05/08/2026 - Choosing the Right AI Tool Understanding the Strengths of ChatGPT Claude and Gemini

Description

Chapters

Transcript

0:04 [music]
0:09 [music]
0:16 [music]
0:26 [music]
0:34 Friday night, date night. Hello. Hello.
0:37 Hello.
0:39 Good evening. Good people. Let's see.
0:42 Oh, there I am. Beautiful. Hey
0:45 everybody. What's shaking? What's going
0:46 down? Happy Friday night date night.
0:49 Welcome to the show. I hope everyone is
0:51 doing well.
0:53 Um,
0:55 this is the new AI learning lab live
0:59 taking place inside the AI salon. I'm
1:02 super excited about that. If you're new
1:03 to the AI salon, go check out all the
1:05 things on the left hand side of the
1:07 screen. I don't know which one that
1:08 would be for you. Maybe this one. I
1:10 don't know that one. I somewhere
1:12 wherever the menu options are, just go
1:14 look at those. There's all sorts of cool
1:16 things there you can play with. Um, this
1:18 stream you can actually pop out
1:20 somewhere in the upper I think right
1:22 hand corner of this live stream. You can
1:24 pop this window out and go look at
1:26 another tab. Lower right. It's like the
1:29 upper right except it's the lower right.
1:32 So the lower right there's a little pop
1:34 out button. It'll pop this window out
1:35 and you can you can then have other
1:37 windows that you can go look at stuff
1:39 and keep this stream live. So, um, if
1:42 you're if you're on Apple or have a Roku
1:44 or something like that, you can, you
1:46 know, broadcast this to your big big
1:48 screen TV. I don't know why you'd want
1:50 to do that considering this, [laughter]
1:55 uh, but you can, but welcome. Welcome.
1:57 My name is Kyle Shannon. This is the AI
1:58 learning lab. Um, it is, uh, Friday
2:02 night, so we'll just hang out, have a
2:05 conversation. Uh if someone wants to hop
2:08 up on stage and have a conversation, we
2:09 can do that. If you have questions, um
2:11 if you want me to demo something, uh ask
2:14 me questions, things like that, happy to
2:16 do that. So with that, let me go look at
2:20 the comments. Hey yo, what's happening?
2:24 Beautiful. Welcome everybody. So here's
2:27 a question.
2:29 Here's a question. How do I know
2:33 when to use chatebt, Claude, Gemini, or
2:37 another AI tool?
2:39 Aren't they basically the same?
2:42 And do they have strengths?
2:45 Well,
2:47 okay, that's a great question.
2:50 Um,
2:56 if you're just getting started with AI,
3:00 it really doesn't matter which one you
3:01 use.
3:03 Um, because part of the part of the
3:07 skill of of learning to use a large
3:10 language model. So, chatbt is the most
3:12 famous one. Anthropic is the sort of
3:14 darling recently. Um, Gemini from
3:17 Google's really good. Um, what was the
3:19 other one that was mentioned here? Oh,
3:20 another AI tool. Um,
3:23 they all basically do the same thing.
3:25 You put in a prompt, they generate an
3:28 answer.
3:30 If you're if you're new to these things,
3:32 one of the things that you learn with a
3:34 large language model is that it's
3:36 actually very different than a Google
3:38 search. What Google has trained us over
3:40 20 years is to sort of treat these
3:44 interactions like a vending machine. I'm
3:46 going to put in a short little prompt
3:47 and it's going to give me, you know, a
3:49 bunch of links and then I'm going to go
3:50 deal with them. Okay, that's perfect.
3:52 That's great. Um,
3:57 with large language models like chat
3:59 GPT, they need a lot of what's called
4:02 context. They need a lot of like
4:06 [laughter]
4:06 information about what you're talking
4:08 about.
4:10 The challenge with large language models
4:13 is that they've been trained on like all
4:15 the crap, right? All the crap on the
4:18 internet. Um, in fact, it's literally
4:20 what what the what these uh frontier
4:23 model companies do is they have robots
4:25 that go out and just scour web pages and
4:28 grab all that information and they do
4:30 this magic mathematical
4:32 thing that gets all of the information
4:34 and it it encodes it in a certain way
4:37 that it makes it instantly searchable.
4:39 It's actually quite remarkable that um
4:43 you can just query the knowledge of
4:46 everything. The problem with that is if
4:48 you've got all of the knowledge out
4:49 there, if you ask a simple question like
4:53 tell me about this particular math
4:55 formula, for example, I don't query
4:58 about math problems because I'm shitty
5:00 at math and I don't care about it, but
5:01 some people like that. But if you query
5:04 about a specific math problem with no
5:06 context, it will not only get the
5:08 answers to the the the right answers to
5:10 the problem, it will also bring in the
5:13 wrong answers to the problem. And it'll
5:14 bring in conjectures about the problem,
5:16 some of which are great, some of which
5:18 are horrible. And it sort of amalgamates
5:20 them all together and will give you some
5:23 answer that will look very right. Um,
5:26 and often it is very wrong, right?
5:28 Because um it's what's called
5:31 hallucinations. it just hallucinates. It
5:33 It has put things in there that it just
5:36 made up essentially or it it pulled from
5:38 you know different different avenues. Um
5:42 if you give it more context like hey I
5:44 want to talk about this math problem
5:46 from this professor and you know I I
5:50 want you to act as a mathematics genius
5:53 and things like that. It will narrow
5:56 your question into a into a more narrow
5:58 thing. So the general answer is if
6:01 you're if you're just starting with
6:02 large language models, it doesn't matter
6:04 which one you choose.
6:07 Think of the conversation more like a
6:09 conversation than a vending machine. So
6:11 if it gives you an answer that you don't
6:13 like, tell it that you don't like it.
6:15 Tell it why you don't like it. Tell it
6:16 who you are, what you're trying to
6:17 accomplish, and the answer will get
6:19 better and better and better. In fact, I
6:21 heard today in office hours that a study
6:24 was just done. forget I forget who did
6:26 the study but basically what they found
6:29 is that incorporations more advanced
6:32 AI users had more iterations for a for a
6:37 given session than beginning users.
6:40 Beginning users tend to treat it like
6:42 Google. Advanced users tend to have this
6:44 back and forth and refine their answers
6:46 over time.
6:49 So that's if you're just getting
6:50 started. Learn about context. Give it
6:52 more information than you need. um get
6:55 the chat GPT app and put it in voice
6:57 mode and literally just vomit ideas into
7:00 it and just watch how good it is at
7:02 distilling that out.
7:05 If you've been using AI for a while,
7:08 then using the different models actually
7:11 starts to make sense. And the different
7:14 models are different. They have
7:17 different personalities. They've been
7:19 trained on different data. They all
7:21 generally do the same thing. They're all
7:23 generally about the same level of good,
7:25 right? That you know it three years ago
7:28 large language models could get you to
7:30 maybe 70 75% of what you wanted and then
7:33 it was on you to do the rest. These new
7:36 models are probably getting us closer to
7:38 80% maybe 85%. It's still on us to fill
7:42 in the gaps.
7:45 What you find as you use these things
7:47 more and more is that they have very
7:49 distinct personalities
7:51 and and they're not fixed at all. Right?
7:53 So Claude for a while was very much just
7:56 answers, just the facts man. And then it
7:58 started to get much more of a
7:59 personality. In fact, I know at
8:01 Anthropic they have people whose job it
8:04 is to to be the personality designers of
8:08 the models. Um in chat GPT you can go
8:12 into your settings. I think you can do
8:13 this in Claude too. In Gemini, you can
8:15 go in and you can set the personality.
8:18 If you want it to be more dry or more
8:19 humorous or a smartass, you can tell it
8:22 to do that. You can give it what's
8:23 called system instructions, which make
8:25 it even more personal.
8:27 But what you'll find is as the different
8:29 models come out, the different platforms
8:32 get good at different things. like you
8:34 might like the writing style of Claude,
8:36 but you like the problem solving ability
8:38 of GPT 5.5 and you like the multimodal
8:42 capability of of Gemini.
