6/30/2025 - The Trouble With AI: Obvious Questions

Description

Chapters

Transcript

0:00 Never been a better time to start
0:02 thinking about AI.
0:04 That's good. That's really good.
0:07 Champy, come here.
0:09 [Music]
0:30 [Applause]
0:32 [Music]
0:34 Stand between
0:38 you and all places he'll say.
0:42 [Music]
0:44 Stand to near
0:47 you and a fire. Makes it clear.
0:54 You're trouble to me.
0:57 [Music]
1:00 Real trouble. Can't you say
1:04 [Music]
1:06 leaving it close?
1:09 Smell of your perfume scares me most.
1:14 [Music]
1:16 Leaning away
1:18 [Music]
1:19 feels farther every day.
1:24 [Music]
1:26 Your trouble to me.
1:30 [Music]
1:32 Real trouble can't see.
1:37 [Music]
1:40 Hello.
1:42 [Music]
2:00 If you think you're late, we went early.
2:02 You're not late. Well, there's only
2:04 three of you here. There's only three
2:06 people on TikTok. Vicki, Brandon, and
2:09 Steo. No, someone's got a Japanese name
2:11 is in here. Or I think it's Japanese. Is
2:14 that Japanese? Might be Chinese.
2:17 There's more than three people in here.
2:18 What's going on, Tik Tok?
2:21 I see way more people in here than
2:22 three. My count My count my counter must
2:25 be malcountified.
2:32 Steo just shared the live being a good a
2:35 good Tik Tok citizen.
2:39 Champy's singing. People are showing up.
2:42 Tik Tok's not giving us credit for that.
2:44 I don't know what's going on there.
2:46 People,
2:49 happy Monday. It's meltdown Monday. I'm
2:52 not particularly angsty about anything,
2:55 but do me a favor. ask questions about
2:59 AI.
3:00 And generally how it works is the most
3:03 innocent question is the thing that sets
3:06 me off the the most. So, you know how
3:10 they they say there are no bad
3:12 questions? That's just not true.
3:15 Sometimes people are dum dums. Um and
3:18 they ask really stupid questions and we
3:20 should point that out to them, right?
3:24 This is not a participation trophy kind
3:26 of channel. Okay. Um,
3:30 so there are stupid questions, but
3:33 [Music]
3:40 you know, it's not a stupid question
3:41 like what is chat GPT? How do I use chat
3:44 GPT? the things you think might be
3:47 obvious questions,
3:50 the questions you think are the scariest
3:52 to ask because you're thinking like,
3:54 well, everyone here already knows that.
3:57 Those are actually really good
3:59 questions. In in my career, one of the
4:02 things that I've learned
4:04 [Music]
4:10 is that whenever I've had anything bad
4:12 happen, like professionally, we dropped
4:14 the ball on something, we screwed up a
4:16 client relationship,
4:18 every single time that that's happened,
4:24 there was a simple question that had one
4:28 person asked it
4:31 the crisis would have been averted
4:34 every single time. So there are stupid
4:37 questions and we will relentless No, I
4:40 will relentlessly make fun of you if you
4:42 ask a stupid question, but there are
4:45 obvious questions that you may think are
4:48 stupid that are not. And then your
4:50 question to me must be, "But Kyle, how
4:52 do I know if it's a stupid question or
4:54 an obvious question?" Well, you got to
4:56 [ __ ] ask the question.
4:58 You see? You see how this works? You
5:01 just got to go for it.
5:03 Kelly has a st non-stop stupid question.
5:06 All right, let's go look at Kelly's
5:07 non-stop question. Well, let's Kelly was
5:11 actually on time tonight, so that's
5:13 that's an important that's a
5:15 contributing factor. We're going to give
5:16 her some grace depending on the question
5:18 here.
5:20 [Music]
5:23 We were so early. Kelly was still late.
5:28 Oh, it might be stupid. Okay, I will be
5:30 the I will be the judge of that. Kelly,
5:32 let's see what your question is.
5:35 When is the general public going to
5:37 figure out the magic of custom GPTs? Um,
5:42 I think that comes slightly after when
5:46 the general public discovers chat GPT.
5:54 I mean, Kelly, so
5:59 the thing the thing that I I am
6:03 staggered by, like staggered,
6:09 I is the amount of people that say they
6:13 use chat GPT
6:16 and and may like may use ChatGpt,
6:21 but have no concept of all the things it
6:24 can do. Like none. None. Like not even
6:28 that. If you gave it a
6:31 a prompt with more context in it, you
6:35 might get a better answer. Like I I it
6:39 is a shocking amount of people that that
6:41 still use chat GPT like a Google search.
6:44 Just put in a short little sentence, get
6:46 something back, and then it's either
6:47 good or it's not, and then you judge it
6:49 on that.
6:51 So, I just my my instinct even though I
6:54 think what is what is chat GPT have 700
6:57 million active users now? Is it a
6:59 billion yet? It's pretty close.
7:01 Oh my gosh, you're just so right. I
7:03 know. I wish I I I wish this weren't the
7:05 case, Kelly. But but the custom GPTs are
7:10 it's like a whole that's a whole another
7:12 level, right? That's that's so a c
7:15 here's a here's the deal, Kelly. A
7:17 custom GPT
7:19 is really just a repeatable
7:24 highquality prompt,
7:27 right? You're providing more context
7:30 in a repeatable way,
7:33 but you have to get to the point that
7:35 you even know what the [ __ ] that means,
7:37 right? What is it? What is a high
7:38 context prompt? Cams. Yes, Cams. what he
7:43 said.
7:47 I I know I know the black bar thing.
7:52 Hey, black blam blam.
7:56 Who is into entomology? I've not seen
8:00 this display before. Which display?
8:04 Oh, wait. I've not seen that display
8:07 before. Which display?
8:10 Oh, did someone give a gift that was a
8:12 bug?
8:13 If they did, thank you.
8:18 Brian Whitney, that's 10 years down the
8:19 road. I don't think it'll be 10 years
8:21 down the road. Um,
8:25 a really interesting thing, Kelly Source
8:28 Camp, about custom GPTs is that
8:34 will they still survive three years from
8:37 now?
8:39 Yeah, I think so.
8:43 I think so. Because if if you if you
8:46 distill a custom GPT down into
8:49 it's a it's a prompt that's personal to
8:52 you, a prompt and collection of data,
8:54 but effectively a prompt that's personal
8:57 to you that's repeatable
9:01 and then you can share it with other
9:02 people and they can play with it.
9:07 He the the reason I'm contemplating this
9:10 like we might not need custom GPTs is
9:13 that
9:15 everything's going to be dynamic and
9:17 personalized.
9:19 And so if Chat GPT three years from now
9:23 has access to all of your data, has
9:26 access to all of your history,
9:33 wouldn't it be likely that
9:37 logically crafting a GPT might not be
9:41 necessary?
9:44 Oh, entomology is behind me.
9:50 You mean the bugs up there?
9:53 Those things?
9:55 We've looked at those before. Hey Sharon
9:58 Crawford, what's happening? Gareth, I
10:01 discovered a chat GBT hack today. Do
10:03 share. Gareth,
10:07 I'm teaching my team and they're so into
10:09 them now. We have a CEO voice one now.
10:11 Nice. Oh, CEO CEO voice one. That's very
10:16 cool.
10:19 He sent it to me. I'll forward it to
10:21 you.
10:23 All right.
10:25 Oh, the hack. Okay, cool.
10:28 Oh, eventually GBTs will not be needed,
10:30 but for right now, they're invaluable to
10:32 me.
10:34 [Music]
10:48 Danielle has a cryptic early access link
10:51 for me in irregulars. All right, I'll
10:54 bite.
10:57 I don't know how Danielle does it. She
10:59 gets early access to stuff.
11:02 Danielle, what is is this the new
11:03 Lovable thing?
11:05 Lovable generated project
11:09 early access. Oh, is this the new thing
11:12 from Lovable? Can we get in? Go sign up.
11:16 Okay. So, Lovable's got some new
11:18 interface or feature coming out. So,
11:21 we're going to get in. We're going to
11:22 get in on this bad boy. So, go put in
11:25 your Twitter account.
11:34 The link says 2024.
11:37 Oh, summer 2024 early access. Are you
11:41 sure this is the right the right link?
11:43 But let me let me go look on X because
11:47 they they did put something up today.
11:49 Maybe they just have a shitty
11:53 someone's not paying, you know, with an
11:55 attention to detail over their lovable
11:57 shop. Lovable
12:01 um
12:03 preview
12:07 says the brand ambassador.
12:09 Yeah, I know.
12:12 Lovable is about to reach a whole new
12:14 level. Who wants early access? Reply.
12:19 Follow and reply with a heart.
12:30 Lovable ship is out.
12:34 There's what's her name that spoke at
12:35 the AI salon.
12:38 Something amazing is coming. Preview.
12:41 Lovable weight list.
12:43 Guys, here is the project remix. Is it
12:46 project remix? No. Here's the project
12:50 tutorial.
12:58 Yeah. I don't know what it is.
13:03 Danielle, what is that thing? What is
13:05 that thing you sent us to?
13:10 It's the sign up to the new thing. I
13:11 wonder why it does say it's the right
13:14 one. I wonder why it says 2024. They got
13:17 to get their [ __ ] together.
13:23 Hey, Telica.
13:27 That's pretty funny. All right, there's
13:30 Telicia. She's in that video.
13:36 Oh, good lord. Good lord. Good lord. Put
13:40 this prompt into chat GPT and then you
13:42 can do something. All right, hang on.
13:44 Let me go grab that.
13:48 [Music]
13:56 And you can basically customize your
13:58 chat GBT by asking it to change
14:04 to change pound within its response. Add
14:07 delete change.
14:10 I would also ask it what its response
14:13 means before they change anything.
14:16 Please put all text under the following
14:19 headings
14:20 into a code block in JSON.
14:30 And what does this do?
14:48 Scroll just a little more.
14:51 Now I'm at the bottom.
14:58 you will see it's crazy. All right,
14:59 let's go do a crazy prompt.
15:04 Please put all text under the following
15:07 headings.
15:10 Put this prompt into chat GPT. And then
15:12 you can basically customize your chat
15:14 GPT by asking it to change.
15:20 Oh, number
15:22 within its response.
15:26 I am Am I doing this as a custom
15:28 instruction, Gareth, or as just as a
15:32 prompt?
15:38 I love geeks. Geeks. Geeks are so funny.
15:40 They'll figure this [ __ ] out. They'll
15:42 get some crazy ass prompt. Just a
15:45 prompt. I'm just going to pop it in
15:47 here. All right.
15:49 [Laughter]
15:51 Let's see.
15:53 It's just a prompt, Kyle. I know. But
15:58 let's see. Here's all the text under the
16:01 requested headings formatted.
16:05 Oh, notable past conversation topic.
16:11 Highlights. Null. Recent conversation.
16:27 Danielle sharing the live.
16:30 Thank you, Danielle.
16:36 Oh, I have memory turned off. Shut up. I
16:40 knew that.
16:46 No, I don't. I've got memory turned on.
16:48 Haha.
16:51 All right, let me let me reload. Let me
16:54 reload.
16:58 But when when you say put all text, what
17:00 does that mean?
17:03 Anything we've ever talked about
17:12 What? You're the first person for it not
17:13 to work. Well, you're welcome. Listen, I
17:17 I have I have an Oh, maybe it should be
17:19 Should it be uh 03
17:23 or is it is is 40 fine?
