
28:00 This
28:01 God, does this feel like 1997?
28:04 This This feels like 1997. Like these
28:07 kind of these kind of graphics,
28:10 you know, learn the world wide web.
28:19 No music on hold?
28:21 That's better than Jeopardy calling.
28:23 Make America AI ready. All right. Well,
28:26 so wait. I'm going to go back to uh
28:29 Wait, I can't see my screen anymore.
28:32 Okay.
28:34 I'm going to stop sharing there. I'm
28:36 going to go back and look at my texts.
28:37 Hold TikTok. I'll be back.
28:43 AI is changing how we work and live. Oh,
28:45 you know what I can do? I know what I
28:46 can do. Hang on.
28:59 Oh, I didn't I don't have it here.
29:02 Oh, that's a drag.
29:09 I should send my resume to the US
29:11 government.
29:14 That's a t-shirt. Can't wait to see how
29:16 much they get wrong. I know. Hang on.
29:20 Seven 7 days 10 minutes. Learn what AI
29:23 is, how it works. This is your AI 101
29:25 course. It's part of the Make America
29:26 Ready initiative, Department of Labor
29:29 in connect collaboration with education
29:32 technology partner Artist.
29:35 What?
29:57 Oh, you put you put it there? Okay.
30:00 Great. Let me Can I go grab that? Yes.
30:04 This is worth Can you grab the whole
30:06 thing, Brandon? This isn't all of it, is
30:08 it? No, it's not all of it.
30:10 Go grab the whole thing.
30:12 And I'll stick it in a in a post it in a
30:14 in a notes.
30:19 Oh, I might be able to do it here. Can I
30:20 do it here? Hang on.
30:24 Copy.
30:25 I got it. Never mind.
30:28 Okay.
30:29 Let's go.
30:31 Notes.
30:41 Okay. Here we go.
30:45 Um
30:46 TikTok, you'll just have to listen
30:48 along.
30:53 And then are we sharing? No, cuz I got
30:56 to share. Wait, stop there.
30:59 Start sharing there. Start doing this.
31:01 That to that. That to that.
31:06 Okay.
31:07 Lesson one, what is AI? AI is already
31:10 working for you. In fact, you probably
31:11 used AI before you finished your morning
31:13 coffee today and just didn't know it.
31:16 Google Maps quietly dodging the
31:18 nightmare in traffic jam? AI. Netflix
31:21 guessing the next series to binge? AI.
31:24 See, this is all
31:34 So, how does it work? It's a system that
31:36 looks at massive amounts of data, finds
31:37 patterns, and makes predictions. That's
31:39 it. That's the real secret. Patterns in,
31:41 predictions out, but it doesn't stop
31:43 there. Oh, okay. There's a newer kind of
31:45 AI called generative AI that doesn't
31:47 just predict, it creates. That's
31:49 actually that's good.
31:51 Give it a short instructions, aka a
31:53 prompt, and it can draft an email
31:55 explain confusing for This is This is
31:56 not bad.
31:58 Think less sci-fi plotting world
32:00 domination and more really smart digital
32:02 assistant that works at 2:00 a.m., never
32:04 judges you for asking the same question
32:06 three time, and never asked, "Did you
32:07 Google it first?"
32:09 When people say I AI today, they're
32:11 usually talking about generative AI. In
32:13 this course, we'll use AI broadly, but
32:15 our main focus will be generative AI.
32:18 That's at least competent.
32:22 Before we continue, let's hear from you.
32:24 How confident do you feel about your
32:26 understanding of it? Respond
32:29 with a zero to 10.
32:31 Wild.
32:33 Wild, wild, wild. Seeing this makes me
32:35 feel pretty secure about my teaching
32:37 gig. Yeah, it's it's okay, Kelly. We
32:39 don't We don't need your expertise
32:42 anymore. The government has it.
32:46 Make
32:47 America
32:49 What was it? AI
32:53 AI ready. Hey, AI readiness.
32:56 They stole my [ __ ]
32:59 Damn it.
33:04 Make America AI ready.
33:06 Do Do I get some sort of compensation
33:08 for that?
33:10 We've had AI readiness in the AI Salon
33:12 for
33:13 years now. Okay.
33:15 So, basically, we are the paid to the
33:17 government class.
33:20 I guess so.
33:22 Oh my god. Oh my god.
33:26 Um
33:27 So, the other thing I thought might be
33:28 fun
33:30 is to
33:38 just surf through the old Twitter.
33:41 Since I can't really build anything. I
33:44 mean, I could, but
33:48 By the way, my
33:50 um
33:55 I'm down here in Orlando on Storyvine
33:57 business. And then Friday, I fly out of
34:00 here
34:01 and I go to Boise, Idaho. And I'm
34:04 speaking at the
34:06 um
34:08 It's called
34:10 um Oh my god, my brain. I'm so tired.
34:12 It's called Hackfort.
34:14 There's this There's this sort of
34:17 Boise version of South by Southwest
34:19 called Treefort.
34:21 And then they created a a sort of
34:23 technology arm of it called Hackfort.
34:25 And I'm speaking about AI
34:28 to the artistic community in Boise on
34:30 Saturday. So, that's pretty cool.
34:34 I got invited there by this woman named
34:36 Kathleen Cohen. She's one of the women
34:38 that is hopefully
34:41 um helping to get Sydney, my musical, up
34:44 and running, mounted, produced.
34:46 So, that's something we're working on.
34:47 So, that's pretty cool.
34:50 Um
34:51 So, that's this weekend.
34:53 Uh all right, let's see what's going on
34:55 here. You all can see my Well, wait.
34:58 TikTok can't. So, TikTok, I'll just tell
35:00 you what I'm seeing. And if you have
35:01 questions, just pop them in there.
35:05 And then Brandon, if you've got stuff
35:07 you want me to look at, go to head and
35:09 go ahead and do that.
35:12 Volunteers,
35:14 CEO says there's only two kind of people
35:16 that will succeed.
35:18 Trade workers or your neurodivergent.
35:20 Neuro spicy for the win, baby.
35:24 There you go.
35:26 All right. So, so basically AI's going
35:29 to completely wipe out the middle class.
35:30 I've actually I've actually noticed some
35:33 some
35:34 things going on in Twitter that I think
35:36 is bots.
35:38 They'll screw it up and make it harder
35:40 and more expensive than it has to be.