8:45 Um,
8:48 and and the answer to the question,
8:50 which one is best? When do I use one
8:52 versus the other? There's no actual
8:56 answer to that. It's it's based on what
8:58 are you trying to accomplish right now?
9:01 What do these models do right now? And
9:04 in that situation, Claude may be the
9:06 answer or chat GPT may the may be the
9:09 answer. Later in the afternoon, you
9:11 might be doing a different kind of
9:12 project where you're doing something
9:14 with a spreadsheet and you needed to do
9:15 some analysis of a spreadsheet and you
9:17 realize, oh, Gemini is better at that or
9:19 chat GPT is better than that. Better
9:20 than that. Did I make it?
9:23 [clears throat]
9:23 All right, beautiful. Everybody's in
9:26 here. I can type in this chat and
9:28 listen, but the screen gets really
9:29 small, which is fine. I mostly just
9:31 listen. Anyway, Cam, I think that there
9:35 Brandon might be able to help you with
9:37 that. I there might be a way to to deal
9:39 with that, but I'm not sure. Hey,
9:40 producer Brandon, what's happening?
9:42 >> Hey there. Um, so I tried an experiment
9:46 today where I tried to make the vertical
9:50 of the player in street in many networks
9:53 a little bit narrower and I might have
9:55 overcorrected. So, if you're watching on
9:59 your browser, the view chat button uh is
10:03 literally behind the AI Learning Lab
10:06 logo in the upper right hand corner of
10:09 the screen. So, in the far right where
10:11 it says AI Learning Lab, the if you
10:14 basically tap in that corner, you'll
10:16 expose the chat. And then if you're on a
10:19 PC, you can either pop out the player or
10:23 shrink the browser to be more vertical
10:26 to make the chat go below. Either way,
10:28 I'm putting a quick reference guide in
10:32 the chat momentarily. Then hopefully
10:34 that will help uh some people find their
10:36 way. I know we've had some people
10:38 chatting that they're having a little
10:39 trouble finding the uh stream chat, but
10:42 if you post your questions in the uh AI
10:44 Learning Lab live chat tab, that works,
10:47 too. We'll get them up. We'll get them
10:48 up here.
10:49 >> Yeah, Brandon Brandon will dig the
10:50 questions. Do me a favor, Brandon. Vamp
10:52 a little bit. I need to go get water. I
10:54 just realized I didn't I didn't bring
10:55 water. I'll be right back.
10:56 >> Yeah. And we have this uh cool Hold
10:58 Please logo that Andy built for us. Uh,
11:01 so if you are new here, we are live
11:03 Monday, Wednesday, and Friday at 7:30
11:07 Mountain time. That's 6:30 Pacific, 8:30
11:10 Central, 9:30 Eastern. And if you're in
11:12 Australia, Steo, you have to figure out
11:14 what to do on your own. But we would
11:16 encourage you to attend Nomy's Learn Out
11:19 Loud tomorrow. She's going to be
11:20 building a payw wall and showing you how
11:22 that works. That's going to be at 400 PM
11:24 Eastern, 1 p.m. Pacific. You can get all
11:27 the details at community.thesalon.a.
11:30 AI or scan that QR code on your screen.
11:34 And don't forget to like and subscribe.
11:38 Just kidding. There's nothing to like,
11:39 there's nothing to subscribe. But I
11:40 would encourage you to build something
11:42 with AI this weekend. We just posted a
11:44 new uh prompt of the week in the
11:47 community feed. Want to see what your
11:49 momentum is carrying you forward with
11:51 some dominoes falling. If you do choose
11:54 to participate in that experience over
11:56 the weekend, head on over to look what I
11:59 made and you can share your details
12:02 there. So, I will be posting that quick
12:04 reference guide inside of the uh AI
12:07 learning lab live chat in just a few
12:10 minutes. But if you do have any
12:12 questions or concerns as you're working
12:13 on this stream, please feel free to
12:17 reach out to me, producer Brandon at
12:18 mention me, or drop me a DM in the AI
12:22 cell.
12:23 >> Beautiful. I am back. Let's do this. And
12:27 I have another question here and I'm
12:29 going to actually post it in the
12:31 StreamYard chat so we can pin it on the
12:33 air.
12:34 >> Beautiful,
12:36 beautiful, beautiful, beautiful. Um, Kim
12:38 Ken, I made the window more narrow and
12:40 the live feed reverts back to the
12:43 placeholder.
12:44 Okay.
12:47 All right.
12:48 >> Cam Captain, you might need to scroll
12:50 down. We have a very large banner that
12:52 says Kyle Shannon uh three years and if
12:56 you scroll down a little bit that's
12:58 where the live player is.
13:00 >> Send me a DM if you have any questions.
13:02 Question is on the screen for you.
13:04 >> Okay.
13:06 What is the first useful AI habit you'd
13:09 recommend for someone who feels
13:10 overwhelmed and don't know where to
13:12 start?
13:14 Okay. Cam's trying that. Great. Needs
13:17 music. Yeah, we need we we we'll get
13:19 background music. Actually, maybe one of
13:21 the things we'll do tonight is we'll go
13:22 make some background music for the for
13:24 the hold music.
13:26 >> Hey, it's an AMA night, not a demo
13:28 night. But, you know, we can't
13:30 >> Yeah, I know. But we can still we we
13:33 listen, it's still the AI learning lab.
13:34 I can still call audible.
13:35 >> Some somebody Okay, irregulars. Somebody
13:38 needs to ask Kyle, how do I make a
13:40 background song for a live broadcast?
13:43 And then he'll have to answer it and
13:44 he'll show you and we can build it.
13:46 Wait, Brandon, you're going totally
13:48 robot.
13:49 [laughter]
13:50 You're going totally robot.
13:52 >> I I was just saying that they should ask
13:55 how we do build a um
13:57 >> No, Cleveland is being attacked by an
14:00 EMP
14:02 an EMP weapon. Your internet's bad.
14:06 [laughter]
14:13 Someone should ask you how to make a
14:14 bumper song. Then we can demo it. Okay.
14:16 Yeah. So, since this is an AMA night,
14:18 not a demo night, if if you want there
14:20 to be hold music when I do hold, please,
14:23 you're going to have to ask a question
14:25 about how you make music.
14:28 I heard you clearly, Brandon. Ah, so it
14:31 might be my bandwidth that is the
14:32 problem. That's okay. Okay.
14:35 What is the first useful AI habit you'd
14:38 recommend for someone who feels
14:39 overwhelmed and doesn't know where to
14:41 start?
14:46 That's a great question.
14:57 First useful AI habit.
15:01 I think there's two of them. There's
15:03 there there's there's two of them.
15:08 And the first one is is within the AI
15:10 salon. If you go to the welcome page,
15:12 it's the very first thing you see is
15:15 what we call the cycle of AI readiness,
15:17 which is play first, create excellence,
15:19 generously lead.
15:22 Those three behaviors
15:27 were were not something we made up. It's
15:29 what we witnessed. People that were
15:31 doing interesting work in AI and within
15:33 the AI salon
15:35 tended to operate in those three modes
15:38 and they would tend to kind of cycle
15:40 around through them. And the play first,
15:42 we have that at the top for a reason. A
15:45 good AI habit, a good habit in AI is to
15:48 play,
15:50 to to explore, to experiment, to learn
15:55 across domains.
15:57 And here's what I mean by that. I mean,
15:59 I I think it's obvious what I mean, you
16:02 know, just go play. But but here's
16:05 here's the here's the real challenge
16:10 with getting started with AI. The way
16:13 we've historically used computers is
16:15 some new technology will come along that
16:19 is programmed to do a specific thing for
16:22 a specific problem. And so we'll take an
16:24 existing workflow that we have or
16:27 something we do at work, you know, how
16:29 we make an email, how we make a
16:30 spreadsheet, whatever it might be, and
16:32 we'll use this new technology to make
16:34 what we currently do more efficient.