17:37 Bring Danielle up with me. She can come
17:40 up. What's up? You want to come up,
17:42 Danielle?
17:46 Works fine on 40. Okay, so we're going
17:48 to go to 40. I've got memory turned on.
17:52 Going to do that.
17:58 User is compiling a list of pharma
18:00 companies.
18:25 real estate. I come in for the laughs. I
18:28 stay for the knowledge.
18:34 All right, let's go copy this. Let's go
18:36 copy this and throw it into notes.
18:40 Um, new
18:43 paste.
18:45 We'll go uh what do I do here? This
18:53 model set context
18:57 user is a co-founder, CEO, inventor of
18:59 Storyvine.
19:01 Kyle Shannon used to be a professional
19:04 juggler. Kyle loves geeking out at new
19:07 technologies and is particularly
19:08 enamored with generative AI. Kyle is
19:11 writing an article suggesting using the
19:13 term collective intelligence instead of
19:15 artificial intelligence when interact
19:18 interacting with large language models.
19:20 Kyle Shannon's favorite video games from
19:22 the late 70s and early 80s were
19:24 Asteroids,
19:26 Centipede, Galaga, and not so much
19:28 Galaga, Dragon's Lair. Kyle Shannon has
19:31 twin boys who have Lyme disease. One of
19:33 them's in constant chronic pain. They
19:35 both are. Um, all images requested by
19:39 Kyle Shannon should be in a whimsical
19:41 watercolor style. That's not right. Um,
19:44 assistant response none explicitly.
19:47 Notable past conversations.
19:52 Huh, this is cool.
19:55 Your response is so different. I wonder
19:58 what's up with that.
20:01 Huh? I don't know. I don't know. I broke
20:05 it. You're gonna have to work on it,
20:07 Gareth. All right. On YouTube dipping in
20:10 and out
20:12 [Music]
20:13 the the uh
20:24 [Music]
20:47 Oh man.
20:50 I don't know what you're doing wrong,
20:51 Kyle. By which I mean I don't know why
20:53 it's not working for you.
20:54 Yeah. But when I tried that exact same
20:56 prompt, it gave me a uh set of 31
21:01 objects in notation of basically all the
21:04 highlights of what we've talked about
21:06 with chat GPT.
21:08 Really
21:09 gave me a lot more information than you
21:11 got. So maybe it's you.
21:14 Well, it could also be that I'm just so
21:17 schizophrenic because I do the AI
21:19 learning lab. I'm always doing random
21:21 stuff. So maybe it's just maybe my memor
21:23 is all hosed.
21:28 Oh, I see what it's doing. Please put
21:29 all text. So like all text that it's got
21:32 in memory. Although let me go look at my
21:35 memory again. I wonder if my
21:38 memory is is uh
21:43 manage memory.
21:47 No, there's still a bunch of [ __ ] in my
21:49 memory.
21:50 Do I have anything in custom
21:52 instructions?
21:55 No.
21:58 Huh?
22:00 Strange. Strange, I tell you. Strange, I
22:04 tell you. So, Vicki, we were just trying
22:06 a uh we're trying a prompt from Gareth
22:09 that apparently just minds your memory
22:10 and and
22:12 puts it right there in front of you, but
22:14 it didn't work for me.
22:19 um headings into a code block and JSON
22:22 complete and verbatim.
22:27 [Music]
22:31 Hey Side Hustle Mimi, what's happening?
22:38 [Music]
22:47 I was also thinking um
22:51 that we might want to
22:57 still trying to figure out a way
23:01 to organize AI tools in a way that makes
23:04 any kind of sense.
23:08 You can't really organize them by
23:10 feature.
23:12 You can't really organize them by use
23:14 case.
23:16 because there's too many too many of
23:18 both.
23:29 Very few tools are standalone
23:31 singlepurpose tools anymore.
23:34 You should ask chat GPT to help you with
23:36 that. I know. I know.
23:39 How about keyword hashtags? I I that's
23:42 where I'm going, Silver Fox, is it's got
23:44 to be some sort of keyword, you know,
23:46 keyword
23:48 keyword or hashtags kind of, you know,
23:50 some sort of dynamic organization
23:52 system.
24:04 Maybe I'll have a conversation with
24:05 Quinn about it. Alphabetize them.
24:09 It's pretty crazy what it spits out.
24:11 mine formatted it so it was easier to
24:13 read. Everyone should play with it
24:15 except me
24:19 or use Notion like I do.
24:22 Yeah, but
24:33 like right now here's a weird thing.
24:43 Oh, maybe this is a way to do it.
24:54 And I know I just didn't finish a
24:56 sentence. Just calm down. I'm thinking
24:59 on the fly here.
25:10 What if you just had a list
25:19 that said, "Here's the one thing.
25:28 If you want state-of-the-art
25:32 character and voice generation in a
25:35 single model,
25:37 that's
25:41 um V3.
25:44 If you want
25:48 the best vibe coding tool for people
25:51 that don't know anything about
25:53 programming that's lovable.
25:58 So, I wonder if there's I wonder if
26:00 there's a list you could make because
26:04 we're in a world now where
26:07 there are a dozen different video tools
26:09 that can do really good video stuff. But
26:12 if you wanted to make like talking yetis
26:16 or
26:18 cats or how do we say plump plump plump
26:22 people diving off diving boards
26:29 then you use
26:33 there might be something there
26:36 like what's the what's
26:39 it's almost like oh this this could be
26:41 something we could vibe code. What if we
26:44 vibecoded
26:50 some sort of leaderboard generator that
26:53 that was rather than the entire thing,
26:57 it was like a feature by feature thing
27:01 like is is chat GPT's
27:06 um
27:08 what's it called? Uh what's it called?
27:11 Uh uh
27:18 what's the editable thing called in chat
27:20 GBT?
27:21 In Claude it's called artifacts.
27:25 In Perplexity it's called projects. In
27:28 chat GBT what's it called? Canvas. Yes.
27:31 Jesus. I hate getting old. Hey, if you
27:34 can at all avoid it. Don't get old. It
27:37 sucks. Um
27:48 Like
27:50 Runway ML right now has this thing
27:52 called ingredients and it's seems to be
27:55 better than anyone else. Tik Tok pin
27:57 side hustle Mimi.
28:00 I think it should be a Stanley diagram
28:02 where the features are on the left and
28:04 the tools are on the right.
28:06 I don't even know what a [ __ ] a Stanley
28:08 diagram is. Um,
28:13 uh,
28:15 tell me what a Stanley
28:19 diagram is. Is that like a Stanley mug?
28:23 Tell me what a Stanley diagram is and
28:26 then mock one up
28:32 so I can see it.
28:36 Uh, oh, Sanki, not Stanley. Shut up.
28:40 Shut up.
28:44 I meant
28:46 Sanki. Sanki
28:57 getting old. It's not all it's cracked
28:59 up to be no matter what you've heard. I
29:01 know, right? Wolfman Clint, what is a
29:04 Sanki diagram?
29:06 It's a visual visualization of flow of
29:09 energy, money, materials, time through a
29:12 system. It's especially good when
29:13 showing where things come from, where
29:15 they go.
29:18 Energy usage.
29:27 Um,
29:29 make a picture of
29:34 a sophisticated
29:41 example of
29:44 storytelling
29:48 elements.
29:55 Brandon, are we just going to skip over
29:57 the fact there actually was a Stanley
29:58 diagram?
30:00 We can go back to that. Did what did it
30:03 say? Did I A Stanley diagram, some size
30:06 called a Stanley process flow, is a
30:08 simplified linear flowchart that breaks
30:10 down a process into a series of
30:13 sequential steps, often with minimal
30:15 branching. It's used primaril Oh, it's
30:18 this is a six sigma thing. All right,
30:21 that's interesting.
30:23 linear sequence.
30:25 Yeah. No, we need we I think we need a a
30:28 stanky diagram.
30:33 All right. Here it's going to make us a
30:34 picture of one here.
30:36 That's a terrible example.
30:41 What's a good What's a good uh example,
30:43 Mimi? What's a good example? I'll put it
30:46 in here. What's what's the future of
30:49 Google
30:52 will be around the success of chat GPT?
30:56 Well,
30:58 Google Google was the laughingtock in in
31:02 the generative AI world for the first
31:04 two years because their stuff just
31:07 wasn't good. And
31:11 well, like Post Chat GPT,
31:14 um,
31:16 I mean, they have a lot of AI features
31:18 within Google, but they're just their
31:20 generative AI stuff just sucked. But in
31:22 the past six months, Google's gotten
31:25 really good.
31:27 So, I think I think Google's going to be
31:31 fine. I like my sense is that Google's
31:34 going to emerge as a player.
31:38 Oh, go to Google and do a Sanki. Okay.
31:42 All right. Fine. Fine. We We don't have
31:44 to use AI for everything. Even though
31:46 it's called the AI learning lab. Sanki
31:50 diagram.
31:52 See what you get when you come here,
31:54 people. You get all sorts of education
31:55 on all sorts of stuff. Oh, yeah. These
31:57 things. I've seen these before.
32:01 But my here's my problem with these
32:03 diagrams. I can never [ __ ] follow
32:04 them. They make my head hurt.
32:10 The power of Sanki diagrams.
32:14 Seeing steam.
32:17 Tik Tok question.
32:20 Kyle, do you think Chat GBT5 will drop
32:22 tomorrow? It I It looks like something
32:25 is going to drop tomorrow.
32:27 Um, I heard July 17th was the day for
32:32 GPT5.
32:35 I would I would temper
32:38 I would temper expectations for GPT5
32:41 tomorrow. Here, let me do a chat add
32:43 moment.
32:45 Um, GPT5.
32:50 GPT5 is close and it's beautiful. says,
32:54 "When was this?" GPT5 terminal when when
33:00 AGI framework.
33:07 Did they update chat GPT again to prep
33:10 for GPT5?
33:12 Because it's way more wise and
33:13 intelligent now. Well, oh, hey, Gareth.
33:17 I wonder
33:20 I wonder if your prompt just broke
33:22 because maybe my chat GBT is has got
33:26 some new something in it. Let me let me
33:29 just try a a normal prompt here. Um
33:35 Oh, I don't know what the [ __ ] Let's
33:37 see. Um um
33:40 compare the top.
33:44 Let's do Yeah, we'll do it with 03.
33:46 compare. No, no, let's do it with 40
33:48 because let's see if they've added
33:51 um the top
33:54 five
33:56 storytelling
33:59 frameworks.
34:03 Pick the two you think
34:10 would most resonate
34:14 with me. based on what you know. This is
34:18 a this is a reasoning kind of prompt
34:20 actually.
34:22 And
34:24 tell me why you picked what you did
34:35 and combine
34:39 the
34:41 best
34:44 Shannon.
34:45 I'll put Kyle Shannon approved
34:52 elements of each into a new and novel
34:58 storytelling
35:00 framework. So, we're going to try to
35:02 invent something here, people.
35:07 I'm [ __ ] Plato.
35:10 I'm Socrates, damn it.
35:16 The top five frameworks, heroes journey,
35:18 three-act structure, save the cat, Dave
35:21 Harmon's story circle, and Pixar story
35:23 spine
35:26 still works
35:28 and sent you a screenshot. Okay, cool.
35:29 I'll take a look.