35:43 Sora say goodbye to Sora.
35:47 Everyone is wrong about why Sora is
35:49 canceled. Sora 2 is the last gen tech. C
35:52 dance 2 is 10 times better.
35:54 And it will be rolled out in the US
35:56 within the next few days. Trained on
35:57 TikTok has much more video data than
35:59 anyone else on Earth. Open AI can't
36:01 compete.
36:03 We'll see.
36:05 Too many AI generated replies, too few
36:08 human ones. We will not accept the
36:10 unacceptable.
36:12 My speech at the stop AI race.
36:19 The fetus slop.
36:32 >> So this is kind of interesting. It's
36:33 like one of the things I just want to
36:35 make sure I'm not in a post.
36:38 One of the ways I look at signal, how I
36:40 separate signal from noise, and again, I
36:41 don't think this is going to last that
36:43 much longer,
36:44 is if I see the same story show up over
36:46 and over and over again in my feed, then
36:48 that generally means it resonated.
36:51 Everyone on X is acting like Sora's shut
36:54 down is the end of AI film making.
36:58 It's not.
37:00 OpenAI's going public. Oh, that was the
37:02 other thing I was going to talk about.
37:03 What OpenAI is likely doing is they're
37:06 likely going public. So,
37:11 OpenAI's going public. That's the entire
37:13 story. When you're preparing for an IPO,
37:15 you go through every product and you ask
37:17 whether it makes you look like a serious
37:19 company or whether you look like you're
37:22 running a deepfake factory for Disney
37:24 fan edits.
37:26 What?
37:27 No. No.
37:30 Oh good lord. All right, whatever. I'm
37:32 not going to read that garbage.
37:34 Marcus Pittman.
37:39 All right, CERN scientists transported
37:41 antiprotons, a form of antimatter, by
37:43 truck
37:45 across their campus.
37:48 If the If the opposites come into
37:50 contact, they will annihilate each
37:52 other, setting off lots of energy
37:54 depending on the mass. All right.
37:58 All sorts of weird science stuff
37:59 happening.
38:02 Did you see Did you see
38:05 Here's a good one. Wait, let me go back
38:06 to
38:07 Let me go back to just me.
38:09 Um
38:10 I think this is real, but it might not
38:12 be. So, if there any physicists in the
38:15 house,
38:16 I saw a story today that they did some
38:19 analysis on light. They split it, they
38:21 trapped it.
38:23 And instead of light having two
38:25 dimensions, which is what they thought,
38:26 light actually has 48 dimensions.
38:31 Anybody Can anyone confirm or deny that
38:34 or Google it or chat GPT it?
38:38 I'm glad you asked about the 48
38:40 dimensions of light, Kyle.
38:45 Son of a [ __ ]
38:49 48 my way sides, I love it.
38:52 48.5. Oh, there's 48 and a half
38:56 dimensions of light.
38:58 Not two.
39:01 Here's what I know.
39:03 I know I'm not smart enough to know what
39:05 that means, but I am smart enough to
39:07 remember that.
39:10 And then the next time I'm at some party
39:11 and someone says something about
39:13 physics, I'll go, "You know,
39:16 scientists trapped light and expected to
39:19 find two dimensions and they found
39:21 48.5."
39:26 And then the physicist will be like,
39:27 "Uh-huh." And they'll start talking
39:29 physicist stuff and then I will slowly
39:32 back
39:34 out of the conversation and go get some
39:35 more cheese dip.
39:40 Unstable connection. Oh boy, that's not
39:43 good.
39:44 Oh, you just made the half part up.
39:46 Okay, see, there's our This is a channel
39:48 for misinformation. So, not only is the
39:50 48 dimensions possibly not true, but the
39:54 half was just a joke that someone on
39:56 TikTok gave me that I absorbed as fact.
40:01 I'm the unstable part.
40:04 Love your camera, so clear.
40:07 Uh it's just good light. Yeah, it's
40:08 clear it's clear tonight.
40:12 Okay, let's go let's go back to uh
40:15 Let's go back to it, shall we?
40:19 Uh-oh, hang on. That's going to be
40:21 driving.
40:28 AI filmmakers of the future
40:37 Ah. This is interesting.
40:40 Holy [ __ ] LeCun's team just cracked
40:44 world models wide open. So, Yann LeCun
40:46 was a chief scientist at Meta until
40:49 Zuckerberg pissed him off and they
40:52 [ __ ] all their AI up. So, he left to
40:55 start a world model company.
40:58 Everyone's obsessed with the next Claude
41:00 update. Meanwhile, Yann LeCun quietly
41:02 dropped a paper that could matter way
41:05 more long-term. It's called Le World
41:07 Model.
41:08 And to understand why it's a big deal,
41:11 I'm not going to
41:14 Let's go. Why is there no Hang on. We
41:17 need to get the
41:21 the link. Here we go. Le World Model.
41:23 All right.
41:26 So, let's go HTML.
41:31 Yeah, perfect.
41:34 All right. Copy that. Oh, I've got to
41:37 share. Hang on. So, I just went to
41:40 archive to get this this world model.
41:45 Let's go to Claude. [ __ ] ChatGPT.
41:47 ChatGPT sucks lately.
41:56 One week left for two times more usage
41:58 outside the hours of 5:00 to 11:00 a.m.
42:01 Pacific time.
42:02 All right, whatever.
42:04 Okay, so there's that paper. So, uh
42:07 please
42:09 read
42:11 this paper and tell me what it means
42:18 for the future of world models.
42:25 Claude is now a cell phone company. No
42:28 [ __ ]
42:30 Actually, you know what? Tokens are the
42:32 new minutes.
42:33 Remember when they sold you minutes? At
42:35 T-Mobile, you can have unlimited minutes
42:38 on
42:39 weeknights between
42:41 nights and weekends are free, exactly.
42:44 Let me fetch that paper for you. This
42:46 paper is from NYU
42:49 not Jan look
42:51 LaCoon's group.
42:55 A new approach. Okay, the problem it
42:57 solves. Jeppa based world models have a
42:59 fundamental instability, representation
43:01 collapse, which means if you're facing
43:04 in this direction and then you turn
43:05 around and you turn back, everything's
43:07 different.
43:10 The model learns to map every
43:12 observation point
43:15 to the same point in latent space,
43:18 which trivially satisfies the predict
43:21 next state objective, but produces
43:23 useless representations.