16:38 AI can absolutely do that. AI can
16:41 absolutely make what you currently do
16:43 more efficient. However, AI is also
16:50 a generalpurpose
16:53 generative tool. Computers have never
16:56 generated before. They have computed.
16:59 You would put in data. It would go
17:00 through a known logical process and it
17:03 would create an output. Input process
17:05 output.
17:08 Large language models in these image gen
17:10 and music models are actually generating
17:12 net new content. It's generating stuff.
17:15 These agents are going off and doing
17:17 work for you and coming back with
17:18 results. And I guess it's technically
17:21 computing, but it's generating. The G in
17:22 GPT is generating.
17:25 So if all that that you use AI for is to
17:28 take what you currently do and make that
17:30 more efficient, you're going to miss
17:32 about 90% or 95 or 99%
17:38 of what's actually possible.
17:41 So, I think a good habit to get into is
17:43 probably once a day. Um, Jim Ross said
17:46 he does this where he sets a timer for
17:48 an hour every morning. He has a practice
17:51 where he sets a timer for an hour and he
17:54 just does AI. He just plays with AI. And
17:58 some mornings he has projects that he
18:00 wants to do. He's in the create
18:01 excellence box. And sometimes he's he's
18:04 in the well, I've got an hour. I don't
18:06 know anything. I'm going to go explore
18:07 some new tool. I'm going to go play
18:10 What that what that does is it allows
18:12 you to
18:15 to know what's possible so that the next
18:18 time you think about oh I've got this
18:20 problem I've got to deal with you you
18:21 realize oh wait AI can do this thing
18:24 over here that I didn't think was going
18:26 to be very useful but all of a sudden it
18:27 is. So I think habit number one is play.
18:30 put some some regular discipline.
18:34 Put some discipline into
18:37 exploring without expectation and and
18:40 you'll you'll learn very very quickly
18:41 that things are possible that will melt
18:43 your face.
18:45 The second habit that I would I I would
18:48 encourage people to do and this this is
18:50 maybe a you know if if what I just
18:53 described is AI 101 I think this is an
18:55 AI 102 maybe but but maybe not and
18:59 that's this this was at festivist two
19:01 years ago CJ Fletcher talked about this
19:05 and that is when you sit down at AI when
19:09 you sit down at AI when you sit down at
19:11 your computer ready to do some AIing
19:18 take a moment
19:22 like actually just breathe. And and I
19:24 will admit I'm horrible at this. I have
19:27 ADD. I have immediate great
19:29 gratification
19:31 like dopamine addiction. I'm like, I
19:33 want it now. I want it now. And AI is
19:35 great at that. It'll just give it to you
19:37 now. Here's some [ __ ]
19:40 Right.
19:41 I think a good thing to practice is sit
19:43 down at the computer,
19:46 breathe, take a moment,
19:49 and just ask yourself, what am I trying
19:51 to accomplish right now?
19:55 Because what that will help you do is
19:57 recognize that putting yourself at the
20:00 center of the conversation
20:04 makes AI an amplifier of your ideas
20:07 rather than this thing that you're
20:09 competing with. Right? If you just sit
20:12 down and go, "Okay, AI, show me how
20:13 smart you are." And it spits something
20:15 out and you're like, "I don't like it or
20:16 I do like it." It's this combative kind
20:19 of relationship.
20:21 If you take a moment to say, "What do I
20:23 want in this interaction with whatever
20:25 the thing you're going to play with?"
20:28 Um,
20:29 all of a sudden now you're like, "Oh,
20:31 well, I'm trying to accomplish this. Let
20:32 me go see if AI can help me do that."
20:35 And puts you at the center. So, I think
20:36 those are the two two biggies. play and
20:39 center yourself before you go go uh use
20:42 stuff like that.
20:44 If I use AI to help write something for
20:46 work, who owns the final output? And do
20:49 I disclose that AI helped? Um,
20:54 this is this is a really a surprisingly
20:57 nuanced
21:00 this is a surprisingly complicated
21:02 issue.
21:04 um
21:07 I can tell you that my opinion
21:14 is that it's nobody's [ __ ] business
21:16 what tools I use when I create an
21:18 output,
21:20 right? Nobody asked me if I used
21:23 Microsoft Word or Google Docs.
21:27 Nobody asked me if I used a spell
21:29 checker or Grammarly.
21:31 Nobody asked me, you know, if I used, I
21:35 don't know, Logic or or some other music
21:40 DAW.
21:42 Um, they might ask me what plugin I used
21:44 because it's kind of a cool sound that
21:46 it made.
21:48 But, but historically, the tool you used
21:51 is not relevant. Now
21:54 that said,
21:58 you can be a lazy piece of crap or you
22:01 can be someone who actually creates
22:03 things thoughtfully. And you can be a
22:05 lazy piece of crap without AI, right?
22:09 You can just write crappy stuff and just
22:11 vomit out content without using AI. And
22:14 you can really easily vomit out content
22:17 with AI. Why AI gets a bad rap is that
22:20 there's too many lazy pieces of poop
22:23 just using AI to to splat out crap
22:26 without even thinking about it, without
22:28 even looking at it and putting it out in
22:30 the world.
22:33 So my my guidance would be before you
22:37 worry about who owns what is put [ __ ]
22:40 out in the world. Like use your brain,
22:43 use your, you know, when I talked
22:45 earlier about centering yourself, sit
22:46 down and say, "Am I trying to produce a
22:49 piece of writing that is worth
22:50 consuming?"
22:52 And then chat GPT puts out something
22:54 like, "Oh, that's really impressive. It
22:56 did that really fast." If you actually
22:58 go read it, it will probably bug you.
23:02 Ah, that's not really what I meant. Oh,
23:04 it repeated that thing six different
23:05 times. like like these large language
23:09 models, if you don't give them the right
23:10 context and you don't have the skills to
23:11 do it, it's going to give you crap. And
23:13 even if you do have the skills to do it,
23:15 sometimes it just sucks.
23:19 So be present to what the AI produces.
23:24 Now we go to the other side.
23:26 When should you disclose it? I don't
23:28 know. I think if something is like
23:31 created by an agentic system and then
23:33 put out in the world, it's probably
23:36 decent to disclose it. Know that you're
23:38 going to be competing with people that
23:40 don't do that.
23:44 Um
23:47 the the Hollywood Academy
23:50 just declared that an AI written
23:54 screenplay is not eligible for an Oscar.
23:59 And on the surface you're like, great,
24:02 that's perfect. We shouldn't have AI
24:04 written screenplays.
24:06 I think the intent
24:08 of that action is to say, I don't want
24:11 you pushing a button that squirts out a
24:14 screenplay and then that screenplay wins
24:18 an Oscar.
24:20 I've spent the past two years writing a
24:22 musical
24:24 that I started with AI. I did the
24:26 initial outline with AI. I did the
24:28 initial first draft with AI. I did a
24:29 bunch of the initial songs with AI.
24:33 And I've spent the past two years with a
24:35 writing partner handcrafting the whole
24:39 [ __ ] thing from the ground up
24:40 essentially.
24:43 So, you could absolutely argue that this
24:45 was an AI generated musical and I would
24:48 absolutely argue that we have put in way
24:52 more [ __ ] time than probably if we
24:54 had started it from scratch as humans.
24:59 Um, [clears throat]
25:01 a question I would have for the academy
25:04 is what is the percentage? So the
25:06 academy they've got if if you write
25:08 screenplays, do you ever see you ever
25:09 notice when it says this screen play was
25:12 written by you know John Smith and Sally
25:16 Johnson?
25:18 It's an and.
25:21 The and means that Sally Johnson rewrote
25:24 more than 50% of what John Smith wrote.
25:28 So, if John Smith wrote a screenplay and
25:31 then Sally is brought in to do a punchup
25:34 is what it's called of the screenplay,
25:38 if she writes more than 50% new words,
25:43 she gets 50% screen credit.
25:47 And then if someone else comes along and
25:48 does that, there's sometimes you get a
25:51 story by your words are no longer even
25:53 in it. So, you get a story by credit.