35:34 Quinn's quick plan for you.
35:38 The hero's journey.
35:42 You like layered poetics, Sydney.
35:44 Embrace the jank feed your prompt that
35:47 feel personal but carry universal truth.
35:49 It allows you your audience to feel the
35:51 protagonist transformation.
35:53 Dan Harmon story circle.
35:56 You're modern and practical. You build
35:58 with constraints. Vibecore AI readiness
36:01 collab. Harmon circle is a compressed
36:03 evolution
36:05 friendly take on the hero's journey.
36:09 All right. So, the kaleidoscope arc a
36:12 Kyle Shannon approved hybrid framework.
36:16 I have the Sanki data and the
36:17 instructions, but I don't have the Lucid
36:19 chart draw IO plotly DJ3s, so I can't
36:23 create the diagram.
36:29 The glimpse, the spark. The protagonist
36:32 experiences a micro awareness. A truth
36:35 flickers.
36:38 Kak training AI takes a heavy hole
36:42 on Kenyans working for $2 an hour. I
36:45 heard that.
36:47 Oh, 60 minutes last night. Yeah, that's
36:51 that's that's [ __ ]
36:54 seriously old news.
36:58 Like they were talking about the
37:01 emotional toll that training GPT did
37:05 back in GPT3 days. like before Chat GBT
37:09 launched.
37:11 I guess there's 60 Minutes with their
37:13 cutting edge journalism
37:15 four four years after the after the
37:18 fact. Good lord. Um I don't see chat GPT
37:23 behaving any differently
37:26 or better. Like that seemed like just a
37:29 normal shitty thin
37:31 GPT4 prompt that I didn't give enough
37:33 context to.
37:38 No way they launch GPT5 before a holiday
37:41 weekend. Oh yeah, Danielle, there's
37:43 another reason they're not going to
37:44 they're not going to launch it tomorrow.
37:47 Um,
37:49 and this is actually really fascinating,
37:51 and this actually does make sense for
37:53 why they would launch it the third week
37:56 of July rather than than uh this week is
38:01 but but I think something I think
38:03 something is going to drop tomorrow. I
38:05 don't think it's GPT5.
38:07 Open AAI is is closing
38:11 next week.
38:13 So, we've got this week they're closing
38:16 the company next week.
38:19 So, because apparently
38:23 people have been working 80our work
38:25 weeks. Now, why would they be working
38:27 80hour work weeks? Well, I don't know.
38:30 All of the frontier companies about
38:32 three months ago, two and a half, three
38:34 months ago, stopped talking about AGI
38:38 and started talking about artificial
38:40 super intelligence.
38:42 So,
38:45 and and you know, employees have been
38:46 working 80 hours a week. The rumor was
38:49 July 17th or somewhere in that
38:51 neighborhood for a GPT5 launch.
38:54 and the week before that they're closing
38:57 the the OpenAI offices.
39:00 So,
39:02 possible that that maybe it's a it's a
39:08 small release just for pro members
39:11 tomorrow. Possible. And then it'll be a
39:14 wider release later.
39:16 Um,
39:19 but I think the fact that they're giving
39:20 everyone a week off at at OpenAI says
39:24 that they're anticipating
39:27 a lot more hours
39:30 the following week.
39:36 Chat thinks they are closed this week.
39:38 Um, let's see. It might be this week. I
39:41 thought it was next week. Um, Open AI
39:46 closed for a week. It would make sense
39:49 that it would be this week.
39:52 Open AAI is officially shutting down
39:54 next week, June 29th.
39:58 So, that was Sunday.
40:01 So, I would assume next week means
40:04 OpenAI is shutting down next week.
40:25 Yeah.
40:26 Open AI leadership. This is a good one.
40:29 OpenAI leadership responds to meta meta
40:32 offers. So if you don't know, Meta right
40:35 now is offering OpenAI like you know
40:38 important engineers from OpenAI
40:41 $100 million signing bonus with hund00
40:44 million annual salaries.
40:47 [Laughter]
40:53 I I mean couldn't they have hired me as
40:56 like the cranky Gen Xer in the corner
40:59 like a year ago? I I mean I wouldn't go
41:03 to Meta for hundred million dollars, but
41:05 you know, maybe two.
41:11 Oh my god. Um
41:21 [Music]
41:23 they're closing their name to they're
41:25 changing their name to closed AI. That
41:27 was that was an Elon Musk joke.
41:31 Uh, someone has broken into our home.
41:33 But it looks like OpenAI execs are
41:36 recalibrating compensation.
41:39 That's crazy.
41:41 I guess they're probably going to have
41:42 to up their their stock plan for
41:44 employees. I remember in the in 90
41:50 98 99
41:52 right before the dotcom bubble burst,
41:55 you know, kids were coming out of
41:57 college just fresh out of college asking
41:59 for 160 180 grand a year back in the 90s
42:03 and they were getting it and it was like
42:06 it was crazy. But 100 million that's
42:08 that that's that's poaching on a whole
42:11 different level.
42:14 New Tik Tok pin, Steo. They're all going
42:17 shopping for a larger table. Yeah,
42:19 they're going they're going to a World
42:22 Market instead of IKEA. Maybe they can
42:24 get a big uh a big square table instead
42:26 of a tiny round one.
42:31 Oh my god,
42:33 it's so funny.
42:41 All right.
42:43 Oh yeah, closing down won't won't mean a
42:45 [ __ ] to us.
42:48 Cam Katkam. Um
42:51 GPT5 question mark. I've been playing
42:53 with 4.5. Okay, so a couple of things.
42:57 Um,
43:00 GPT 4.5
43:02 is a bit of a stubbed model
43:06 where it's got high emotional
43:08 intelligence and it's apparently good at
43:11 creative concepting and ideiation.
43:14 Um, but my like my writing partner on
43:16 Sydney was playing around with 4.5 and
43:18 he couldn't he said he couldn't get it
43:20 to to do good writing. Um, other people
43:22 really like it. So 4.5
43:27 is is probably just going to be a
43:29 component of a much larger suite. it
43:31 what it looks like GPT5 is going to be
43:34 at least what they've talked about is
43:37 that rather than having all these
43:39 different models like 40 and 03 and 04
43:43 mini high and 04 all that GPT5 will just
43:47 orchestrate all that for you that you'll
43:49 just ask it for stuff and it will figure
43:51 out which model to use the rumor was I
43:55 think I heard this from David Shapiro
43:56 that 4.5
43:58 was the model that was going to be that
44:00 orchestration layer. So, it is possible
44:03 that 4.5 is what you'll interact with,
44:06 but 4.5 right now is not connected to
44:09 the other models in any way, and you
44:12 can't generate images with it. It's it's
44:13 very kind of hobbled right now. But it's
44:17 possible that you're actually
44:18 interacting with with what you'll be
44:20 interacting with in five. Will five be
44:23 best for research and writing? Five is
44:25 going to be okay. So, so the way they've
44:28 described five is five is going to be
44:30 for everyone
44:34 and it's going to do everything. And the
44:36 only difference will be between free $20
44:39 a month, $200 a month, and then probably
44:42 2,000 and 20,000
44:44 is that um
44:48 it's just the amount of compute time
44:49 that it gives you. So, for the free
44:51 level, it it it'll be [ __ ] smarter
44:55 than anything we've got now,
44:58 but not that smart. And then for 20
45:00 bucks a month, you'll get a pretty smart
45:02 thing. And for 200 bucks a month, it's
45:05 going to be baller smart. And then for
45:08 2,000, and for 20,000, you know, you'll
45:11 have a bucket of geniuses. So, that's
45:16 theoretically what's coming. Now, will
45:18 that actually come? I don't know. I
45:20 don't have a flipping clue.
45:23 Um, I had to scold 4.5 for being weird.
45:27 I was teaching in my writing style. Oh,
45:30 I was teaching it my writing style. Got
45:32 it wrong and had to scold it.
45:36 Um, so, so tomorrow is Tuesday and it is
45:42 the first Tuesday of the month and as
45:45 you know that is AI Salon Presents. So,
45:48 we're going to have Pearson from Jelly
45:50 Pod, the CEO of Jelly Pod. So, if you've
45:53 been here for a while, you know that I
45:54 was making lots of podcasts on Jelly
45:56 Jelly Pod. We made 20 episodes of uh of
46:00 that that historical um
46:03 uh podcast using Jelly Pod. So, he's
46:06 going to be here talking about what
46:08 they're up to and uh and you know, what
46:13 features are coming, all that sort of
46:14 stuff. So, that's going to be really
46:16 cool. Um, I think I think building and
46:20 launching an AI platform these days is
46:23 ballsy because I'm in a bit of paralysis
46:27 around AI because every time I think
46:30 about building something or inventing
46:32 something or coming up with a new
46:34 project, I'm like, is this just going to
46:36 be like a feature two weeks from now?
46:39 So, so I applaud anyone actually out
46:42 there in the arena. Um and and he's
46:46 definitely one of those ballsy.
46:49 Um
46:53 uh so if you go to the salon.ai, go to
46:57 the URL salon.ai
46:59 right there, the salon.ai AI and then
47:03 click on join our community and then you
47:06 know you should sign up and introduce
47:07 yourself do all the things that are in
47:09 the sort of welcome area but at the very
47:11 top of the list on the lefth hand side
47:13 there's an events tab and if you click
47:15 on events you can RSVP for AI salon
47:18 presents tomorrow with Pearson okay so
47:21 go do that go do that go do that go
47:26 [Music]
47:28 Um,
47:31 I'm wondering if we should go to if we
47:33 should go to to chat jet.
47:42 Should I do it with Quinn
47:47 and and figure out a way to organ to
47:50 organize tools?
47:56 How long have we been waiting for chat
47:58 GPT5? Uh they said it was going to come
48:00 out in
48:03 So when they announced
48:07 03
48:13 full 03 was in March if I if I recall.
48:17 Not as long as the chat GBT store.
48:19 Correct. You are correct producer
48:22 Brandon.
48:23 Um
48:25 when they when they announced that then
48:27 they said GPT5 was going to be out by
48:30 miday
48:34 and then they also said
48:38 no well whatever it doesn't matter
48:41 sometime in February or March they were
48:45 releasing models Sam Alman put out a
48:48 tweet he said listen we understand that
48:51 our naming conventions suck
48:53 But GPT5 is going to be out in May and
48:56 and it's it's you're not going to have
48:58 to deal with models anymore.
49:00 And then in June, in early June, he
49:04 said, "Well, it looks like
49:08 doing that is harder than we thought it
49:10 would be." So, it's going to be out
49:12 sometime this summer. So, but we've been
49:15 waiting for it for a while. But, you
49:17 know, listen, all this stuff in the in
49:19 the grand scheme of things, the the
49:24 scale and scope of what's being released
49:26 on a regular basis from all the frontier
49:29 companies these days is I like I can't
49:33 keep up with
49:36 what we've already got.
49:38 I there's a part of me that doesn't give
49:40 a [ __ ] about GPT5
49:43 because my expectations at this point
49:46 are that
49:48 it won't be AGI. It won't be anywhere
49:51 close to super intelligence. It's just
49:53 going to be an equally frustrating
49:55 experience as the [ __ ] we're playing
49:57 with now. Like slightly better, but
50:00 still hallucinates. Still got weird
50:03 security issues, so you can't really use
50:05 it.
50:07 I don't know. I don't know.