43:25 Okay, I
43:27 uh
43:29 explain this as if I have a degree
43:37 in acting,
43:39 but like to act like I understand
43:44 science.
43:46 Do not
43:47 make theatrical
43:51 references.
43:53 I'm
43:54 indicating
43:56 a
43:57 level of intelligence.
44:08 Got it. Smart person, no technical
44:11 background. Here's the deal. The big
44:12 picture problem. AI researchers want to
44:14 build
44:15 systems that can imagine what happens
44:17 next before they act.
44:20 That way you mentally simulate catching
44:23 a glass before it falls off a table.
44:25 These are called world models. Instead
44:27 of reacting to every moment, the AI
44:30 builds an internal model of how the
44:31 world works and plans inside its own
44:34 head.
44:36 The catch. You could try to imagine
44:39 the future in full detail, every pixel
44:41 of every frame, like rendering a movie
44:43 in your mind, but that's incredibly
44:44 expensive and mostly wasteful.
44:47 Okay, the problem with the problem. When
44:49 you train a system to compress
44:52 what everyone else has done about it,
44:54 various combinations of duct tape. What
44:56 this paper does, they found a single
44:58 clean,
45:00 mathematically grounded fix.
45:03 They add one penalty term that
45:05 essentially says the compressed
45:06 representations must stay spread out in
45:09 a bell curve shape across all possible
45:12 directions.
45:14 They test this using a classic
45:15 statistics trick. Project the data under
45:19 random lines and check if each
45:21 projection looks like a normal
45:23 distribution. This is still too smart
45:25 for me to [ __ ] understand.
45:28 There's an actual theorem, the
45:29 Cramer-Wold theorem, that guarantees if
45:32 all the 1D projections are bell curves,
45:35 then the whole high-dimensional
45:36 distribution is too. No collapse
45:39 possible. That's it. One prediction
45:41 term, one anti-collapse collapse term,
45:44 one knob to turn down from seven terms
45:47 and six knobs in the previous best
45:50 approach.
45:56 Oh, yeah.
45:57 Time to go to fifth grader.
46:01 The model is tiny, but it means it's
46:04 early. Yeah, I was going to say, okay.
46:06 How long
46:08 before you think
46:11 this is
46:14 buildable?
46:16 Um look at Reddit and
46:20 X and tell me
46:23 your prediction.
46:29 So, that's pretty cool.
46:31 Um a lot of
46:33 how we use AI today is because of Yann
46:36 LeCun. He was he was
46:38 I didn't he's a he's a bit of an Eeyore.
46:40 He's a bit of a drag to listen to.
46:43 Um and he also he was really down on
46:46 large language models when he was at
46:48 Meta. Largely because their large
46:50 language models suck.
46:53 Okay, Kyle. The context you need. Paper
46:55 dropped 2 weeks ago.
47:04 Where's
47:06 That's still gone.
47:09 LeCun's startup AMI labs just raised 1.3
47:12 billion at a 3.5 billion valuation.
47:18 World labs
47:21 Wow, World labs closed a billion dollars
47:24 just weeks before that.
47:27 Two billion dollars in 3 weeks both
47:30 betting against LLMs. So, this was
47:33 Whoever asked the question on TikTok
47:34 earlier,
47:36 is text going to be the primary
47:37 interface for AGI? It's not.
47:40 World models are. Cuz once you
47:42 understand the world, you can just stick
47:44 it in a robot and it can go be in the
47:45 world cuz it understands worlds.
47:48 And we live in one.
47:52 Right? Where is he tonight? I'm in
47:54 Orlando, Florida. I agree. His dunking
47:57 on LLMs is tiring. Yeah, it's just
47:59 tiring. It's just shut the [ __ ] up, you
48:01 know? I'm like
48:02 I know you've been in it for a while and
48:04 you're cynical and you hate the fact
48:05 that Sam Altman was kicking your ass at
48:08 meta.
48:09 But, you know, lighten up. He's probably
48:12 lighter now that he's got this this cool
48:14 world model thing.
48:16 What the community is saying on X
48:18 archive or alpha 9 or XIV noted that one
48:24 of the biggest bottlenecks of Jeppa
48:27 is they're hard to train.
48:29 And this new research changes that.
48:33 Which is the right read.
48:35 Humanoids daily framed it as vindication
48:37 of LeCun's long-standing argument that
48:39 industry is LLM pilled
48:42 and heading toward a dead end for
48:44 physical intelligence.
48:46 The latent space AI newsletter captured
48:48 the broader sentiment well. Supporters
48:50 view it as an overdue next para-
48:52 paradigm.
48:54 All right.
48:57 Well, listen. Yann was the one that that
49:00 built a lot of the infrastructure that
49:02 led to chat GPT. So, the dude knows what
49:04 he's talking about.
49:09 LeCun himself, we are going to have AI
49:11 systems
49:12 that have human-like and human-level
49:14 intelligence, but they're not going to
49:16 be built on LLM's.
49:18 And it's not going to happen next year
49:20 or 2 years from now. It's going to take
49:21 a while.
49:22 There are major conceptual breakthroughs
49:24 that have to happen.
49:26 This paper's environment are flat 2D
49:29 grids and single object 3D scenes. The
49:32 real world has occlusion, deformable
49:35 objects, blah blah blah. The honest
49:37 bottom line.
49:39 The world model is buildable today as a
49:41 research tool proof of concept.
49:45 It will be buildable as a component
49:46 inside real robotic systems within about
49:49 a year. So, 3 to 5 years. Okay.
49:52 But, here's the thing that matters for
49:53 your world, Kyle.
49:55 The humans amplified by AI framing,
49:58 that's Claude is remembering the [ __ ]
50:01 I've been talking about, which is cool.
50:03 Humans amplified by AI.
50:05 The fact that this runs on one GPU,
50:07 trains in hours, and the code fits in a
50:09 single file means it's not locked behind
50:11 billion-dollar labs anymore. Well, it
50:13 but it will be if it's a big-ass model,
50:16 a big-ass world, the moment world models
50:19 become something
50:21 a motivated individual can build and
50:23 experiment with on their own hardware,
50:26 they become part of a toolkit for the
50:28 kind of people AI Salon serves. That
50:31 inflection point just happened with this
50:33 paper. All right, cool.
50:35 Groovy.
50:36 All right.
50:38 That's powerful.