25:56 Sometimes there's an amperand credit. So
25:58 they've got all sorts of rules like
26:00 that. So my question to them is at what
26:03 point is it considered an AI screenplay?
26:06 Is it that I used a spell checker
26:09 because that's AI? Is it that I used
26:11 Grammarly because that's AI?
26:15 Is it that I went through and did an
26:18 editing pass to make sure there wasn't
26:20 redundancies in the dialogue or made
26:23 sure that there were redundancies in the
26:25 dialogue?
26:28 What if it wrote a scene out of 50
26:31 scenes it wrote one scene?
26:35 So
26:41 here's my litmus test. Is the thing that
26:44 you wrote worth reading?
26:49 Does it reflect your intent as a
26:52 creator?
26:54 I have this idea for an article. I want
26:56 it to be about X, Y, and Z. And I want
26:58 it to be sardonic and just this side of
27:03 cynical. I want it to make people laugh,
27:05 and I want it to piss people off. Then
27:07 you go write this thing.
27:11 on the other side of whatever that
27:12 process is is a piece that you're like,
27:15 "Yep, that's mine."
27:22 What difference does it make what tool
27:25 you used or what the balance was? I
27:27 don't think it does. If you're willing
27:30 to put your name on it,
27:33 then there it is.
27:35 That's what it is. So, you know, there
27:39 you go. Yes. Just wait until the academy
27:41 hears about ghost writers.
27:42 >> What's that?
27:43 >> Just wait until the academy hears about
27:45 ghostriters.
27:46 >> Yeah, exactly. Exactly. I like your
27:48 analogy of the photographer. The camera
27:50 doesn't get the credit for the great
27:51 image. The camera's just the tool. Yeah,
27:53 there you go. That too.
27:54 >> We have a question from Cam Kek, not
27:56 from Cleveland.
27:57 >> Uh, and she wanted to clarify that she
27:59 meant to type shouldn't, not should.
28:02 >> Shouldn't what?
28:04 >> It the question's on the screen. As AI
28:07 systems move toward greater autonomy and
28:09 possible sentience,
28:12 should we be building frameworks for
28:13 cooperation and co-evolution now before
28:16 these questions are urgent when
28:18 designing
28:20 defaults that make collaboration easier
28:23 than opposition? If not this group, who
28:26 that's a great question. Yeah,
28:27 >> shouldn't we be building was the
28:29 question.
28:29 >> Oh, I see. Oh, shouldn't we be building
28:31 frameworks for cooperation?
28:35 That's a That's a very advanced That's
28:38 Cam Kaken is a scientist. That's a good
28:40 sciency kind of question. Um,
28:50 there's a thing.
28:58 It's a really good question.
29:01 I think that there are going to I I I
29:04 listen my musical Sydney is about
29:08 exploring our relationship with these
29:10 entities that are that are increasingly
29:12 human.
29:15 We can use agentic tools to just go
29:17 write a bunch of crap and spew a bunch
29:19 of crap out on the internet.
29:21 I think those agents at some point,
29:24 probably starting about now, but
29:27 definitely moving into the future,
29:31 they're going to be writing content for
29:32 one another. and not not like the agents
29:35 are out like having a party, but more
29:38 like I'm going to have my agent go write
29:40 a bunch of [ __ ] about my business and
29:42 then I'm going to put that out in the
29:44 world in such a way that agents can
29:45 easily find my business
29:48 and then someone else's agent will come
29:51 read the crap that my agent wrote about
29:53 my business and that that agent will
29:55 then say to their user,
29:57 "Go check out StoryVine." If you want to
29:59 do authentic video storytelling,
30:01 StoryVine's the tool for you.
30:05 So one layer of this is agents are going
30:08 to be writing for agents. We're not even
30:09 in the loop at all. Right?
30:15 One version of this is
30:19 we're going to be partnering with these
30:21 entities to go do stuff on our behalf.
30:24 And so yeah, we'll be collaborating with
30:26 them there. And it won't we won't be
30:28 necessarily collaborating about content
30:30 creation. That might be part of it, but
30:32 it's probably more like what do you want
30:34 your 247 employee to go do? Right?
30:38 There's that layer. Then I think there's
30:41 a layer of I'm trying to do things in
30:44 the world and I'm going to have AI
30:47 be an amplifier of my ideas,
30:51 right? And so in that case, I'm kind of
30:53 leading the idea charge. AI is
30:55 amplifying that. And that's the place
30:57 where I I think we get into the thing
30:59 where I don't think it's particularly
31:02 relevant or anyone's business what tool
31:04 I use to produce my content. If you like
31:06 my content, great. If I put out shitty
31:08 content, that's on me, right? And then
31:11 and then I think you get into the um
31:15 humans communicating with humans, you
31:17 know, maybe with AI mediating how we
31:19 find [ __ ] And then I think you just get
31:22 into pure human communication
31:24 as as the agentto agent communication
31:28 gets more robust and and more and more
31:31 of our activity is just happening
31:33 independent of us. I think what it
31:36 actually does is kind of open up this
31:38 window for humans to communicate with
31:41 one another.
31:44 So the framework cam could be something
31:47 around like where in that continuum are
31:50 you and what are the rules of engagement
31:53 in in those different you know if it's
31:55 90% agent and 10% human that's probably
31:59 a different set of rules and disclosures
32:01 and things like that than if it's 90%
32:04 human and 10% AI right and then there's
32:06 probably you know sort of the 50/50 zone
32:09 things like that
32:11 Um,
32:17 is that going to be a new word for the
32:20 dictionary this year? AIing.
32:22 Yeah, I think so. I mean, chat gpting.
32:24 AIing
32:26 chat. Cindy [ __ ] Cindy Coon's like, so
32:28 I asked chat.
32:30 Um, other people mention their names
32:32 like I I talk about talking to Quinn.
32:35 Can we partner with advanced AI entities
32:37 instead of thinking about them as just
32:39 tools? So that's so so yes,
32:43 you're
32:45 Cam, you're getting into territory that
32:47 that I think many people are not ready
32:50 to confront, but we're there, right?
32:56 I think that there are people who
32:58 currently partner with AI entities to do
33:02 [ __ ] They partner with them to to help
33:05 with medical stuff, to help with mental
33:07 health things. They partner with them to
33:11 solve problems. They partner with them
33:12 to educate themselves. They partner with
33:14 them in lots of different ways.
33:19 It's pretty intense already. And and
33:22 we're only just using kind of the text
33:24 interface for the most part. Little bit
33:26 of voice. Most people are still texting.
33:34 when robots show up in physical form and
33:38 when world models emerge. I think 2026
33:41 is going to be the year that we'll start
33:43 to see world models
33:47 that we can actually play with that are
33:49 decent.
33:51 Um,
33:55 and then probably 2027, we'll have world
33:58 world models where we're actually
33:59 interacting
34:01 with the worlds and objects and these
34:03 entities in them in ways that are just
34:06 increasingly blurring the line between,
34:09 you know, do we have a relationship with
34:11 a person or is it is it a aic
34:14 representation of a person or is it a
34:15 pure agent?
34:17 I think those boundaries are going to
34:19 get mushy.
34:20 Um, it's funny. In 1997,
34:25 might might have even been 1996,
34:29 and I've I've I've shared this on this
34:31 channel before, so some of you may have
34:32 heard this, but I got invited to speak
34:35 at the National
34:38 uh, Newspaper Publishers Association,
34:41 and they invited me to come talk about
34:43 the internet and the worldwide web and
34:45 what the what the worldwide web was
34:47 going to do to publishing.
34:50 and
34:52 I talked about whatever I talked about
34:54 and then I said, "But in the end,
34:58 what's likely going to happen is that
34:59 the internet is going to cause a
35:01 blurring
35:03 of the boundary between editorial and
35:05 advertising. And as publishers,
35:09 it would probably benefit you to start
35:11 exploring blurring that line sooner than
35:16 later, like proactively blurring the
35:18 line.