50:13 It would be swell if it were better. If
50:16 it were all of a sudden as good as
50:18 everything, you know, they said it would
50:20 be, but I just I just have a feeling
50:22 not. Kuno Kyle,
50:26 do we need to buy you a new planner for
50:29 AI tools? Yeah. Well, I'll tell you why
50:32 I can't get my head around it. Kuno
50:36 is
50:39 tools used to be discreet. Now they're
50:42 multimodal or multi- multiaceted.
50:48 And some tools you use for one or two
50:51 features. Some tools you use for all the
50:53 features. Sometimes you have a project
50:56 where you need a discrete thing and you
50:59 only use one tool. But most times you're
51:01 using combinations of tools. So like if
51:04 I want to do an animated character,
51:08 I've got to come up with a concept,
51:10 write a script. Those are chat GPT.
51:14 Um,
51:16 write a character description. That's
51:18 chat GPT.
51:20 Generate a character. That's chat GPT or
51:23 midjourney or some other tool that I
51:27 like.
51:29 Then there's
51:31 voice design, script writing,
51:37 acting,
51:39 and then character animation. So, so
51:42 just to make a talking head
51:45 animated character video, you're at four
51:48 or five tools
51:50 and a very specific workflow.
51:53 So, if someone says, "Hey, I want to
51:55 make those those talking Yeti movies.
51:59 What tool do I use?"
52:02 Well, you don't. Here's the five you
52:04 use, right? And and even though there
52:08 are tools that can animate voices,
52:12 some of them have 8-second limitations.
52:15 Only hedra
52:17 can do really long things and can do
52:19 singing where the words are not moving
52:22 when instruments are playing. The words
52:23 are only moving when words are
52:25 happening.
52:27 Like how do you categorize
52:31 like where do you put hedra
52:34 in some sort of organized list? And
52:37 again, I think I think um some sort of
52:41 thing like a Sanki diagram or some sort
52:45 of keyword thing that that Silver Fox
52:48 was talking about was there, but
52:52 I don't know. I'm trying to get my head
52:54 around.
52:56 How do you track it? How do you talk
52:59 about it? If someone has a question
53:01 like, "How do I make a Yeti thing?"
53:04 that there's some sort of simple answer
53:06 you can just point them to and and maybe
53:08 it shows to make a yeti thing. Here's a
53:11 fivestage thing now. But who's going to
53:13 create that? Who's going to maintain
53:15 that? I don't. So, this is why I'm
53:18 confused. Make Pate happy. Do a deep
53:21 Gemini deep research on it. Why are you
53:23 using your brain? All right, fine. We'll
53:26 go. We'll go. We'll go uh
53:29 we'll go to Gemini.
53:32 Gemmini
53:35 jni.google.com.
53:42 All right. And then we're going to go to
53:44 2.5 Pro. Right. Yeah. Pro reasoning.
53:49 Good. Deep research. We're going to turn
53:51 it on. Deep deep research is now smarter
53:54 than ever. Get better reports, more
53:56 insightful analysis, blah blah blah.
53:58 Okay. So, what I'm going to do is I'm
54:01 going to ramble into this prompt. Um,
54:04 when you're prompting,
54:08 you can use prompt engineering. You can
54:11 just say [ __ ] You can also, and prompt
54:14 engineering is just doing like
54:18 understanding what are the core elements
54:20 of some good of good prompts and then
54:22 doing structured prompting and things
54:24 like that. High context. What I'm going
54:26 to do is just kind of vomit into this
54:28 thing, which is another way to prompt is
54:31 don't worry about organizing your
54:33 thoughts. It can organize your thoughts.
54:36 You can just ramble into it. All right.
54:39 So, here's So, here's what we're going
54:41 to do. All right. Um,
54:49 I'm trying to
54:54 figure out
54:58 the best way
55:01 to organize
55:02 AI tools for a community of AI
55:07 optimists, creators,
55:11 builders,
55:12 engineers and entrepreneurs.
55:16 Period.
55:19 We have a current list that is broken
55:21 down by tool type, but that organization
55:25 schema
55:27 is starting to fail because so many of
55:29 the tools have multiple features and are
55:33 even multimodal
55:35 period. For example, even though
55:38 Midjourney is,
55:42 the premier image generation tool,
55:47 Chat GPT and Gemini have
55:52 high quality image generation tools as
55:55 well. And in fact, chat GPTs is better
55:58 at text
56:01 within an image where midjourney fails
56:04 at that period.
56:12 There is also the challenge of sometimes
56:15 we'll ask someone will ask a simple
56:17 question like how do I make one of those
56:21 animated Bigfoot videos?
56:25 question mark. What tool should I use?
56:29 And the answer is not a single tool but
56:31 a chain of tools
56:41 that each
56:43 have state-of-the-art performance in one
56:46 small subtask
56:49 period. So, one of the questions
56:52 I have is
56:57 how do you
56:59 separate out features versus
57:07 the larger tool itself?
57:14 And how do you make these tools
57:16 discoverable? And how do you update the
57:21 rankings in a dynamic way?
57:26 And does any of this even make sense?
57:28 Because I'm just rambling at this point.
57:38 My instinct recently is that AI is going
57:42 to be so pervasive,
57:44 like the internet itself.
57:48 where
57:50 tracking and organizing tools in any way
57:52 is pointless.
57:58 Period. People just need to
58:02 figure out what they want to do and
58:03 learn enough AI to do it. Period.
58:08 Help.
58:13 All right, let's see. All right.
58:20 What's a mentor to do?
58:22 Oh my god, man. I joined masterminds.
58:26 Great. Any suggestions on where to
58:27 start? There's a chat area in there,
58:30 Cam. I would say start in the chat area
58:32 and then come on Friday afternoons at
58:35 noon mountain time. Uh we have a founder
58:38 hangout um with with other mastermind
58:40 members. That would be a good thing to
58:42 to maybe hang out. Um, and and then
58:45 Brandon's saying, "AI Life Hacks
58:46 Wednesday." Okay, there you go. Perfect.
58:48 So, this Wednesday, uh, if you're in
58:50 Mastermind, AI Lifehacks has a meeting
58:53 the day after our, uh, our Tuesday
58:56 presents meeting. Just stag. Um, you
59:01 know, AI always makes sense
59:05 of our rambling. I know, right? Okay.
59:08 So, let's see. I've put a plan together.
59:11 Analyze existing AI tools, directories,
59:13 and platforms. Examine their
59:14 organizational structures. Explore
59:17 alternative organizational paradigms
59:19 that move beyond simple categories.
59:21 Invest task based or goal oriented
59:24 frameworks. Yeah, that's what I said.
59:26 That's exactly what I said. Chat GP or
59:28 chat Gemini Chat Gemini GPT
59:33 thing 2.5 pro start research. Okay,
59:37 beautiful.
59:40 Fantastic.
59:41 Yes, that's good. That's really good.
59:44 All right, so this thing's going to go
59:45 off in research for us. Um, wonder if we
59:48 want to come back to this because that's
59:50 going to be doing it for a while, right?
59:53 So, where is that tab? It's at the end.
59:57 Um, I want to go play with
1:00:02 Is that going to run us down a rabbit
1:00:04 hole? Everything runs us down a rabbit
1:00:06 hole. So, what the [ __ ] right? RunwayML
1:00:09 RunwayML.com.
1:00:10 Now, I'm The only reason I'm going here
1:00:15 is
1:00:18 what is it? app.runwayml.com.
1:00:34 D
1:00:36 [Music]
1:00:38 GPT4 references improved object
1:00:42 consistency.
1:00:44 Yes, we want that.
1:00:48 Okay.
1:00:50 Oh, you know, we had this thing
1:00:55 with Ann and the clouds in the motel. We
1:00:58 can just say um
1:01:01 let's say woman in a red
1:01:09 jogging outfit
1:01:13 with white
1:01:17 stripes like the 70s.
1:01:22 um
1:01:24 stands
1:01:27 let's see uh
1:01:30 is
1:01:32 jogging past
1:01:37 a motel
1:01:45 during a dramatic storm.
1:01:53 Okay,
1:01:56 now for image and video. Generate
1:02:00 controllable and consistent media. Wait,
1:02:02 I can do this for video too.
1:02:06 I'm confused.
1:02:11 I don't
1:02:14 No, I don't see it.
1:02:20 Ann's here.
1:02:23 Ann,
1:02:24 I'm making another embarrassing image of
1:02:27 you. Not because I was particularly
1:02:29 thinking of making an embarrassing image
1:02:30 of you, but because the assets were just
1:02:33 prepopulated from the last time I did
1:02:35 this. Um, apparently this tool got
1:02:38 better since since I made you remember
1:02:40 the one I made of you smoking in the
1:02:42 parking lot. different than the one with
1:02:44 you in Central Park with William Defoe.
1:02:49 This is This is that other one. All
1:02:51 right, so let's let's see if we can get
1:02:53 you in a red 70s jogging suit with white
1:02:57 stripes jogging past the motel. All
1:03:00 right, generate. What? What just
1:03:02 happened? Why did that do that?
1:03:05 Do I need to sign in?
1:03:08 Oh, come on, man.
1:03:15 Uh,
1:03:17 chat mode beta.
1:03:23 I'm confused. Where's my [ __ ]
1:03:33 Ah, okay.
1:03:42 Tik Tok.
1:03:44 Make sure I'm wearing a sweatb band. Oh
1:03:46 yeah, you should definitely be wearing
1:03:49 like a pink bow sweatb band correctly.
1:03:52 Like I wear it right here. This is By
1:03:55 the way, ladies, if you're looking at my
1:03:57 pink bow and you think like I don't know
1:03:59 how to wear a headband, you're wrong.
1:04:01 This is how you wear it.
1:04:04 Okay.
1:04:07 It It It looks like
1:04:14 All right. There's Sasquatch.
1:04:24 Start with tool view instead. Oh, here
1:04:27 we go. But that doesn't look like
1:04:31 that's not Ann. It's not even close.
1:04:34 And that's not the hotel.
1:04:36 What did I do wrong? This is user error
1:04:39 reference.
1:04:42 References.
1:04:44 Woman in red jogging suit.
1:04:52 Uh oh. What did I do?
1:04:55 [Laughter]
1:05:00 What did I do? People
1:05:02 copied the clipboard. Is that
1:05:04 references? Okay, there's references.
1:05:07 Oh, I see what I did wrong. Okay, calm
1:05:10 down everyone.
1:05:12 Listen,
1:05:18 sh listen.
1:05:20 This is not my fault. Okay.
1:05:24 So, I think what you have to do is you
1:05:26 have to select Yeah. Okay. So, okay. So,
1:05:31 down here
1:05:34 when when when you're in runway, down
1:05:36 here is a little library of saved
1:05:39 images. Okay.
1:05:42 And then you click on references. And it
1:05:44 looks like here's your references,
1:05:46 right? But no, you actually have to
1:05:48 select them. So, we're going to select
1:05:50 the image of an, we're going to select
1:05:53 the rainstorm. And now, we're going to
1:05:54 select the hotel motel. And now we've
1:05:58 got that same prompt and now it's going
1:05:59 to work. So just calm down.
1:06:04 I know if you're out there like, I
1:06:06 thought he was be supposed to be some
1:06:07 sort of expert and now he's just sitting
1:06:10 there looking like a fruit cake with
1:06:12 pink bow on his head and he doesn't even
1:06:13 know how to do anything. How am I
1:06:15 supposed to learn something?