50:39 What's going on? Questions, thoughts,
50:41 what's going on over here in YouTube?
50:44 I'm scared. My husband is believing all
50:46 the obvious AI videos, not
50:49 even the good ones. This is going to be
50:50 a long life.
50:55 I'm sorry about that.
50:57 Oh, I lost I lost screen share audio.
51:01 I totally would have Shakespeare's its
51:03 response if you hadn't prompted like
51:05 that.
51:07 Uh uh uh
51:09 Had a friend warn me about the
51:11 government
51:13 offering AI education.
51:16 Well, hey, at least at least they're
51:18 [ __ ] trying. At least the government
51:20 is not sitting there going, "We don't
51:21 know what AI is."
51:28 Fittymint in the house, what's
51:30 happening?
51:34 How's Adam?
51:45 Adam
51:50 So, so there's two issues with Adam.
51:52 Well, well, so okay.
51:55 Adam doesn't suck too bad.
51:57 So, there's this guy, Alex Finn,
52:00 and he's been he's been all in on Open
52:03 AI for
52:04 3 months now.
52:07 And he's got a YouTube channel, and you
52:08 can go watch his YouTube channel. And he
52:10 And he just put out a video that said
52:13 3 months of
52:15 Open Claw Learning in 27 minutes or
52:18 whatever it was. So, it's a half-hour
52:20 video.
52:21 And I thought, "Oh, I'll go look at all
52:23 the things I haven't set up correctly."
52:25 And basically, I had set Adam up mostly
52:28 correctly. Like, there wasn't a lot that
52:29 he showed
52:31 that I haven't done. I've just done it
52:33 poorly.
52:35 Um
52:38 And so, for example,
52:41 so when I watched
52:43 Alex Finn's video, I got some
52:44 confidence. I was like, "Oh, okay. I
52:46 know what to do with Adam here, and I'm
52:47 going to do this, I'm going to do that."
52:48 And I designed a whole ass
52:52 I deleted some cron jobs that were
52:54 corrupted. Cron jobs are just repeating,
52:57 automatically running things that run on
53:00 whatever schedule you want.
53:02 As in chronology. I think that's the the
53:04 base of cron.
53:06 Or cron is the base of that.
53:08 Um
53:10 And so, I said, "I'm going to do a daily
53:13 Great Repurpose signal report, and and
53:17 we're going to have Adam go out, and
53:19 he's going to read on Reddit, and he's
53:20 going to read on X, and he's going to
53:22 read blogs.
53:23 And he's going to go find us on a daily
53:25 basis five articles that are
53:29 in line with the Great Repurpose." And
53:32 so, I spent couple hours within 3 4
53:34 hours on Sunday
53:36 designing this newsletter. And it was
53:38 just going to be internally for for
53:40 Brandon and Andy and myself. It was just
53:42 like an internal test.
53:44 And so
53:46 I designed this thing and it was
53:47 beautiful. It looked great. It had it
53:50 would had nice formatting. It was really
53:52 well done. The logic was done. It was
53:54 well written.
53:57 And so I said, "Okay, great. Put it in
53:59 the cron job." And like every morning at
54:01 7:30
54:03 East Coast time
54:05 real time, like where I am right now,
54:07 real time
54:10 send this newsletter.
54:13 Cool.
54:14 So, the next morning
54:18 you're talking about your AI son like
54:20 that. My AI son is sitting is sitting
54:23 outside the
54:26 the school on the steps with drool
54:29 dribbling down his chin. He's not the
54:31 brightest He's not the brightest one in
54:33 class.
54:36 So, the next morning
54:38 like 7:30 comes and goes.
54:41 And it's like there's no [ __ ]
54:43 newsletter.
54:45 And then I go look at the cron jobs and
54:47 there's two cron jobs in there. One of
54:49 them's at 8:00 Mountain, one of them's
54:51 at 7:30
54:53 New York time.
54:56 It missed the New York one. I said,
54:57 "What's going on?" It said, "Oh, I'm
54:58 about to run the one at 8:00. I couldn't
55:00 run the one at 7:30 cuz whatever." It It
55:03 made some [ __ ] up.
55:05 And then at 8:00 it said, "Oh,
55:08 I've got this
55:10 thing that you want.
55:12 And and I ran that at 8:00 and here's
55:14 the output of it." And it was a
55:17 paragraph with some HTML in it,
55:20 unformatted HTML.
55:25 And then I said, "No, you're supposed to
55:27 send
55:29 the
55:30 the
55:31 the
55:32 thing, the newsletter that we talked
55:34 about."
55:35 And so it emailed that paragraph to
55:38 Brandon and Andy and me.
55:41 And I was like, "No."
55:45 And so, I deleted the one cron job and I
55:47 tried to get it to
55:49 and it just every morning it goes, "I
55:51 tried to email you, but I can't. I don't
55:53 have access." And I'm like, "No, you do
55:55 have access. Try to email me right now."
55:57 And he'll email me just fine.
56:00 And then the next morning he goes to
56:01 email me with his cron job and he can't
56:03 do it. So,
56:04 so, I need I need someone who doesn't
56:07 suck to help me
56:09 fix Adam.
56:10 Adam sucks.
56:12 Thank you for asking about my horrible
56:15 digital son.
56:18 Tough to market a new good idea that can
56:21 be so easily replicated now. Yes.
56:25 So, Fiddyman, this is this is one of my
56:27 This is one of my new new insights. So,
56:30 next Tuesday, I want you If you're in a
56:33 regular, please go register for this.
56:36 If you're in the mastermind.
56:39 So, in the mastermind, we are now adding
56:41 We've We've got the practice lab, right?
56:43 Which is creating a daily practice.
56:45 We're going to spin up this new area in
56:47 the mastermind
56:49 for the great repurpose. And one of the
56:52 first things that I'm doing in there
56:55 is going to be creating your own
56:58 personal mission statement using AI that
57:01 can be that can address what you're
57:02 talking about, Fiddyman.
57:05 Which is
57:06 we're coming
57:08 into a world we we
57:10 probably are actually there,
57:13 but it's it's going to take a while for
57:15 it to feel real.
57:17 We're coming into a world that
57:20 if you market
57:24 the thing you're trying to sell,
57:27 it will fail.
57:31 Period.
57:33 And you're like, "What?
57:36 Huh? Kyle, what are you talking about? I
57:40 don't understand."