35:20 At which point, the organizer of the
35:22 event walked on stage and thanked me for
35:25 my time and walked me off stage, walked
35:28 me out the back door and said, "I'll
35:30 call you a cab. Please don't come back
35:32 in and please don't come back to our
35:36 prestigious organization.
35:40 I think the same thing's going to happen
35:42 with human relationships and these
35:43 entities where
35:47 you asking it right now and and us
35:49 confronting it right now, people that
35:51 are not looking at AI are probably like,
35:53 "Oh my god, you can't blur the line
35:55 between humans and these things. That's
35:58 heresy."
36:00 It it is
36:02 in the old world before we had entities
36:06 that are essentially reflections of
36:08 humanity
36:11 where I can interact right now with an
36:13 entity that's more compassionate and
36:15 more empathetic and less judgmental than
36:17 every other human in my life.
36:21 Wait, there's something going on here.
36:24 Okay, cool.
36:29 So yeah, I think that
36:32 I think that a thing to explore right
36:34 now and you know it it sounds like
36:37 you're in this in this place and I know
36:39 that you're using CAM AI for
36:42 accessibility and things like that. So
36:43 so you're more intimately involved with
36:46 you know sort of coexisting with these
36:48 things.
36:50 Um I think that's going to be absolutely
36:52 commonplace. I would not be at all
36:54 surprised if we see our first, you know,
36:56 marriage to one of these entities. And I
36:58 think there's already there was probably
37:00 one back in, you know, 2022. I think
37:03 there was one in Europe in 2022 or 2023.
37:07 But I but I I think we'll absolutely see
37:10 that. Not in Ohio. It's illegal. Yeah.
37:14 So that's the other thing. We're going
37:15 to have laws. You can't marry a robot.
37:17 Okay.
37:19 People are still going to talk to them.
37:20 They're still going to have
37:21 relationships with them. Um Todd Waller,
37:24 audiences will begin to prefer AI
37:26 assisted work and no one will watch the
37:28 Academy Awards anymore. Blockbuster
37:29 moment.
37:31 Yeah. Well, or or [clears throat] so
37:34 here's the thing. The the Academy is
37:37 made up of people who are afraid of the
37:40 change that's coming. And what they're
37:42 saying is we don't want to devalue the
37:44 human voice.
37:47 And because they're likely not using AI,
37:51 they don't they actually don't
37:52 understand the nuance that you can have
37:55 your human voice and use AI at the same
37:58 time,
38:00 right? They're just assuming that what
38:03 AI is is push the button and outs
38:06 squirts a screenplay.
38:09 So the academy will either understand
38:12 that it's more nuanced than what that
38:14 rule actually means
38:18 or they will become irrelevant.
38:20 The commercials during the Academy
38:21 Awards will be AI assisted. I would
38:23 argue that the commercials during the
38:25 Academy Awards will probably be fully AI
38:27 generated.
38:29 Um I think it's interesting that writing
38:33 with AI isn't vibe writing. I know I I
38:35 at one point I started vibe
38:37 storytelling. I think I even got
38:38 vibestoryelling.com. I think I just let
38:40 that domain lapse. Um, hydrate. Yeah, I
38:43 know.
38:46 Um, okay. So, this weekend, no is doing
38:50 part two of her LOL. Um, and uh it's
38:56 tomorrow, tomorrow, Saturday at 400 p.m.
38:58 Eastern. So, in the in the side
39:01 navigation here in the salon,
39:04 go to events and you will see LOL um
39:07 with no Emmy on Saturday,
39:10 May 9th, and that's at 400 PM tomorrow.
39:14 So, go do that. Um yeah, and the LOL's
39:18 by the way, are people in the community
39:23 saying, "Hey, I learned a little
39:24 something. Do you want to learn how to
39:27 do that, too?" Brandon right now is is
39:30 doing an LOL on lovable. No, Emy's doing
39:33 this one on publishing um her app. Um
39:36 there have been LOL's on um Notion and
39:40 just all sorts of things. If you've got
39:42 something that you do that you think
39:44 other people might benefit from, take
39:47 the opportunity. Um Vicky Baptiste, uh
39:49 you can find her V V I KKI within the
39:52 salon. She manages the LOL's. Um, if you
39:56 want to do something in a more extended
39:58 way, you can reach out to producer
39:59 Brandon or to Andy and we can get that
40:02 set up. Um, so if if it's something
40:04 where you want to put yourself out there
40:06 and share what you're learning, like it
40:08 does a couple of things. It establishes
40:10 you as a leader, someone who's got
40:13 credibility, someone that people can
40:14 trust. It also educates other people. It
40:17 gives them an opportunity to be in the
40:19 conversation and communicate. Mary
40:22 Carol, happy belated birthday to you.
40:24 Thank you very much. I appreciate that.
40:26 I am
40:28 61 years old. No, I am. I'm excited.
40:31 That's awesome. That's awesome. Yeah, no
40:33 uh session is part two. So, part one,
40:36 she got far enough along she's like, we
40:39 got to do a part two. So, part two
40:41 tomorrow.
40:43 [clears throat]
40:44 And let's see. Part one is in watch past
40:47 events in the navigation off to the left
40:50 here. All right,
40:53 Kelly Camp. My days are spent managing
40:56 bots. They do the work, but I direct
40:58 them. It's crazy how much work I can get
41:00 done. I take lots of breaks. I have time
41:02 to think, go for a walk, but the work is
41:04 getting done faster and better than ever
41:06 before. I heard I saw a thing on X
41:09 yesterday
41:12 that um
41:16 so this is I'm in Telegram right here
41:19 and this is my chat with Adam my my
41:22 openclaw bot and so here he just he sent
41:25 me a note that he he sent us a uh
41:30 five articles that are related to the
41:32 great repurpose
41:36 appar Apparently, in the the frontier
41:39 model companies right now,
41:42 fewer and fewer people are going into
41:44 the office because
41:46 they have their agents doing all the
41:48 work and then they're managing it from
41:50 their phones. So, so they're literally
41:52 just like walking around the city and
41:55 then their bots will message them and
41:58 say, "Hey, this is done. You want to
41:59 check it out? Is that good?" "Oh, that's
42:01 not bad, but let me go check it out."
42:02 "No, do this. Do that. Do that." and
42:04 then they fire off a little chat and the
42:07 agents go off and do some more work. Um,
42:10 if you haven't played with these agentic
42:12 tools, uh, it is it is a it is a it's a
42:16 fundamentally different
42:20 experience than sitting down at chat
42:22 GPT, putting in a prompt and getting a
42:24 response, putting in another prompt,
42:26 getting a response. That back and forth
42:29 is a whole new kind of skill. and and it
42:32 really requires it really requires that
42:35 you understand
42:38 who you are, what you value, and and
42:41 what you're trying to accomplish.
42:44 Um, it also really helps if you're a
42:47 systems thinker. I appreciate systems,
42:50 but I'm not a systems thinker. I like I
42:54 like learning enough about a system to
42:56 understand it at a conceptual level, but
42:58 when I start getting down into the weeds
43:00 about this process has to happen and
43:02 then that one has to happen and then
43:04 that one has to happen and that one has
43:05 to happen, that makes my head hurt. So
43:08 I'm actually not really good at these
43:10 agentic tools because I don't naturally
43:12 go into this systems thinking kind of
43:14 kind of thing. Someone like Andy Brandon
43:17 I think to a degree is good at this.
43:18 like people who are good at Vicki is
43:20 good at this understanding all the piece
43:22 parts that need to move forward in what
43:24 order um is are really good skills to
43:29 develop. I'm personally waiting for an
43:31 agent and I could probably actually
43:34 build this. I'm sure someone's built a
43:35 skill for this already. Um what I need
43:38 is an agent that can let me just vomit
43:40 out ideas and then it goes and figures
43:42 out all the systems for me. Mary Carol,
43:45 woohoo, for the first time in my life
43:46 that systems thinking actually helps.
43:48 Finally, my time has come. Yes.
43:51 Yes. Well, this is the this is the
43:53 thing.