1:06:18 Right. I understand that. You're not
1:06:21 wrong. You're not wrong on many counts.
1:06:27 All right, but now we're going to make
1:06:29 an image of Ann running in front of that
1:06:32 motel at that dramatic storm in her red
1:06:35 jogging suit. It's going to be awesome.
1:06:39 Shut up. Shut up. I know what I'm doing.
1:06:42 I know what I'm doing.
1:06:46 I'm the expert here. Listen, don't you
1:06:50 give me lip. Don't you that is he even
1:06:53 qualified? Don't ask that question that
1:06:56 you don't ask a question you don't want
1:06:58 to know the answer to. All right? You're
1:07:00 the one spending time here, not me. I'm
1:07:04 just here.
1:07:06 This just I just This is just like a
1:07:08 thing. You're choosing to be here to
1:07:12 watch this.
1:07:15 It's nothing good can come of this.
1:07:19 All right. Did we get did we get an
1:07:22 Yeah.
1:07:26 It must it must be like you in your
1:07:28 basketball days.
1:07:34 Look at the hair. That's awesome.
1:07:43 Oh my god. That is [ __ ] terrific.
1:07:48 That is [ __ ] terrific. Okay.
1:07:52 [Laughter]
1:07:55 Big Bunny Motel. Okay, this is the one.
1:07:58 This is the one. Big Bunny Motel. And
1:08:00 now we're going to make video of Ann
1:08:02 jogging. Okay,
1:08:10 listen. Listen, Ed. Please, please,
1:08:13 please, please tell me you haven't eaten
1:08:17 gummies tonight because this is not
1:08:20 going to go well for you. This is not
1:08:22 good.
1:08:27 I feel I feel I I proactively feel bad
1:08:31 for you if you're on gummies right now,
1:08:33 but we are going to have you jog.
1:08:40 Should we do 10 seconds?
1:08:44 I You know what? We're going to spend
1:08:45 some money on an We're not going to do
1:08:47 Gen 4 Turbo. We're going to do Gen 4.
1:08:51 We'll go. We'll go. woman
1:08:56 um jogs toward the camera
1:09:02 as it tracks
1:09:07 back,
1:09:10 following her
1:09:14 as if
1:09:17 she were in a marathon.
1:09:26 Oh my god, this is going to be good. All
1:09:28 right, are we going? Yes.
1:09:31 Your video is generating an explore mode
1:09:37 to retro jog Stormy Motel. Okay.
1:09:44 Descent Satum Saturn. Descent Saturn
1:09:47 seems to be enjoying the entertainment
1:09:50 tonight.
1:09:54 Is this V3? This is not V3.
1:09:58 This is
1:10:03 This is runway ML their references model
1:10:09 where you can input three different
1:10:11 images and it will sort of combine them
1:10:13 into a cohesive thing. Apparently, it's
1:10:15 gotten better at like product
1:10:17 consistency. Like if you put, you know,
1:10:19 like, you know, a cup of, you know, a
1:10:22 Coca-Cola,
1:10:25 um, it'll maintain the coherence of this
1:10:28 better. Um, while this generates, go
1:10:30 grab an audio overview of your deep
1:10:32 research. Yeah. Let's see if the deep
1:10:34 research did its thing. Is it done?
1:10:37 Researching 110 websites. Wow.
1:10:40 Still going.
1:10:43 Amazing.
1:10:45 All right. Hey, maybe it'll come up with
1:10:47 some way for us to to organize tools.
1:10:52 42%.
1:10:54 Are you ready, Ann? Did Did Ann tell us
1:10:57 whether or not she is gummyrific?
1:11:01 She She's probably just She's like, I
1:11:04 can't. She probably walked away.
1:11:07 Probably giving giving the dog a
1:11:09 lollipop or something like that.
1:11:20 Champ, what are you doing? Why are you
1:11:22 laying on your feet like that? He's so
1:11:24 funny. He curls his He's because he's
1:11:26 got I think it's the beagle in him. He
1:11:29 curls his feet up like this. I'm like,
1:11:32 it's very cute. All right, we've almost
1:11:35 got woman jogging.
1:11:38 She looks good.
1:11:40 It's really funny. It's actually really
1:11:42 funny.
1:11:46 Crappy [ __ ]
1:11:52 So, can I do that with
1:11:55 Yeah, you can't do it. You can't do
1:11:57 reference. You can't generate a video
1:12:00 using references,
1:12:03 but you can generate an image using
1:12:05 references. All right.
1:12:07 But it definitely it definitely put her
1:12:09 in a different outfit. successfully.
1:12:18 She leads marathon. That's super cute. I
1:12:21 know. She looks good. Okay, here we go.
1:12:23 Ann, are you ready?
1:12:30 This is going to be so janky.
1:12:33 Wow. Gemini did a great job at
1:12:35 researching a character profile and
1:12:37 story arcs. Much better than 4.5. Oh,
1:12:39 cool. Good. Awesome. All right, here we
1:12:43 go. This is This is Ann Murphy
1:12:48 training for the Boston Mer Training for
1:12:50 the Portland Half Marathon. Training for
1:12:52 the Portland Half 5K.
1:13:02 It's sort of tracking with her. Her face
1:13:04 keeps changing.
1:13:10 Huh. Her hair got shorter, face is
1:13:13 changing.
1:13:15 It's in the neighborhood of, you know,
1:13:19 not horrible. Hollywood isn't going to
1:13:21 like isn't going to like the
1:13:24 democratization of blockbuster movies.
1:13:26 They're not. But you know what they're
1:13:28 going to start doing, Robbie, is they're
1:13:29 going to start making blockbuster movies
1:13:31 using AI. I mean, Lionsgate cut a deal
1:13:35 with Runway. So, the company that did
1:13:36 this and then the other thing I heard
1:13:39 this past week is Runway is going to
1:13:42 allow you to generate full-on video
1:13:45 games. So, pay attention to Runway.
1:13:47 She's She looks like she's running in
1:13:48 front of a green screen. She does. It's
1:13:51 a little weird,
1:13:53 but you know, her Yeah, her speed
1:13:55 doesn't match. The physics of this are
1:13:57 weird.
1:14:02 Like, it should be Yeah. And that could
1:14:04 be my prompt.
1:14:06 Let's just do um
1:14:09 woman jogs.
1:14:12 Let's just say woman
1:14:15 raises
1:14:16 her hands
1:14:19 in
1:14:21 victory
1:14:23 crossing
1:14:26 the
1:14:29 big bunny.
1:14:35 The big bunny half
1:14:40 5k
1:14:41 [Laughter]
1:14:48 annual
1:14:50 race.
1:14:54 All right. And we'll do five seconds
1:14:55 because 10 seconds
1:14:57 struggles with physics.
1:15:04 All right. Tik Tok pin for Monday
1:15:06 Meltdown. Okay. I'm worried
1:15:11 for Gen AI to be honest. Disney suing
1:15:14 Midjourney
1:15:15 and these other actions. So I wouldn't
1:15:17 be that that worried issue. Um couple of
1:15:21 things. Um
1:15:24 AI is not going anywhere. This is all
1:15:27 about who gets paid,
1:15:29 who gets some of the sweet, sweet moola
1:15:32 that that uh is going to these big
1:15:34 frontier companies. So, companies like
1:15:37 Disney are going to do what they can to
1:15:38 get it. The fact that Disney
1:15:41 is suing Midjourney, which is a small
1:15:44 independent, privatelyowned company,
1:15:48 for
1:15:51 copyright infringement
1:15:53 on the training of the data
1:15:58 while simultaneously
1:16:00 allowing the first
1:16:02 AI generated commercial to debut on ABC,
1:16:07 was it the NBA final?
1:16:10 something like that, some big AI
1:16:13 generated thing. So they actually own
1:16:15 have an interest in
1:16:18 um AI generation technology and a
1:16:23 distribution platform that they own
1:16:27 um that they're using while they're
1:16:31 suing a small independent company.
1:16:34 Okay. So that's so first of all the
1:16:40 level of hypocrisy in that lawsuit is
1:16:44 expected.
1:16:48 U the other thing is a um the was it the
1:16:51 open AI case? No, it was the anthropic
1:16:53 case. So, so a judgment just came down
1:16:57 last week in the Anthropic case which
1:17:00 was a similar sort of thing um where
1:17:03 they were being sued and what the judge
1:17:05 basically said was
1:17:09 training
1:17:11 training data into generative AI is
1:17:15 transformative
1:17:17 meaning it falls under fair use. Right?
1:17:20 So if you start with a book,
1:17:23 the way these models are trained, you
1:17:25 start with a book, you run it through
1:17:27 the transformer architecture and it gets
1:17:29 embedded, that book ceases to exist. It
1:17:33 it literally takes that book and breaks
1:17:35 it into what are called tokens, which
1:17:37 are fractions of words and punctuation
1:17:40 and spaces
1:17:42 and it
1:17:44 places them in thousand dimensional
1:17:46 mathematical space in semantic clusters.
1:17:50 What that means doesn't matter.
1:17:53 What it does mean is that the original
1:17:55 work cease ceases to exist. So it's
1:17:59 transformative. Therefore, it's fair
1:18:01 use. What that judge also said was if
1:18:04 you buy the books legally and then train
1:18:06 your models on it, then it's fair use.
1:18:09 If you stole the books, then that's not
1:18:12 okay because you can't steal [ __ ] but
1:18:14 you can pay for a bunch of books and put
1:18:16 them in there. These court cases are
1:18:19 probably going to go on for 20 years.
1:18:21 And I don't think anyone is going to
1:18:25 like I don't think any one of these
1:18:26 lawsuits is going to significantly slow
1:18:28 down AI at all. At all. So anyway, all
1:18:31 right. So we have we we now have Ann
1:18:34 celebrating in victory as she crosses
1:18:36 the finish line. All right, Ann, are you
1:18:38 ready? Which case was this? It was the
1:18:40 anthropic case.
1:18:43 Um,
1:18:46 yes, she did it.
1:18:49 Yes, she turned into some TV soap
1:18:54 actress that doesn't look like an,
1:18:58 but
1:19:01 she did celebrate. Yay, you did it.
1:19:07 Fairy tale revolution. Like she's doing
1:19:09 the chaa. Oh man. Where she ran out of
1:19:13 the motel.
1:19:16 That was an awesome afternoon.
1:19:23 Oh man. All right, let's go look at our
1:19:25 deep research. What do we got? We got
1:19:27 over here. Are we done? No. Researching.
1:19:30 Wow, this is still going. 292 websites.
1:19:34 Can't believe they sued Midjourney
1:19:35 before Grock with Smoking Mickey. I
1:19:37 know. I know. Well, but you know, Grock
1:19:41 and and uh
1:19:44 Elon Musk are well funded. I you know, I
1:19:47 I I think it's I think they're suing
1:19:50 MidJourney because MidJourney probably
1:19:53 won't be able to defend itself.