57:43 If everyone can replicate everything,
57:47 and if everyone's going to get laid off,
57:49 and the skills that they used to trade
57:52 their time for money for are devalued,
57:57 then everyone's going to be selling
57:59 something, a course, a workshop,
58:02 consulting,
58:05 some version of themselves.
58:08 And what we've always done,
58:11 for the most part,
58:13 is when we're selling something,
58:15 we sell the thing.
58:18 I'm doing a workshop. You're going to
58:20 learn these six things. You're going to
58:22 learn them
58:24 faster, better. They're going to You're
58:26 going to make money because of the
58:27 things I'm going to teach you.
58:32 The problem with marketing the thing
58:35 is that everyone can replicate the
58:36 thing. So, if your thing is so good that
58:38 it rises above the noise, which it
58:40 won't, but if it does,
58:42 everyone else down here will look at
58:44 that and go, "Oh, I'm going to sell one
58:45 of those. That rose above the noise."
58:48 And they will just instantly replicate
58:49 you.
58:51 And so, you go from competing against no
58:52 one to competing against everyone in a
58:54 week.
58:58 And no one's going to pay attention cuz
58:59 everyone's going to be making the same
59:01 claim. I make it easier. I make it
59:09 But what they are going to pay attention
59:10 to,
59:11 I think,
59:16 is the individual behind
59:18 the thing.
59:21 So,
59:23 what I think we all need to get clear on
59:25 is what is the thing that we have it as
59:28 an individual
59:30 that no one else in the world has.
59:35 Cindy [ __ ] is like this wacky futurist
59:38 with this art
59:39 thing and
59:41 her crazy glasses and her mind that's
59:43 like all over the place and pulling in
59:46 things from
59:48 you know, her research world and from
59:50 her creative world and she's she comes
59:53 at
59:54 what she does in a very particular way.
59:58 And so for example, she's got a workshop
1:00:00 that's coming up at the end of March.
1:00:05 And we were talking about that and I'm
1:00:07 realizing
1:00:08 the workshop itself
1:00:11 people are a lot of people are making
1:00:13 the same claim she is. I mean hers is a
1:00:15 bit different cuz it's about mindset and
1:00:17 there's some different messaging in
1:00:18 there.
1:00:19 I was like, wait a minute.
1:00:21 They're not going to buy the workshop.
1:00:22 They're going to buy wanting to hang out
1:00:24 with Cindy [ __ ]
1:00:26 and understand what she's talking about
1:00:28 when she talks about mindset shift and
1:00:31 what does it mean to be
1:00:34 in this AI future?
1:00:37 It's about her and so
1:00:39 the workshop that I'm giving next
1:00:41 Tuesday which is I think
1:00:47 11:30
1:00:49 Mountain.
1:00:51 I think that's right.
1:00:54 Scary if your individual story isn't
1:00:56 compelling enough to justify a business
1:00:58 not to copycat you.
1:01:00 No, but here's the thing. You're still
1:01:01 going to you're still going to
1:01:02 ultimately talk about the thing you're
1:01:04 selling.
1:01:05 But the hook to get there is not the
1:01:07 thing. The hook to get there is you. Who
1:01:11 are you? Why did you make that thing?
1:01:13 What is it about that thing? Is there
1:01:15 some story about someone's life changed
1:01:18 and that made you make that thing. It's
1:01:20 going to be all about tying people's
1:01:23 personal relationship to you and their
1:01:25 trust of you to the things you're
1:01:28 selling.
1:01:30 Then you can talk about the things. It's
1:01:32 just It's the initial marketing
1:01:36 push.
1:01:38 If you talk about the thing, it's just
1:01:40 going to
1:01:41 you're just going to be against
1:01:42 everyone.
1:01:44 So, it's not the workshop on being AI
1:01:47 ready, it's Kyle Shannon bringing his
1:01:49 particular flavor of what he does to
1:01:52 teach those same six bullet points that
1:01:54 everyone's going to be teaching.
1:01:57 Make sense?
1:01:59 So, that's coming up. So, if you're in
1:02:00 the mastermind, go please go RSVP for
1:02:02 that.
1:02:05 I'm getting into vertical gardening
1:02:06 mainly because I can't squat anymore.
1:02:13 Silver Fox is the best.
1:02:19 That's a Sounds like a stand-up the
1:02:22 beginning of a stand-up act.
1:02:26 They will pay for scaling and
1:02:27 reliability. They'll pay for trust. I
1:02:30 agree, they'll pay for trust.
1:02:32 But if But here's the thing, Uncanny
1:02:34 Uncanny Valley,
1:02:35 um
1:02:37 if you say
1:02:40 I'm going to give you scaling and
1:02:41 reliability, everyone's going to say
1:02:43 that.
1:02:45 And everyone's going to be able to
1:02:46 deliver that.
1:02:47 Why is yours different? Well, ours
1:02:48 actually works. Okay.
1:02:51 But how do you rise above the noise? If
1:02:52 you want to market that thing,
1:02:56 then you
1:02:58 you know, troll out your your
1:03:01 chief scientist to talk about why yours
1:03:03 is different.
1:03:05 Here's why he's passionate about it.
1:03:07 He's, you know,
1:03:09 he's not very good in social situations,
1:03:11 but you know what he knows?
1:03:13 Scaling and reliability.
1:03:15 He has one little quote from him. Uh I'm
1:03:18 very passionate about scaling and
1:03:20 reliability.
1:03:21 Right? And then you're like, oh, that
1:03:23 guy. That guy I I would trust him for
1:03:27 scaling and reliability cuz he's got
1:03:29 that personality for someone that would
1:03:31 be into that.
1:03:33 Right? That's what this is about. Anchor
1:03:35 anchoring the thing you're selling
1:03:39 in the person or the stories attached to
1:03:41 that person who's got some credibility.
1:03:44 And we all like everyone has a thing
1:03:49 that you're most qualified to talk about
1:03:51 and to do and to be.
1:03:54 And so that's what this this workshop's
1:03:55 going to be on next Tuesday.
1:04:05 Okay, let's go back to this thing. Let's
1:04:09 go back to
1:04:13 All right, that was Jan LeCun's world
1:04:15 model. That was a fun little
1:04:16 dig dig down into the hole.
1:04:19 I had Siri write its own eulogy. Oh,
1:04:21 that's pretty funny. Oh, you should be
1:04:23 able to hear
1:04:41 The spoon too suggestive.