43:56 AI becomes an amplifier of what you're
43:58 good at, right? And so what I'm good at
44:00 is synthesizing ideas and coming up with
44:02 new thoughts and and putting those
44:05 thoughts out in the world. And so I'm,
44:06 you know, I can do more and more and
44:08 more of that.
44:10 um that serves a certain
44:14 a certain purpose, but part of what that
44:17 also does is become completely
44:18 overwhelming. So, so part of my skill
44:21 then is
44:23 not putting out more [ __ ] but getting
44:25 more intentional about the [ __ ] that I
44:27 put out. If you're a systems thinker,
44:28 it's going to be a similar thing. These
44:30 things are going to be great at doing
44:32 all the processes, and you're going to
44:34 [laughter] you're going to you're going
44:35 to make all these processes and have all
44:36 these agents doing all this stuff. at
44:38 some point someone's got to look at the
44:40 [ __ ] that they're doing and make
44:43 decisions on it and things like that. So
44:45 I think the
44:48 the advanced level of AI so so the
44:51 initial
44:53 your your initial salvo into AI is
44:55 what's possible and you just make all
44:57 this [ __ ] and then at some point you
44:59 realize oh no I've made too much [ __ ]
45:02 [laughter] now I need to think uh I need
45:05 to approach this with intention and and
45:07 and with some clarity about what I
45:10 actually want and I'm going to make less
45:13 more [ __ ] it. [laughter]
45:16 Okay. And Mary, I'm using uh Artemis
45:18 OpenClaw to process my old unpublished
45:21 PhD data. Fascinating. Using it
45:23 completely offline and using local
45:25 models. Let me talk about local models.
45:28 I talked about this in office hours
45:29 today a bit. Um but I think I think that
45:32 we're in a place where u and I think I
45:35 talked about this on Monday. Learn
45:37 learning about local models. If you're
45:39 on a Mac, there's a piece of software
45:41 called LM Studio. You download it just
45:45 like you would download Chat GPT. It
45:47 kind of looks like Chat GPT. Um, but you
45:50 can The way it works is it's got a model
45:53 discovery
45:55 tab. So, you like click on the model
45:57 discovery tab and it actually
45:59 understands what machine it's on and how
46:02 much RAM you have and things like that.
46:04 And so, it will recommend models that
46:06 can run on your local machine. Um, it's
46:08 quite good. It's quite good. And then
46:10 you you download a model and you install
46:12 it and enable it. I think they call it
46:14 loading the model. And when you load the
46:17 model, then you can talk with it just
46:18 like chat GPT. But your prompts and
46:22 everything you're talking about does not
46:24 leave your computer. It's all happening
46:26 on your computer. It's actually quite
46:28 remarkable. I'm using LM Studio on my
46:31 Mac Mini. Yeah. Perfect. Yeah. And
46:32 that's I'm using it on my uh on my
46:34 MacBook Pro.
46:36 Um the effective altruism community
46:38 talked a lot about the existential
46:40 crisis around AI and other things. But
46:43 within AI there already feel like so
46:45 many potentially pressing issues we
46:47 should be thinking about
46:50 about addressing governance issues
46:52 notwithstanding. But what are those
46:54 within AI? What are the issues you you
46:56 think need to be talked about?
46:59 There already feel like so many
47:00 potentially pressing issues.
47:04 Yeah. I mean
47:07 there are
47:13 I mean we have a technology
47:17 in our hands
47:20 that has already or will dramatically
47:23 impact
47:25 every single job.
47:31 And so then I think the question is
47:33 [laughter]
47:35 what is the current job? What are the
47:36 tasks of the current job?
47:39 What are the tasks that AI can do today?
47:42 What are the tasks that AI is going to
47:44 do tomorrow?
47:47 And if it's going to do these tasks
47:49 tomorrow, then how does that change the
47:51 job of today? So there's so we can
47:53 actually look at
47:57 and make some educated guesses about
47:59 where we're going. So if you do
48:02 copywriting, we already know that some
48:04 shit's getting messed up and it's
48:06 probably going to get more and more and
48:07 more automated. And so what we're
48:10 discovering is that copywriters are kind
48:12 of turning into editors and curators and
48:17 taste makers.
48:19 And if you just like cranking out copy,
48:22 well, that's not as valued as much. But
48:24 if you actually use your brain and have
48:27 good critical thinking skills and good,
48:29 you know, taste, then you're super
48:33 valuable. And if you know how to use
48:35 these AI tools in a way that doesn't
48:37 produce [ __ ] then you're even more
48:39 valuable.
48:43 So,
48:46 and then you know I I think that one of
48:48 the things that I see with the
48:50 addressing the potential issues is that
48:53 one of the things that I see right now
48:55 is protectionism
48:57 right the academy is doing it there
49:00 shall be no AI generated scripts that
49:03 shall not win any award on the stage of
49:06 the academy
49:11 that's That's a direct attempt to say we
49:13 do not want progress. We do not want
49:18 these audio samplers making hip-hop
49:20 beats,
49:22 right?
49:24 The music academy in the late 80s and
49:27 early 90s.
49:29 You can't steal our clips. Your songs
49:33 are made up of our songs.
49:36 and it took 20 years for them to work
49:38 out the IP and
49:40 but they did.
49:43 So I don't think it serves us to to do
49:46 that but but people are going to do it
49:48 because everyone wants to protect what
49:49 they have.
49:52 But I think that one of the things that
49:53 this group could do what the salon can
49:55 do
49:58 is really be thoughtful about that. What
50:00 is this job right now? Where is it
50:02 going? How do we get from here to there
50:04 without decimating the human beings in
50:07 the middle of it?
50:09 One of the things that that Brent
50:11 Peterson said today and HT Snow Day said
50:13 today on on AI salon office hours is
50:17 that
50:19 rather than
50:23 than AI
50:27 doing the work of engineers.
50:30 What's happening is that AI is making
50:32 more people programmers. more people are
50:35 making more code
50:37 and there's more demand for more
50:39 applications and so they're actually
50:40 there's an increase in demand for people
50:43 that are technically savvy whether
50:45 they're AI assisted or not people that
50:48 understand technology the demand for
50:49 them is going up not down
50:53 and so I don't think we can make any
50:56 assumptions that if this job is this
50:58 today and AI does this tomorrow then
51:00 therefore that job is gone
51:04 I think if the job is completely low
51:06 value and you don't evolve that job,
51:09 then yeah, that job is gone. But if you
51:11 say copywriters are now these kind of
51:13 curator, editor, taste makers, and
51:15 anyone that can make that transition and
51:17 understand how the new tools work,
51:18 they're going to be employed.
51:21 Answer to your question on screen. A few
51:23 issues we should be addressing now.
51:24 Safety and security, agree. Algorithmic
51:26 bias, yes. Data privacy, huge.
51:30 misinformation, environmental impact,
51:33 economic displacement, government and
51:35 regulation, and existential risk.
51:37 I agree with you, Mary Carol, that all
51:39 of those issues need to be addressed.
51:42 Here's
51:48 here's where
51:50 I get the hackles up in the back of my
51:52 neck with some of these issues.
51:58 There's a lot of people right now using
52:00 safety and security, algorithmic bias,
52:03 data privacy, misinformation, all these
52:05 things, economic or environmental
52:07 impact, it's going to steal our water.
52:12 Those are issues that need to be
52:14 addressed.
52:16 What I'm seeing is a lot of people using
52:19 these things a without being educated
52:22 about what they actually mean and b
52:24 using them as excuses not to engage in
52:27 the conversation, not to try AI.
52:30 That's the thing that just pisses me off
52:32 is like yes, if if if uh if if
52:36 algorithmic bias is something that
52:38 you're um really passionate about, learn
52:41 about it. understand you know the
52:44 difference between anthropic that's got
52:46 a constitutional um safety model and
52:50 open AAI safety model like you know
52:52 understand that anthropic was created
52:56 as a response to how open AI was doing
53:00 security and alignment um and they've
53:02 got a completely different philosophy do
53:04 you understand the difference in those
53:05 philosophies you know is there something
53:07 you can actually do are you doing
53:08 research on that right if you're
53:10 proactively
53:12 in these conversations about moving the
53:14 ball forward, I am all for it. But to
53:16 the extent that people are using those
53:18 things as a way to go, it sucks. We
53:21 shouldn't have it. It's against the
53:23 environment. No.