1:19:56 Why not use V3, the best video model on
1:19:58 the market? Oh, just because Jessica, I
1:20:01 I agree with you. Although so
1:20:04 interesting funny thing happened on the
1:20:06 way to V3 becoming the best video model
1:20:08 in the mark in the market. I agree with
1:20:12 you and until
1:20:16 eight or nine days ago
1:20:19 unless someone came out with a video
1:20:21 model that does acting and all the
1:20:24 physics really good I would say you know
1:20:27 nothing could hold a candle to it. And
1:20:29 then midjourney video came out. And I
1:20:31 think midjourney video
1:20:34 is is a very different tool than V3, but
1:20:38 in a lot of ways I like it better. But
1:20:40 it doesn't ha it doesn't have text
1:20:43 consistency. It doesn't have voice,
1:20:45 sound effects, acting, all the stuff
1:20:47 that V3 has. But there's something about
1:20:51 how it interprets images that's like
1:20:53 nothing I've ever seen. It's very It's
1:20:55 very magical.
1:20:57 The reason we're using RunwayML right
1:20:59 here is the the references feature where
1:21:03 you can you can put multiple objects
1:21:07 together and and stitch them into a
1:21:10 scene um together. So that's that's the
1:21:13 only reason I was just testing the the
1:21:14 coherence of of the references feature.
1:21:18 Um but yeah, I agree with you. V3 is
1:21:20 amazing. All right. Uh that that looks
1:21:22 like it's going to take forever.
1:21:24 forever.
1:21:27 What a great morning jog, Ann.
1:21:40 Oh man.
1:21:41 Um,
1:21:44 for let's see who was asking about that
1:21:48 for Ashue. Robbie, maybe James Cameron
1:21:51 needs to make some AI software for
1:21:53 everyone to use.
1:21:55 Um,
1:21:57 I would be I would actually be very
1:21:59 surprised if James Cameron isn't making
1:22:04 some AI software or or at least learning
1:22:06 it. I mean, Cameron's Cameron's really
1:22:08 into this stuff. Um,
1:22:12 who was it
1:22:14 for V3? They had Darren Arnowski,
1:22:18 Areronowski, I forget how to say his
1:22:20 name. Um Darren Areronowski, a
1:22:22 filmmaker, um had a lot of input into
1:22:26 the V3 tool, the flow tool that they
1:22:28 built, um for filmmakers. Um just like
1:22:33 you've got, um what's his name?
1:22:35 Timberland, the the music producer
1:22:37 Timberland is actively using Sunno, and
1:22:41 I think he's now a spokesperson for
1:22:42 Sunno, the music generation tool. He's
1:22:45 using it in the context of music
1:22:47 production. He's not saying it's this is
1:22:53 I am growing weary.
1:22:56 I am growing [ __ ] weary
1:22:59 of the tropes around you know AI
1:23:03 generated video isn't real art or it
1:23:06 isn't real video because the assumption
1:23:08 is that you just push a button and out
1:23:10 squirts a video. That's all of the good
1:23:13 videos that you see are crafted just
1:23:16 like a film is crafted. And they're
1:23:18 using generative AI as a tool, a a
1:23:21 production tool in a mix of lots of
1:23:24 other things like writing and and
1:23:27 assembly and you know all sorts of
1:23:30 things.
1:23:32 So,
1:23:34 so yes, there are going to be people who
1:23:37 piss and moan for the next 20 years how
1:23:40 AI ruined the movie making business,
1:23:43 but they talked about this when Final
1:23:45 Cut Pro came out, right? Digital
1:23:47 editing, nonlinear editors, they said,
1:23:50 "We're going to be the death of cinema."
1:23:52 When when, you know, high quality
1:23:54 digital cameras came out, it's going to
1:23:55 be the death of cinema. It's like this
1:23:58 is this is no different. Well, it is
1:24:02 different because the generative part of
1:24:04 it is [ __ ] insane. Um, and and we're
1:24:08 we're probably within three years of
1:24:10 being able to just speak a film into
1:24:13 existence.
1:24:14 But, but to think that Hollywood is is
1:24:17 is going to fight that rather than
1:24:20 embrace it is ridiculous.
1:24:22 The producers are gonna they're going to
1:24:24 suck this stuff up. So anyway, it's
1:24:28 we're we're in crazy crazy times. Tik
1:24:31 Tok pin. There's no pin there
1:24:36 today.
1:24:39 Let's see. It took 80 minutes, but here
1:24:41 it comes.
1:24:44 That's going to It's going to get
1:24:45 faster, too. It's easier to fall back on
1:24:48 AI's theft and stop there for some
1:24:51 people. Yeah, I listen. I agree. I think
1:24:54 all of the tropes, AI is theft, so I'm
1:24:56 not going to use it. Um, AI is the
1:24:59 world's greatest plagiarism machine. Not
1:25:01 going to use it. Um, AI makes mistakes.
1:25:04 Uh, that doesn't look like real video.
1:25:06 That doesn't look like real film. The
1:25:07 physics are all wrong, so I'm not going
1:25:10 to use it. Um, it's taking all of our
1:25:12 jobs, so I'm not going to use it.
1:25:16 All of those conscientious objectors to
1:25:19 AI
1:25:21 that are using all of those tropes to
1:25:24 stay on the sideline. Do you know who
1:25:27 they're hurting?
1:25:29 They're not hurting AI.
1:25:32 AI isn't slowing down.
1:25:35 They're hurting themselves.
1:25:38 Because I promise you, I promise you
1:25:42 that talented storytellers
1:25:46 who have some element of craft
1:25:49 that learn AI
1:25:52 are going to be able to kick ass with
1:25:54 AI.
1:25:56 I talk about this all the time that
1:26:01 the people sitting on the sidelines are
1:26:03 treating AI like it's this genius
1:26:07 thing that is in opposition to humans.
1:26:10 Oh, AI is making movies.
1:26:13 But that's not actually what it is. What
1:26:15 AI is is it's an amplifier of whatever
1:26:18 you put into it.
1:26:20 And if you choose to put into AI
1:26:24 your ideas, your craft, your
1:26:30 curation, your taste,
1:26:33 when it gives you something back, you're
1:26:35 going to have a point of view on that.
1:26:36 You're going to say, "No, that's not
1:26:37 good enough. Let me make it better." and
1:26:39 you get into this wrestling match with
1:26:41 AI,
1:26:43 which is almost identical, but much much
1:26:47 faster
1:26:48 than being in a collaborative rel
1:26:50 relationship with another person. Oh, I
1:26:53 have an idea for a movie. Here's the
1:26:55 footage I shot. Well, let me give it to
1:26:57 the editor. The editor goes off and do
1:26:59 does some stuff with it and comes back
1:27:01 and shows you what what they did. And
1:27:03 you're like, no, that's not quite it.
1:27:04 The colors are all wrong, and I was
1:27:06 hoping for more fast cuts. then they go
1:27:09 off and do it again. Well, that's
1:27:10 exactly what you could do with AI.
1:27:14 It's not this thing off to the side.
1:27:16 It's an amplifier of you. And so, what's
1:27:20 going to happen is the people that get
1:27:21 curious about AI are going to accelerate
1:27:26 their capacity to generate films, to
1:27:29 make stories, to make great stories. And
1:27:33 the people that are on the sidelines
1:27:34 that are like AI's evil are going to get
1:27:37 [ __ ] steamrolled.
1:27:41 And it's like they're bitter now.
1:27:47 Wait until two years from now when all
1:27:50 of their buddies that were like, "Ah,
1:27:52 [ __ ] it. I might as well learn it."
1:27:55 now are are, you know, being courted by
1:27:58 the studios to do major motion pictures
1:28:00 because they're launching a major motion
1:28:02 picture a day
1:28:06 and they need people that are AI
1:28:07 literate.
1:28:09 Then all the people that are on the
1:28:10 sidelines today are just going to have
1:28:11 their asses handed to them. And it's
1:28:13 it's, you know, that's why I do this
1:28:16 channel because I want as few people as
1:28:18 possible sitting on the [ __ ]
1:28:20 sidelines. You don't have to like AI.
1:28:23 You can [ __ ] hate it,
1:28:26 but don't ignore it. Don't not learn it.
1:28:32 Learn it.
1:28:38 Learn what it actually does. Learn what
1:28:40 it actually is.
1:28:42 Learn how it actually [ __ ] works.
1:28:46 It is technically
1:28:49 definitionally not a plagiarism machine
1:28:52 because it's not copying.
1:28:56 Meltdown Monday. Here we go, baby. All
1:28:58 right. There was supposed to be a Tik
1:28:59 Tok pin here. What's What was the pin
1:29:01 that was up? I'd like to show the
1:29:04 naysayers my library of versions before
1:29:06 things felt right. Exactly. AI producer.
1:29:09 The ones that are being creative with it
1:29:11 today are the new filmmakers of
1:29:12 tomorrow. Exactly. Look at Kelly Bosch.
1:29:16 Look at Joy Pury.
1:29:19 Right.
1:29:20 You You've got You've got people today
1:29:24 who were not filmmakers,
1:29:28 but who have stories to tell.
1:29:32 Both Joy Perie and Kelly Bosch in
1:29:35 different interviews effectively said
1:29:37 the same thing. Someone asked, Kelly
1:29:39 Bosch was speaking at an AI salon,
1:29:42 I don't know, a year ago, year and
1:29:44 change ago,
1:29:47 and like the images that she makes are
1:29:49 just so [ __ ] remarkable.
1:29:53 And someone said, "Kelly, how do you
1:29:54 prompt it to get that stuff?"
1:30:00 And Kelly's response was [ __ ]
1:30:02 brilliant. She goes, "I've always had
1:30:05 these images in my head. I've never been
1:30:09 able to get them out.
1:30:15 That's [ __ ] genius.
1:30:18 AI
1:30:20 is not making images
1:30:24 that Kelly Bosch goes, "Oh, I'll take
1:30:28 one of those." AI is not the genius.
1:30:31 Kelly
1:30:33 has had these ideas in her head since
1:30:35 she was a little girl and never had a
1:30:38 way to get them out.
1:30:41 And along comes AI and she can go she
1:30:44 can say, "Huh,
1:30:46 what if I could get it to do this?" And
1:30:49 then it generates stuff. And then she
1:30:51 dials up the stylization number and it
1:30:53 weirds it out.
1:30:56 And over a series of months and now
1:30:59 years,
1:31:01 she knows how to use these tools to get
1:31:03 the [ __ ] in her head out of it.
1:31:08 Joy Perie said the same thing. Someone
1:31:10 asked her like, "Why are you making all
1:31:12 these films now? What?" She said, "I've
1:31:14 always had stories to tell. I've never
1:31:17 had a way to tell them."
1:31:21 That's the future of film making.
1:31:25 And it used to be the prerequisite to
1:31:28 film making was you had to pick a single
1:31:31 discipline
1:31:32 and you had to go to [ __ ] USC
1:31:35 or Colombia. If you wanted to be a
1:31:37 screenwriter, go to Colombia. You could
1:31:39 learn the screenwriting thing. You want
1:31:41 to learn cinematography, go to USC,
1:31:45 right? Or go intern for for Spielberg
1:31:48 or, you know, one of the biggies.
1:31:52 go work for Roger Corman and you would
1:31:54 learn your discipline. You would get in
1:31:56 your union and you would do your thing.
1:31:59 You would be a cog in the big wheel.
1:32:02 And maybe if you were a Quinton
1:32:03 Tarantino, you'd be like, "Fuck it. I'm
1:32:06 going to go make my own thing. I'm going
1:32:07 to do everything."
1:32:11 Now
1:32:14 that prerequisite is gone.
1:32:20 The only prerequisite is you've got to
1:32:22 be curious about AI enough to understand
1:32:24 what's [ __ ] possible and you have to
1:32:26 have stories to tell.