1:04:49 Oh my god. That's hilarious.
1:05:12 I've seen that video before.
1:05:14 Open AI should just remove restrictions
1:05:16 on Sora for a day before discontinuing
1:05:18 it. Satoshi never existed.
1:05:21 If you think Claude killed Open Claw,
1:05:24 you clearly don't understand Open Claw.
1:05:26 That was actually an Alec Finn post that
1:05:28 someone just stole.
1:05:31 After 8 months in stealth,
1:05:34 Hark is the new thing.
1:05:38 The billionaire founder of the $39
1:05:40 billion humanoid robotics company Figure
1:05:43 AI
1:05:44 announced his newest venture, an AI
1:05:46 intelligence lab named Hark.
1:05:50 First time
1:05:51 >> Yes, sir.
1:05:53 Am I not sure right?
1:05:54 >> unexpectedly and suddenly have 30 people
1:05:57 over on TikTok. So, we might want to
1:06:00 take a moment from doom scrolling and
1:06:01 tell them about the AI Salon.
1:06:04 We should probably do that. So, do me a
1:06:06 favor there, Jan Jan What's your name?
1:06:10 Brandon.
1:06:11 Wait, how am I going to show this? Let's
1:06:14 see.
1:06:16 Pop up uh
1:06:19 Here, I'll do it.
1:06:21 Cuz I know what I want. Yeah, that. Is
1:06:24 that it?
1:06:27 No.
1:06:29 Do do do
1:06:33 There we go.
1:06:34 I have the uh link pins to TikTok.
1:06:38 Okay, cool.
1:06:40 So, yeah. So, if if you're new here on
1:06:42 TikTok, and actually if you're new here
1:06:44 on uh YouTube and LinkedIn and Twitter,
1:06:48 um
1:06:53 2026 is about to get really weird.
1:06:58 Jobs are going to get [ __ ] up.
1:07:00 People are going to lose jobs. People
1:07:02 that keep jobs are going to find that
1:07:04 those jobs are just It's normal.
1:07:07 They're not what they used to be. You're
1:07:09 not having as much fun as you used to,
1:07:11 or maybe you're having more fun cuz you
1:07:12 kind of hated your job.
1:07:14 But it's just going to get weird. Things
1:07:15 are going to change.
1:07:17 Um
1:07:21 It's going to start in 2026. I think
1:07:23 2026 is clearly the year
1:07:26 that we all confront that this AI
1:07:29 stuff's not going away, and it's going
1:07:31 to mess some stuff up.
1:07:33 And
1:07:35 that's the beginning.
1:07:37 Right? The next 5 years are going to be
1:07:39 a
1:07:40 an acid trip.
1:07:43 And so
1:07:45 the AI Salon is a thing that we
1:07:49 started the week that ChatGPT came out,
1:07:52 and the whole idea was to create a
1:07:54 community of people who are curious
1:07:56 about AI.
1:07:57 And so what you've got now in in the
1:08:00 salon is a community of about 4,000
1:08:02 people
1:08:04 that
1:08:06 have been
1:08:07 trying to figure it out.
1:08:10 Learning about prompting early on,
1:08:12 learning about custom GPTs, learning
1:08:14 about reasoning models, learning about,
1:08:16 you know, all these new multimodal
1:08:18 models, and what that means, learning
1:08:20 about video and images and sounds and
1:08:24 just all the different things.
1:08:27 And it's a group that
1:08:32 while we explore a lot of the tools is
1:08:35 really talking about the human impact of
1:08:38 AI.
1:08:40 So the one thing I wouldn't do if I were
1:08:42 anyone right now, if I did have any
1:08:44 advice for anyone, it's get your ass in
1:08:46 a community. Get your ass in an AI
1:08:49 forward
1:08:51 empathetic, compassionate,
1:08:54 patient, generous community full of
1:08:58 people like the people that hang out on
1:09:00 this his channel and on these
1:09:02 these YouTubes
1:09:04 that are people that are just trying to
1:09:06 figure it out.
1:09:08 And they don't have ego about it and
1:09:09 they're not being proprietary about it.
1:09:13 One of our one of our tenants in the AI
1:09:16 Salon is generous leadership, generously
1:09:19 lead.
1:09:20 And what that looks like is as you learn
1:09:22 stuff, share stuff. And so we have a a
1:09:24 very generous community
1:09:26 where if you come in and you're like, I
1:09:28 don't know nothing.
1:09:30 People won't judge you for that. They're
1:09:31 just like, cool. I was there like 2
1:09:33 months ago.
1:09:36 It's almost like whoever in the
1:09:38 community knows more than anyone else,
1:09:40 they're the ones that feel the most
1:09:41 clueless. Even no matter how clueless
1:09:43 you think you feel, we all feel that
1:09:45 way.
1:09:45 No matter how behind you think you are,
1:09:48 we all are.
1:09:49 So, community.thesalon.ai.
1:09:53 Um it's free. Go check it out. Um
1:09:56 there's a thing called the AI Salon
1:09:58 Mastermind, which is a subscription.
1:10:01 And that's got just different level of
1:10:04 commitment. The people in there are kind
1:10:06 of different. They're up in their game.
1:10:08 And uh
1:10:10 go check that out. All right?
1:10:12 Juanita, see in the D. What do you think
1:10:15 of the prompt queens and kings on
1:10:17 TikTok? They're cool videos.
1:10:21 Um they post in lives. Um
1:10:38 I don't have my button here. I have a
1:10:39 little button that says, you can make
1:10:40 money with ChatGPT.
1:10:43 Um I think anyone that's selling prompts
1:10:45 right now,
1:10:52 I'm not a fan of that.
1:10:57 For one thing, we all have large
1:10:59 language models.
1:11:03 I don't travel with the button, no.
1:11:07 We all have access to large language
1:11:08 models. So, if you want to know how to
1:11:10 prompt Claude,
1:11:12 ask Claude to teach you how to prompt.
1:11:17 What I think is more important than
1:11:20 spoon-feeding someone a thousand prompts
1:11:22 that they're not going to go through and
1:11:24 test
1:11:26 is is, you know, teach someone how to
1:11:28 fish. Don't give them a fish. Teach them
1:11:31 how to fish. Or don't sell them a fish
1:11:32 for $149.99.
1:11:34 You can have the ultimate prompt
1:11:35 package.