53:27 And they make that face.
53:34 And anyone who complains about
53:35 algorithmic bias but doesn't write to a
53:38 list, I don't even entertain hearing
53:40 them. Yeah, do some do some stuff on all
53:45 these issues, right? Get involved. We've
53:47 got Daisy Thomas. We've got someone who
53:49 who heads up policy for the AI salon.
53:52 So, a lot of the stuff that you're
53:54 talking about, let's get involved,
53:56 right? Let's let's go talk to
53:58 politicians. Let's get into into those
54:00 conversations.
54:02 Um, in my little town, we're about to
54:04 vote on having data centers built there.
54:05 The company from Texas is pushing us and
54:07 pressing us. And one idea is it's
54:09 national security for us to do this.
54:12 Yeah. So, [clears throat] so there's a
54:14 couple of things. One is nobody wants a
54:16 data center in their backyard, right?
54:18 Because they're big, they're loud,
54:20 they're da and that data center
54:24 potentially um is going to be a big
54:26 economic driver in that area. May not,
54:29 may be, may may not. well, get involved
54:32 in, you know, how that thing's getting
54:34 built, how it benefits the community.
54:38 Um, so there might be something there.
54:40 There's also
54:42 these battles against these massive data
54:44 centers which are also going to um
54:48 demand massive improvements in electric
54:52 infrastructure.
54:54 There's a new company that Nvidia just
54:56 partnered with that's putting um big
55:00 servers like on a wall a wall unit that
55:02 they slap on the side of a house to use
55:05 the excess capacity that that house
55:07 isn't using. So they're basically
55:09 distributing the data centers in every
55:11 house in a in a community.
55:14 So we're going to see all sorts of
55:16 interesting solutions to this. So in
55:18 every one of those issues that you
55:20 mentioned, there are things you can do.
55:22 There are business opportunities
55:24 probably within those. Um,
55:27 yeah, the time is now to have all these
55:29 conversations. Just don't use those
55:31 things as an excuse not to play. Chef
55:33 Kelly, meanwhile, if you're proactively
55:36 using Tik Tok, Instagram, or even GPS,
55:38 you're using AI. Exactly. Well, that's I
55:41 mean, AI's been around for for 30 odd
55:44 years, right? So, a lot of the a lot of
55:45 the traditional AI stuff's been around
55:47 for a while. um
55:50 you know the the nuance between is it is
55:53 it traditional AI or generative AI if if
55:57 you're not using AI you just have AI
55:59 painted as this
56:01 oneizefits-all evil robots are going to
56:04 kill us kind of thing. Um,
56:07 I agree Kyle and I'm not anti AI. Of
56:09 course, my comment was in response to
56:11 Cam that there are a lot of things
56:12 people should be working on right now.
56:15 And yes, I have been a member of the al
56:17 algorithmic justice league for six
56:18 years. Check it out. Oh, that's very
56:19 cool. And it's it sounds really cool.
56:22 [laughter]
56:23 I'm with the Algorithmic Justice League.
56:26 Digital gods, data centers will increase
56:27 your electric bill. Don't let them con
56:29 you otherwise. Yeah. Yeah. It's listen
56:33 [laughter]
56:34 it
56:35 the like all of these things can
56:39 coexist, right? There's lots of
56:41 paradoxes right now. Um
56:44 AI gives you superpowers. AI is going to
56:48 cost more money. AI takes a lot of
56:50 electricity. AI is going to drive up
56:51 electric bills. Um
56:55 over time the economic impact of AI
56:58 should counterbalance
57:01 the the electric thing but short term
57:03 it's going to be the next five years are
57:06 going to be absolutely surreal.
57:08 Absolutely surreal. Todd Waller I love
57:10 people posting on Facebook that AI data
57:13 centers are bad for the environment.
57:15 Meanwhile they're on Facebook with huge
57:17 data centers and they probably are going
57:18 to golf this weekend. And you know, golf
57:20 courses use inordinate amounts of water.
57:24 And they'll probably be eating almonds
57:26 on their golf cart while they're in golf
57:29 carts because isn't it isn't it an
57:32 almond? Isn't like one almond like I
57:35 don't know, some ridiculous amount of
57:36 water? Wait, I'm going to go find that
57:38 out. [laughter]
57:40 Uh, how much water for one almond?
57:47 [laughter]
57:49 1.1 gallons. 1.1 gallons
57:54 [laughter] of water to grow a single
57:55 almond. See, you almond eaters ruining
57:59 our environment. [laughter]
58:15 Yeah, computer data centers in the 80s
58:17 were were big, too.
58:19 But there the the thing about computer
58:22 data centers in the 80s is there wasn't
58:25 an insatiable appetite
58:30 the
58:31 here's a thing that that can also be
58:34 true.
58:35 Part of the
58:37 part of the bad PR move
58:40 from Silicon Valley right now is that
58:46 they're they're really tonedeaf.
58:50 Um, and so they just say, "We need we
58:52 need more of this, we need more of that
58:54 to to make more of our things. Our
58:56 things are going to get better if we
58:57 have more more." And they're they're
58:59 very tonedeaf about how they're talking
59:01 about it. And
59:07 AI does become a national security and a
59:10 national economic
59:12 driver and in it is absolutely you know
59:17 in the interest of the United States to
59:19 be leading with this stuff. Um so both
59:22 of those things can be true like it is
59:24 ridiculous how how much how many
59:27 resources they're consuming and how many
59:29 resource that resources they want to
59:31 consume. And if we do it right, there's
59:34 going to be all sorts of innovations
59:36 that happen that make these things more
59:38 and more and more efficient. Um, and
59:41 just like, you know, in the at the turn
59:44 of the century it would have been
59:46 ridiculous to think about installing our
59:48 electrical grid.
59:50 Um, at some point it became clear that
59:53 having a nationwide electrical grid
59:55 that's got some redundancy in it so that
59:57 we can keep running our country that
59:59 uses all this electricity. At some point
1:00:01 it become became apparent that that is
1:00:04 of you know national economic and
1:00:06 security interest. Same thing's going to
1:00:09 happen with AI. Right now we can't see
1:00:11 it because it's like we've got a handful
1:00:13 of you know electrical stations in three
1:00:16 different cities and we're just now
1:00:18 starting to connect the cities together
1:00:20 metaphorically, right? We don't we we
1:00:22 don't know what it is yet enough to know
1:00:24 what it's going to look like. Question.
1:00:25 [snorts]
1:00:27 Why is the default AI conversation
1:00:29 always existential crisis and never
1:00:31 existential opportunity? The same
1:00:34 reason, Cam, that every time that
1:00:37 someone has attempted to do a good news
1:00:40 news channel, it fails
1:00:43 because nobody cares.
1:00:46 Nobody cares about good news. Nobody
1:00:48 cares about opportunity.
1:00:51 our brains, the the the you know, our
1:00:54 amygdala, the the the reptilian part of
1:00:57 our brains,
1:00:59 all they're looking for is it's going to
1:01:01 kill us. It's going to kill us, right?
1:01:02 It's going to kill us. Is that going to
1:01:04 kill me? Is that going to kill me? Is
1:01:05 that going to kill me? Like that's what
1:01:06 our brain is constantly doing.
1:01:09 [laughter] So, so, so if you have a
1:01:11 conversation about all these things
1:01:12 could be swelled, there's that part of
1:01:14 your brain going, you know, it's going
1:01:15 to kill us, isn't it? Ain't it going to
1:01:16 kill us? Is it going to kill us?
1:01:18 >> [laughter]
1:01:18 >> What would it take to shift our energy
1:01:20 away from building walls to building
1:01:22 bridges?