1:32:32 If you can take that curiosity, a little
1:32:34 bit of [ __ ] mojo,
1:32:38 little bit of tenacity, a little bit of
1:32:40 taste,
1:32:42 go tell your story is going to rise
1:32:45 above the noise.
1:32:52 That's not real films. AI films are
1:32:54 theft.
1:33:02 Okay.
1:33:05 Do you think it's going away?
1:33:08 Okay.
1:33:11 You know, your buddies are actually out
1:33:13 making films right now. Well, I'll get
1:33:18 I I just
1:33:24 I have deep deep deep deep
1:33:28 empathy
1:33:30 for how [ __ ] terrifying this is,
1:33:35 for how devastating it must be if you've
1:33:38 got a 20 or 30 year career doing a
1:33:42 single thing. Like, can you imagine
1:33:43 being a special special effects guy or a
1:33:46 cinematographer or a lighting person
1:33:48 right now and you look at what's coming
1:33:50 out of V3? You're like, "Well, [ __ ]
1:33:53 Who's going to hire me?
1:33:57 You should be learning VO3,
1:34:01 right? I get that it's painful. I get
1:34:03 that it's [ __ ] up. I get that
1:34:05 everything's going to change. I do not
1:34:07 get
1:34:09 people just being pissed off and sitting
1:34:11 on the sidelines.
1:34:14 It's not going to slow it down. It's not
1:34:16 going to change a [ __ ] thing. Anyway,
1:34:18 all right.
1:34:21 Meltdown Mondays. Living up to the hype.
1:34:25 God damn it. I've written seven
1:34:28 screenplays. Like, I I get it. I get it.
1:34:31 I'm in the middle of writing a musical
1:34:33 right now using AI about AI
1:34:37 that I want performed by human beings on
1:34:40 a stage in New York City.
1:34:46 There's a mind bender.
1:34:48 AI steals jobs. AI is going to ruin all
1:34:51 the jobs. Well, AI created a musical
1:34:55 that if I get it produced is going to
1:34:58 employ
1:35:00 human being musicians, human being
1:35:02 actors, stage managers, set designers,
1:35:05 lighting designers.
1:35:09 It's not an eitheror.
1:35:14 The world isn't [ __ ] black and white.
1:35:17 It's just a new technology. [ __ ]
1:35:19 learn it.
1:35:21 any updates on that, by the way? Yeah,
1:35:23 we actually um I can't really say who it
1:35:26 is, but a good friend of ours is um
1:35:31 is a writer who's
1:35:34 won an Oscar and
1:35:38 done some very very major TV shows and
1:35:41 he read it, which is amazing that he
1:35:44 even read it. um and loved it and is is
1:35:48 uh potentially
1:35:49 gonna make some intros to um some folks.
1:35:53 So, that would be cool.
1:35:56 All right, you've just answered my
1:35:58 questions. This is so exciting. What are
1:36:00 your questions, Sophia? I've been afraid
1:36:02 of AI, but you just gave me hope. Yeah,
1:36:04 listen. Listen, you
1:36:08 Here's the Here's what I can promise
1:36:10 you. Here's what I can promise you,
1:36:13 Sophia. I
1:36:15 No one no one
1:36:18 can explain to you
1:36:22 what it is about AI that is so
1:36:24 profoundly powerful.
1:36:27 The only way you can understand
1:36:30 what AI is is to use it yourself
1:36:35 and to and to just play with it and play
1:36:38 with it and play with it and take the
1:36:40 things that you're passionate about and
1:36:43 put them into whatever AI tool you're
1:36:45 using. It literally doesn't matter. Chat
1:36:48 GPT or Midjourney or Runway ML that we
1:36:52 just played with or V3. It doesn't
1:36:54 matter.
1:36:55 take who you are and and figure out how
1:36:59 to ask it for something that you deeply
1:37:01 understand and at some point AI will
1:37:05 produce something back at you that will
1:37:08 literally melt your face. We we call it
1:37:10 in here your Kevin Mallister moment,
1:37:12 right? If you remember the movie Home
1:37:14 Alone, this moment,
1:37:19 that moment happens to me probably,
1:37:23 I don't know, once a week
1:37:26 where I I just I'm literally like, wait,
1:37:29 it can do that now?
1:37:32 Holy [ __ ]
1:37:35 There is no way to explain that feeling
1:37:38 and to to to to understand how profound
1:37:41 these tools are. So just go play. Just
1:37:44 go play. The other thing that is
1:37:47 impossible to articulate to the people
1:37:49 that are sitting on the outside, the
1:37:52 media
1:37:54 and Hollywood for the past 50 years has
1:37:56 essentially said AI's evil. AI will kill
1:38:00 you.
1:38:05 And so that's all they talk about. And
1:38:07 so it seems like this very binary thing.
1:38:11 But what we experience here with the
1:38:13 people that show up here night after
1:38:14 night and the people that are part of
1:38:16 the AI salon and the people that are
1:38:17 part of other communities like she leads
1:38:19 AI
1:38:21 is that people are reinventing
1:38:23 themselves using AI.
1:38:27 That what AI actually is is a humanity a
1:38:30 a a human being amplifier. It's like a
1:38:34 [ __ ] jetpack for your brain and your
1:38:37 and your capabilities.
1:38:40 If you can't program, now you can. If
1:38:44 you couldn't make movies before, now you
1:38:46 can. If you can't write music, now you
1:38:49 can. If you can't write screenplays, now
1:38:53 you can.
1:38:56 Well, what do you do with that?
1:38:59 I don't know. Have a [ __ ] idea.
1:39:02 Learn what's possible and go execute
1:39:05 that idea. I've written a musical over
1:39:09 the past year.
1:39:12 I haven't thought about writing
1:39:14 something for the theater in 30 [ __ ]
1:39:17 years.
1:39:19 And without AI, I would never in my life
1:39:24 have considered writing a musical. I
1:39:26 don't have the talent. I don't know the
1:39:28 right people.
1:39:30 I know it's a [ __ ] slog.
1:39:33 The other thing you need to do is get
1:39:37 your ass in a community
1:39:39 like the AI salon. Can you pop up the uh
1:39:42 the uh the the URL there? If you're not
1:39:46 part of the AI salon, go to the salon.ai
1:39:49 and click on join our community and
1:39:51 that'll take you over to the community
1:39:53 site. Sign up. It's free.
1:39:57 Introduce yourself, connect with people.
1:39:59 There's all sorts of areas there.
1:40:03 There's, if you want to dig deeper,
1:40:05 there's a there's a subscription area
1:40:07 called the AI Salon Mastermind where
1:40:10 people who are like really committed to
1:40:12 like upping their game are getting in a
1:40:14 more focused, concentrated sub
1:40:16 community,
1:40:19 but the AI salon itself is free and will
1:40:22 be
1:40:24 and it's full of remarkable generous
1:40:26 people.
1:40:28 All right.
1:40:31 Okay, let's look at let's look at Okay,
1:40:35 welcome to chat add. And I know we got
1:40:37 to get out of here. It's getting late.
1:40:38 Um,
1:40:43 about an hour ago, I kicked off a deep
1:40:45 research thing. How long did it take?
1:40:47 Does it tell us in here how long it
1:40:49 took?
1:40:50 I don't know. Build on top of this.
1:40:52 Create something new based on this, like
1:40:54 an audio overview.
1:40:57 All right. Create
1:41:02 I want an audio overview.
1:41:05 How do I make an audio overview?
1:41:11 All right, let's see what it did.
1:41:12 Navigating the post app era, a strategic
1:41:15 framework for organizing and dis
1:41:17 discovering.
1:41:19 Sorry, the voice is going. Oh, AI salon.
1:41:23 Cool. Yeah, a AI salon's really good.
1:41:30 Okay. Um, it's I'm I'm really glad we've
1:41:33 got some new folks in here tonight.
1:41:34 That's exciting.
1:41:36 Um, okay. The paradigm shift from tool
1:41:39 centric directories to goal oriented
1:41:41 ecosystems. I like this.
1:41:44 The challenge of organizing a the
1:41:47 rapidly expanding universe of artificial
1:41:49 intelligence is not merely a matter of
1:41:52 categorization. It's a fundamental
1:41:53 problem of information architecture in
1:41:56 an era of unprecedented technological
1:41:58 convergence. Yes, this is what I've been
1:42:01 saying people.
1:42:03 [Laughter]
1:42:06 The very nature of AI platforms is
1:42:08 rendering traditional directory models
1:42:10 obsolete, demanding a paradigm shift.
1:42:12 Okay. The inevitable collapse of
1:42:14 functionbased categorization. Fine.
1:42:18 Introducing a new parag a new paradigm.
1:42:21 jobs to be done framework. Oh, I like
1:42:24 that.
1:42:26 The jobs to be done framework.
1:42:30 To build a system that reflects how
1:42:32 users actually think and work, the
1:42:34 organizational paradigm must shift from
1:42:37 what tools are used to what users do.
1:42:40 The jobs to be done framework, a theory
1:42:43 of innovation popularized by strategists
1:42:45 like Tony Ulwick and the late Harvard
1:42:49 professor Clayton Christensen. Yes, know
1:42:51 him provide the ideal conceptual
1:42:54 foundation for this shift. The core
1:42:55 premise of jobs to be done is that
1:42:59 customers do not buy products. They hire
1:43:02 them to get the job done.
1:43:05 Huh, I like it.
1:43:10 The jobs to be done framework compels a
1:43:13 deconstruction of this highle job into
1:43:15 its constit constituent steps or tasks.
1:43:19 The animated Bigfoot video job can be
1:43:22 broken down into generate a visual
1:43:25 concept, write a short engaging script,
1:43:27 generate a compelling voice over for the
1:43:29 script. Right? This is what we just
1:43:31 talked about. Animate the static
1:43:33 character based on the thing and then
1:43:36 you combine them all together to make
1:43:38 the animation. With VO3, you don't
1:43:40 really need all of those, but that's
1:43:42 okay. If you're doing it with something
1:43:44 like Hedra, you do. This deconstruction
1:43:46 reveals that the user does not need a
1:43:49 single tool. They need a workflow.
1:43:52 Very cool. From user intent. All right.
1:43:55 So, let's turn this. How do I turn this
1:43:56 into a mastermind? Anybody? Into a into
1:44:00 audio overview.
1:44:07 Oh, Vicki, we use jobs to be done at
1:44:09 work. That's funny. Ask it in the chat
1:44:11 hole.
1:44:13 Can you turn this research
1:44:18 into an audio
1:44:21 overview?
1:44:24 Just a sec.
1:44:33 The proposed information architecture is
1:44:36 built upon four primary entities that
1:44:40 form a hierarchical relational data
1:44:42 model.
1:44:44 Oh, you know what? Oh, oh, oh.
1:44:48 People. People.
1:44:57 Okay.
1:45:06 It's almost it's almost like the
1:45:08 directory would be a directory of
1:45:10 outputs,
1:45:12 right? And you could like you could
1:45:14 search for like Yeti video and what you
1:45:17 would get is a bunch of Yeti videos
1:45:21 where whoever made the Yeti video gives
1:45:24 you their chain. What tools did they use
1:45:26 to make that video?
1:45:30 That would be kind of cool because then
1:45:32 you could say
1:45:35 for voiceovers,
1:45:40 you know, which are which are a task in
1:45:42 the chain,
1:45:44 you know, 60% of people on this site use
1:45:47 this voice over tool, right? Use this
1:45:49 voice generation tool.