1:11:40 Understand
1:11:42 how these things work. Understand that
1:11:44 giving context is important. Understand
1:11:47 that the prompt methodology that you
1:11:49 just paid for was probably created a
1:11:51 year ago
1:11:53 before the technology changed.
1:11:56 How you prompt an open Claude agent
1:12:03 is not how you prompt ChatGPT. I'm
1:12:06 learning that the hard way. It's a
1:12:07 completely different skill.
1:12:11 Don't make money, make a difference.
1:12:13 Yeah, you make a difference by
1:12:14 educating. And you don't, you know,
1:12:16 I I just
1:12:18 The The one thing I know about being
1:12:20 this early in an industry that's moving
1:12:22 as quickly as AI
1:12:24 is that anyone Anyone Anyone that tells
1:12:28 you right now that they're an expert in
1:12:30 any way is full of [ __ ]
1:12:35 They might be an expert on a particular
1:12:37 kind of prompting, a particular kind of
1:12:41 technical setup, a particular type of
1:12:43 management of AI systems.
1:12:48 But,
1:12:50 as the technology changes, and if we
1:12:53 move if if indeed we're about to
1:12:56 transition from LLMs, large language
1:12:59 models, to world models
1:13:01 over the next 3 years,
1:13:04 all of that prompting [ __ ]
1:13:07 like it may help you conceptually,
1:13:10 but it's just going to be fundamentally
1:13:12 different skills over here. So, what we
1:13:14 like to talk about in the AI Salon is AI
1:13:17 readiness.
1:13:19 And AI readiness looks like mindset
1:13:21 shift. Like get curious. Play first.
1:13:26 Play without expectations.
1:13:29 Learn across domains.
1:13:31 Why?
1:13:34 Because if all you're doing is taking AI
1:13:36 and applying it to what you know,
1:13:40 you're limiting your your lens
1:13:44 of what you think AI can do to the [ __ ]
1:13:46 you already know.
1:13:52 What about the other 99.999%
1:13:55 of [ __ ] that AI can do right now?
1:13:58 If you're just like, well, I
1:14:00 write technical reports, so I'm just
1:14:01 going to do technical reports. Cool.
1:14:07 But maybe you take that technical report
1:14:09 knowledge and you go into Claude and you
1:14:11 say, "Turn this into an interactive
1:14:13 dashboard for me." And it'll just code
1:14:15 that for you.
1:14:17 And if you're just looking to make a
1:14:19 report, it'll make a report, but you'll
1:14:21 miss out on this cool other thing you
1:14:23 can do.
1:14:26 Like that's a mindset shift.
1:14:29 Instead of thinking about AI as this
1:14:31 thing we compete against, which by the
1:14:33 way you're going to lose,
1:14:37 if Garry Kasparov couldn't beat Big
1:14:39 Blue,
1:14:41 you're not going to compete against
1:14:44 an AI that's as smart as 274 PhDs.
1:14:51 You're just not.
1:14:52 I don't care how smart you think you
1:14:54 are. I'm special. Well,
1:14:57 we're all special.
1:15:00 It's going to be smarter than you. So,
1:15:01 if you compete against it, that's going
1:15:03 to suck.
1:15:04 What if
1:15:08 rather than say, "I'm smarter than AI."
1:15:10 What if you just say,
1:15:12 "AI does what it does.
1:15:15 I'm who I am. Here's what I care about.
1:15:18 I'm going to put me at the center of the
1:15:19 conversation.
1:15:21 And rather than compete against AI, I'm
1:15:23 going to take AI and strap it on like a
1:15:25 jetpack.
1:15:26 And I'm going to amplify my ideas. I
1:15:29 stay at the center. AI amplifies me."
1:15:33 That's a mindset shift. So, that's what
1:15:34 the that's what the salon is about.
1:15:37 It's about being around people that are
1:15:38 thinking about AI in
1:15:43 thoughtful,
1:15:45 high-integrity,
1:15:47 ethical,
1:15:50 human-centered
1:15:53 ways of being.
1:15:56 And that's cool, cuz not a lot of people
1:15:58 are doing that. A lot of people are
1:15:59 talking about, "Oh, the tools and the
1:16:00 algorithms and the math and who gives a
1:16:02 shit?"
1:16:05 What can it do for me right now?
1:16:09 Well,
1:16:10 learn what it is. Don't buy a [ __ ]
1:16:13 prompt library.
1:16:16 If you just bought one, use it. Try it.
1:16:19 You'll just learn something from it, but
1:16:23 what you'll likely learn,
1:16:24 yeah, time to go. Yeah, my voice is
1:16:26 going.
1:16:28 My voice is going, good people.
1:16:32 Um
1:16:35 All right.
1:16:37 Fantashtic.
1:16:42 Um the AI Readiness Project podcast is
1:16:46 happening tomorrow. I will not be there.
1:16:49 Um
1:16:50 Ann Murphy Ann Murphy is um
1:16:53 doing the podcast tomorrow
1:16:56 with a guest co-host. I had Chef Kelly
1:16:58 on I think last week
1:17:00 um or the week before and Ann couldn't
1:17:02 do it and then I can't do it tomorrow
1:17:04 cuz I'm here at the at my client thing.
1:17:06 So um really enjoy your powerful
1:17:09 content. Thank you so much. I appreciate
1:17:11 that. Do you use AI locally? If so,
1:17:13 what's your setup? Um I don't really
1:17:15 have the iron to do anything interesting
1:17:17 in my open class setup. I have Ollama
1:17:21 running the 4 billion parameter Quinn
1:17:24 3.5
1:17:27 mobile optimized model.
1:17:33 Um I'm probably going to buy
1:17:39 a MacBook G5 or M5
1:17:43 with 64 gigs of RAM
1:17:46 and then I don't know a couple of
1:17:48 terabyte. Oh, sure the water bottle.
1:17:49 Yeah.
1:17:51 This is our new AI Salon
1:17:54 collapsible water bottle and we have
1:17:55 little contests and things like that and
1:17:57 Kelly Camp just won one of these. Did
1:17:58 you get Kelly Camp, did you get your
1:18:00 bottle I sent you?
1:18:06 Okay.
1:18:08 Kyle, my main melon.
1:18:13 Um so I'll probably do one of the
1:18:14 higher-up Quinn models when I get a when
1:18:16 I get an M5.
1:18:18 All right.
1:18:21 Have you heard of A step 1.5? I've not.
1:18:25 But I'll check it out or yeah, A step
1:18:26 1.5. I haven't.
1:18:28 I mean, here's the thing. Um
1:18:31 the the local
1:18:34 local large language models are getting
1:18:37 better and better and better.
1:18:39 Um
1:18:40 probably by the end of the year we'll
1:18:42 have something that is truly remarkable
1:18:44 that you can just run on a regular GPU
1:18:47 or regular you know, regular desktop
1:18:49 setup. So
1:18:53 Have you transferred
1:18:55 that Claude to mobile yet? Well, I've
1:18:57 got I've got
1:18:59 um
1:19:00 Adam and I talk on Telegram which
1:19:03 So
1:19:05 Adam and I talking on Telegram basically
1:19:07 consists of me yelling at him for being
1:19:10 shitty.
1:19:12 So So I use tele I use Telegram to rant
1:19:16 at my shitty digital son.
1:19:25 If If you want the real experience of
1:19:27 Open Claude, come to the AI learning
1:19:29 lab.
1:19:31 Oh, you didn't get it yet? Uh, I've got
1:19:34 a tracking number somewhere for you.
1:19:37 Um
1:19:38 I always give my regular spanking.
1:19:41 Nvidia announced Nemo Claude not too
1:19:43 long ago. Yeah, I know.
1:19:44 Um Yeah, so So we'll see. All this stuff
1:19:48 All these Claude's are going to get
1:19:49 better and better and better. Right
1:19:51 Right now, if you want to be in early,
1:19:52 you got to deal with the jank.
1:19:55 What's it costing in daily tokens? It's
1:19:57 not costing me anything because I'm
1:19:58 using OAuth.
1:20:00 Um
1:20:02 This is weird. My brain thinks it's
1:20:03 midnight because you're tired. I scoff
1:20:06 at your tired ness.
1:20:08 It's really good. Um when you set up
1:20:10 Open Claude, you can either set it up
1:20:12 with API tokens or with OAuth which is
1:20:15 basically you're signing into your
1:20:17 Chat GPT or Claude account
1:20:20 or Gemini account. Like your monthly
1:20:22 Chat GPT 20 buck a month thing.
1:20:27 In the terms of service of Google
1:20:31 Anthropic and Open AI, it is against the
1:20:34 terms of service to use OAuth
1:20:37 for any API
1:20:39 situation.
1:20:41 Anthropic is enforcing this. I think
1:20:45 Google might be enforcing this. Like and
1:20:47 and by enforcing like shutting down your
1:20:49 account.
1:20:51 Open AI, because they just hired Peter
1:20:53 Steinberger who created Open Claw, is
1:20:56 actually encouraging this.
1:20:58 So, when you set up Open Claw, you set
1:21:01 it up with the Codex
1:21:04 Open AI Codex and then do like 5.4
1:21:08 and then you OAuth
1:21:10 um
1:21:12 authenticate
1:21:14 and then you can use 5.4 until you burn
1:21:18 through your weekly credits, which I do.
1:21:23 I I generally burn through my weekly
1:21:25 credits. I'm only paying 20 bucks a
1:21:26 month right now. I generally burn
1:21:28 through my my weekly credits in 2 days.
1:21:33 And and then I just don't use Adam for a
1:21:35 while
1:21:37 because he pissed me so off so much in
1:21:40 the 2 days. And then so, I'm not
1:21:42 spending a lot of money on tokens.
1:21:46 I haven't tried Open Claw yet. Pros,
1:21:48 cons, I hear the safety issue. Um here's
1:21:51 the deal.
1:21:55 If you just go barreling into Open Claw
1:21:58 um
1:21:59 and just turn on all the [ __ ]
1:22:03 bad things will probably happen to the
1:22:05 [ __ ] you have it connected to.
1:22:08 Just go slow. Use Grok. One of the nice
1:22:11 things about Grok, if you go to
1:22:12 grok.com, grok.com, or use Grok within
1:22:15 chat G or within X
1:22:19 you can have Grok give you Open Claw
1:22:22 like a a conservative Open Claw setup
1:22:25 based on the latest thinking about what
1:22:27 people are talking about on X. That's
1:22:30 what I would do.
1:22:31 Um just go slow and just don't be dumb.
1:22:35 Just
1:22:37 tiptoe, little steps at a time. That's
1:22:39 kind of how I'm doing it. I have I use
1:22:41 Claude as as my coach.
1:22:44 And then I just go do the command line
1:22:45 [ __ ]
1:22:47 to set up open Claude and I just go back
1:22:48 and forth and it's very frustrating. Um
1:22:51 pros are
1:22:53 if and when you get it right, it just
1:22:55 proactively 24 hours a day does stuff
1:22:57 for you. Cons are
1:23:00 it proactively 24 hours a day does stuff
1:23:02 for you.
1:23:04 Which means you have to keep up with it,
1:23:06 which you can't.
1:23:08 Cuz it's working 24/7.
1:23:12 And you're a human, so you're not even
1:23:13 working 8 hours. I know you get paid for
1:23:15 8 hours, but let's face it. 4 and 1/2 of
1:23:18 that is you surfing Tik Tok.
1:23:22 Right?
1:23:23 So as a human, you've maybe got two good
1:23:26 hours a day and this thing's working 24
1:23:28 hours a day. So assuming you get it all
1:23:30 dialed in, you're going to be buried in
1:23:32 reviewing your open Claude bot's work.
1:23:37 Ah,
1:23:38 the Crustifications are pleased with
1:23:40 your progress.
1:23:45 All right, kids. I'm out of here. Um
1:23:47 I don't think I'm going to be here
1:23:49 tomorrow night. I should be here
1:23:50 Thursday night.
1:23:52 Uh and then I'm definitely not here
1:23:53 Friday night and Friday there's no
1:23:55 office hours. I'm going to be on an
1:23:56 airplane headed to Boise, Idaho.
1:24:00 All right, everybody. Everybody you you
1:24:03 take care now, okay? Okay?
1:24:06 Yeah, I'm right.
1:24:07 Yeah, I'm right.
1:24:09 I'm going to go.
1:24:11 I'm going to go in nice my nice hotel
1:24:13 room.
1:24:14 I'm going to get some sleep. All right.
1:24:17 Peace out. Bye.
1:24:20 Mhm.