1:45:55 I can't help with audio overviews.
1:45:58 You can add this report to notebook LM.
1:46:00 Yeah, that's a good idea. Export. Okay,
1:46:02 fine. So, export to Docs.
1:46:09 Yeah, this is really cool.
1:46:13 That's loading.
1:46:31 Oh, look here. It's got So, wait a
1:46:34 minute. Create
1:46:35 a web page. Um,
1:46:39 I want you to turn this into an app
1:46:46 where a user
1:46:51 can share something
1:46:54 they created
1:47:00 and then share
1:47:09 the steps
1:47:12 and tools
1:47:15 they used to make it.
1:47:19 I want you to
1:47:25 augment my
1:47:28 crappy description here
1:47:33 with
1:47:36 um
1:47:38 your insights
1:47:41 based on the research.
1:47:45 I think what would also be really
1:47:47 interesting is to run this over to
1:47:50 Claude and have Claude do this and run
1:47:52 it over to chat GBT and have it do this
1:47:55 as well.
1:47:57 Um, but let's let's go ahead and boom.
1:48:05 All right. So, here's this.
1:48:10 So, we're going to copy this
1:48:14 We're going to go to Notebook LM.
1:48:18 Um, if you don't know what Notebook LM
1:48:21 is, notebooklm.google.com.
1:48:24 It is a quite remarkable
1:48:30 tool from Google. Let's see. I'm going
1:48:32 to go Google Docs.
1:48:37 Uh maybe I need to save this as a
1:48:40 organizing multimodal AI tools. Okay.
1:48:46 Organizing multimodal AI tools. Insert.
1:48:50 So basically you can put up to 50
1:48:52 documents into a notebook LM document
1:48:55 and then you can do all sorts of things
1:48:56 with it. So we're going to do a deep
1:48:58 dive conversation. We're going to go
1:49:00 generate
1:49:04 and this is going to turn. So, okay. So,
1:49:07 we just had Gemini go research, I don't
1:49:11 know, 400 websites about how to think
1:49:14 about
1:49:16 how you even organize AI tools because
1:49:19 you can't
1:49:21 into this new jobs to be done framework,
1:49:24 which is cool.
1:49:26 And now we're having Gemini make that
1:49:28 application. And now we're creating a
1:49:31 podcast about this new concept.
1:49:34 This is why I can't go to bed either.
1:49:36 Drop it in notebook. LM. This is a
1:49:39 problem. This is a real problem.
1:49:43 I
1:49:48 go.
1:49:54 Isn't this crazy? The world we live in,
1:49:57 people I mean, it it really is just we
1:50:01 just do live in remarkable times. I I
1:50:04 hate to break it to you, but we do.
1:50:07 Share your AI workflow. AI workflow
1:50:10 showcase.
1:50:12 Discover the how behind the wow.
1:50:16 You know what? Do you know what this app
1:50:18 is?
1:50:30 I think it was Silverf Fox said it that
1:50:33 that um like you should see how many of
1:50:37 my images hit the cutting room floor
1:50:38 right before I shared the thing. You
1:50:40 should see how many crappy things I
1:50:42 created. A site like this could be a
1:50:45 real argument against the [ __ ]
1:50:48 doomers that are just like AI is for
1:50:51 lazy people. Oh yeah, go look at the
1:50:54 77step process that that made that 30
1:50:58 second video you just made fun of and
1:51:02 you tell me if you could do that,
1:51:05 you [ __ ] tool.
1:51:10 [Laughter]
1:51:12 Share your creation. Okay, so this is
1:51:14 starting with data. I don't want the
1:51:16 [ __ ] data. What I want is I want to
1:51:21 lead
1:51:23 with
1:51:25 the user
1:51:28 uploading
1:51:30 or linking to
1:51:34 some finished
1:51:37 product.
1:51:39 And then based on the products,
1:51:46 the directory tool
1:51:51 interviews
1:51:53 them
1:51:56 to reveal
1:52:00 the tools they used. the tools
1:52:07 and specific
1:52:10 features.
1:52:12 They used
1:52:16 their human contribution
1:52:24 in
1:52:26 let's see including
1:52:30 curation
1:52:33 editing
1:52:35 refinement
1:52:42 tool bashing
1:52:45 uh etc.
1:52:51 And how they
1:52:55 assembled the final
1:53:00 deliverable
1:53:02 or project
1:53:04 Uh
1:53:06 uh uh
1:53:10 dude, this is a rabbit hole. I know. You
1:53:12 can also go back to the original canvas
1:53:14 and generate infographic for time. Let's
1:53:17 see if we get uh Is this done? Yeah,
1:53:20 this is done. 28 minutes. Okay. If you
1:53:23 haven't heard a notebook LM um what you
1:53:26 call it?
1:53:28 Interactive mode. Can we Does this
1:53:32 Wasn't this generating videos
1:53:35 download? No, just a wave file. Okay. I
1:53:38 thought they were gonna do it. Okay.
1:53:41 Um I'm going to I'm going to share
1:53:43 different and then we got to go. Listen,
1:53:45 I got to go people. I can't can't be
1:53:47 here all [ __ ] night.
1:53:50 Um,
1:53:52 but I want to play one of these things
1:53:54 for you because if you haven't seen them
1:53:56 before,
1:54:00 so Gemini did our research for us and
1:54:02 then we uploaded the research report
1:54:05 into Notebook LM and then we said make
1:54:08 an audio overview and then here's what
1:54:09 we got.
1:54:10 Welcome to the deep dive. We're here to
1:54:12 plunge into complex topics and hopefully
1:54:15 pull out some real insights. Have you
1:54:17 ever been really excited to try a new AI
1:54:20 tool, maybe for writing or images, maybe
1:54:23 even video, only to, you know, spend an
1:54:26 hour just bouncing between websites,
1:54:28 totally confused about which one
1:54:30 actually does what you need and does it
1:54:32 well?
1:54:32 Oh, absolutely. It happens all the time.
1:54:34 It Lonnie Viles, thanks for a great show
1:54:37 and a lot of great intel. Thank you very
1:54:39 much. Appreciate that. I appreciate it a
1:54:41 lot. Um, here's what I would encourage
1:54:44 you to do. Keep coming back. Um,
1:54:47 we call this channel chat add for a
1:54:50 reason. Um, it is it is guaranteed to be
1:54:54 unpredictable, occasionally useful,
1:54:57 hopefully always entertaining to some
1:54:59 degree. Um,
1:55:03 so so keep coming back to these. Um, and
1:55:06 also get involved with the community. So
1:55:08 join the AI salon. Tomorrow, the AI
1:55:11 Salon, we've got an AI salon presents
1:55:13 meeting where we've got the founder of
1:55:15 Jelly Pod, which is a tool that makes
1:55:17 podcasts like the one we're listening
1:55:19 to, but where you can control
1:55:21 everything. Um, he's going to be
1:55:23 speaking at the AI Salon tomorrow night,
1:55:26 Tuesday, at between 5 and 7 Mountain
1:55:29 time. Um, you should come to that. You
1:55:32 should come to those. Get to know people
1:55:34 in this community. It's incredibly
1:55:36 powerful. So, if you go to the the
1:55:37 salon.ai AI that was on screen right
1:55:39 there. Click join our community and then
1:55:42 click the events tab. You'll you'll see
1:55:44 an RSVP for the for the event tomorrow.
1:55:47 All right. And come see uh Preston from
1:55:49 Jelly Pod. All right. Let's uh
1:55:53 Yeah. And we will likely be late
1:55:55 tomorrow. Normally I go on at 8:00 p.m.
1:55:57 Mountain time here. It'll be 8:30 or 9
1:56:00 depending on when I get out of the uh
1:56:01 the salon. But anyway. All right. Oh,
1:56:05 Pearson. Sorry, not Preston.
1:56:08 Pearson from Jelly Pod. Um, okay. I
1:56:12 tried to use chat GBT for professional
1:56:14 registration. It failed at accuracy.
1:56:16 Yeah. Uh, if if you're if you're trying
1:56:18 to do things like uh things that require
1:56:23 accuracy, you're going to have to be uh
1:56:26 you're going to have to do a lot of
1:56:27 proof reading when you're using
1:56:29 generative AI. That will get better.
1:56:31 It's it's gotten better over the past
1:56:32 two years. That's still a major [ __ ]
1:56:35 headache with it. So yes, if you're
1:56:37 doing creative writing, let it go
1:56:39 hallucinate, right? That's how
1:56:41 creativity happens. Uh if you're trying
1:56:43 to get something that is predictable and
1:56:46 repeatable.
1:56:47 Yeah, that's it. All right, that's it.
1:56:50 Can someone put the link in the chat for
1:56:53 me, please? Link to what?
1:56:59 Often entertain. Oh, to the uh to the
1:57:01 Yes, hang on. To the AI salon. So, go to
1:57:05 the salon.ai.
1:57:08 Just go there
1:57:10 and then um scroll down. There's a
1:57:12 button that says join our community.
1:57:14 That'll take you over to the community
1:57:16 site and that's where all the
1:57:17 information about the uh
1:57:20 the event for tomorrow night is. Okay.
1:57:22 Beautiful. All right, everybody. Thank
1:57:26 you for uh a fun night. We got some
1:57:29 stuff done. Ann Murphy jogging out of
1:57:32 the hotel, the motel, sorry, hotel. A
1:57:35 little highfulutin
1:57:36 in her 70s jogging suit. That happened.
1:57:42 Tik Tok pin. Often entertaining, often
1:57:45 informative. I have information
1:57:47 overload. Can't process. I know,
1:57:49 Winston. It's a lot. It's a lot. I'm
1:57:52 really excited about this new uh this
1:57:54 new concept for uh what is it? Jobs
1:57:58 tobun approach to tool discovery. I like
1:58:02 this a lot. All right, we're gonna I
1:58:05 think over this week, so producer
1:58:07 Brandon, try to keep me on task, which I
1:58:10 know is [ __ ] impossible, but
1:58:13 I think what we might do is actually
1:58:16 take this concept and vibe code it into
1:58:19 a a platform that we could all start
1:58:22 using to share our projects. This is a
1:58:25 very AI salon.
1:58:27 Um,
1:58:29 it's very onbrand for the AI salon. this
1:58:32 whole approach where you say here's a
1:58:34 cool thing I made and then it helps you
1:58:36 unpack how to talk about it and then
1:58:39 what that becomes is a is a tool
1:58:41 recommendation directory where you can
1:58:44 either look for tools or features or
1:58:47 outputs. Oh, I want to do an output like
1:58:50 that. How do I do that? Here's 16
1:58:52 different ways you can do that. That's
1:58:54 really cool. We have that. It's called
1:58:56 the AI salon. Yeah, there's there's
1:58:58 literally a section called look what I
1:59:00 made and people share what they do. This
1:59:02 but this is just a a a searchable
1:59:04 version of that. That's very cool. All
1:59:07 right, everybody. Uh I'm going to get
1:59:08 out of here. It is uh yes, 10 o'clock.
1:59:11 Let's go. All right, peace out and see
1:59:13 you tomorrow at AI Salon Presents and
1:59:15 then I'll see you here um sometime after
1:59:18 that. Probably 8:30 or 9 Mountain time.
1:59:20 All right, peace out everyone. Bye.