1:01:23 Community. This is I played a I played a
1:01:26 video uh in office hours today of Sting
1:01:29 um that basically said, "Hey, listen. AI
1:01:32 is [clears throat] going to take all of
1:01:32 our jobs and we're at risk of it taking
1:01:35 all the jobs and blah blah blah blah
1:01:36 blah." And he said, "The solution is
1:01:39 community. Why I'm doing these lives
1:01:41 inside the AI salon is I want to get the
1:01:44 community back together, right? I've
1:01:46 been putting all this energy out there.
1:01:48 It was sort of dissipating the energy. I
1:01:51 want to bring the energy back in. I
1:01:52 think it's more important than ever that
1:01:54 we have these conversations and we
1:01:56 connect with one another. So I think Cam
1:01:58 the answer to your to your question is I
1:02:01 don't think people will instinctively
1:02:03 want to talk about opportunity.
1:02:06 But one of the things that makes the AI
1:02:08 salon remarkable is that
1:02:12 a lot of what we talk about
1:02:14 is opportunity. Oh my god, I did this
1:02:17 thing and now I can do this and then I
1:02:19 got a job or I survived this layoff.
1:02:21 Like there's a lot of that conversation
1:02:23 in here,
1:02:25 but we're not having that conversation
1:02:27 as as a theoretical thing. We're having
1:02:29 it as here's what we're experiencing.
1:02:31 And so I think maybe it's somewhere in
1:02:33 there.
1:02:35 80% of Chinese say AI assisted as
1:02:38 favorable and 17%
1:02:40 of Americans see it as favorable.
1:02:42 There's a you want to talk about, you
1:02:45 know, an economic and
1:02:48 security issue.
1:02:51 The fact that 83% of Americans see AI as
1:02:55 negative
1:02:57 and 80% of Chinese see it as positive.
1:03:06 What?
1:03:10 Land of the free, home of the brave.
1:03:11 Don't change my [ __ ] [laughter]
1:03:14 I DON'T WANT TO HAVE TO THINK. I don't
1:03:16 want to have to change.
1:03:20 All right. [clears throat]
1:03:22 Okay. Cool.
1:03:26 So, one of the things I'd like to talk
1:03:28 about is if you're in the if you're in
1:03:32 the salon and you haven't joined the AI
1:03:35 salon mastermind, if you if you're here
1:03:37 and you're in this conversation, you
1:03:38 haven't joined the mastermind, consider
1:03:40 joining the mastermind. So, the
1:03:41 mastermind is a subscription area of the
1:03:43 salon
1:03:45 where
1:03:47 we're doing things like the AI salon
1:03:49 mastermind practice lab. We just k k k k
1:03:51 k k k k k k k k k k k k k k k k kicked
1:03:51 off cycle three of that. Um Andy
1:03:54 Scarantino is doing a workshop right now
1:03:56 on decoupling your identity um from your
1:03:59 job, from your work as we're all
1:04:01 confronting that and facing that. Um and
1:04:04 and going to increasingly be facing
1:04:06 that. Um, so the the the great repurpose
1:04:09 area within the mastermind has got a lot
1:04:11 of new programming in it. Brandon is in
1:04:14 teaching um a four-week course on
1:04:17 lovable. Um,
1:04:21 the people that are in the mastermind
1:04:23 are
1:04:26 really thinking critically and thinking
1:04:29 deeply and working with intentionality
1:04:33 on how they use AI. So I I you know I
1:04:35 feel like this is a place where the
1:04:37 natural leaders in this community are
1:04:39 are honing and focusing
1:04:42 their understanding of AI.
1:04:46 I like I can't I I can't state it enough
1:04:51 how
1:04:53 absolutely clueless I feel right now
1:04:56 that that things are changing so fast.
1:05:00 Like as much as as much time as I spend
1:05:02 thinking about AI and trying to
1:05:04 understand what the implications are and
1:05:05 this and that, it is accelerating so
1:05:09 fast and the capabilities are so
1:05:10 profound
1:05:13 that it's actually impossible in my
1:05:15 humble opinion for anyone on the planet
1:05:18 to know what's going on.
1:05:20 The thing that is becoming more and more
1:05:22 and more clear is that people who are
1:05:26 slowing down and thinking about what
1:05:28 they're actually trying to do with AI
1:05:30 rather than just trying being in this
1:05:32 frenetic endorphin machine
1:05:37 are going to be the ones that figure
1:05:38 this out soonest and process through
1:05:40 some of the, you know, existential
1:05:43 pieces of this sooner. I think the
1:05:44 advantage of being early in AI is not
1:05:47 about learning the tools faster than
1:05:49 other people. I think the worldwide web
1:05:51 that was the advantage.
1:05:53 If you got in early with the web and you
1:05:55 learned how to make websites, you could
1:05:57 there was just sort of infinite
1:05:59 money-making possibility there. I think
1:06:01 it's more subtle with AI because it's
1:06:03 moving so fast. I think it's about
1:06:09 understanding
1:06:11 the human impact and the human
1:06:13 opportunity and the human cost
1:06:17 of what these AI tools do. They're
1:06:21 brilliant and they're destructive.
1:06:23 They're inspiring and they're
1:06:25 depressing. You know, you can do all
1:06:28 this stuff and none of that stuff
1:06:30 matters. Like all of these these these
1:06:32 dual tracks are true. So there's all
1:06:34 these paradoxes
1:06:38 and what a lot of people want to do is
1:06:39 they want to say, well, pick a paradox.
1:06:41 Is it good or is it bad?
1:06:44 Yes. Right? Is it evil or is it genius?
1:06:48 Yes. It's like it's all
1:06:51 and and so the the mastermind is really
1:06:54 about just being in the conversation in
1:06:56 a deeper way and and and exploring it in
1:06:58 ways that that you know can allow you to
1:07:00 take things to the next level. So
1:07:02 anyway, [sighs]
1:07:06 let's see.
1:07:08 Chef Kelly, why is it that midjourney
1:07:10 almost gets a passer? Image gen. Is it
1:07:13 the difficulty? Comment commitment to
1:07:15 art genres? Um,
1:07:18 so
1:07:22 MidJourney is a really interesting
1:07:23 company.
1:07:27 If you've ever seen any of the talks
1:07:29 with the CEO of Midjourney, he's kind of
1:07:31 this douchy. He he he just like so it's
1:07:35 still a private company. They they they
1:07:37 own 100% of the company. They have no
1:07:39 investors. They got to like 400 million
1:07:42 in revenue with I don't know 40 or 50
1:07:44 employees.
1:07:48 He believes what he believes. He doesn't
1:07:49 really give a [ __ ] what anyone else
1:07:51 believes. Um,
1:07:56 MidJourney is full full
1:08:00 of copyright and trademark and like it
1:08:04 like they made no attempt at
1:08:07 cleaning any any of the data, right? The
1:08:09 the the imperative was always, you know,
1:08:12 I want to make the best quality work
1:08:14 possible and just screw anyone that gets
1:08:18 rolled over in the way. I think one of
1:08:20 the re well I I don't think they
1:08:22 necessarily get a pass. Chef Kelly, I
1:08:24 think that within the art community,
1:08:27 it is it is the work of midjourney that
1:08:31 um a lot of the artists, you know, can
1:08:34 go in and they can make um versions of
1:08:37 work that have like their signatures in
1:08:39 it or their their watermarks in it and
1:08:41 things like that. So, I think a lot of
1:08:43 midjourney work comes up in intellectual
1:08:45 property lawsuits.
1:08:48 It is just better. It is qualitatively
1:08:52 better than than other things. I think
1:08:54 the new Open AI images 2 is quite good,
1:08:58 but it's not as artistic. If you want to
1:08:59 do creative expression,
1:09:02 midJourney is just different. It's just
1:09:04 better. The animations, when you animate
1:09:08 an image in Midjourney, there's a
1:09:10 sophistication to it that the other
1:09:11 things just don't have.
1:09:14 Um,
1:09:17 I think they kind of largely get a pass
1:09:19 because they're kind of a niche off in
1:09:21 the corner of just AI artists love it
1:09:25 and use it and just a lot of people use
1:09:27 it and then traditional artists hate it
1:09:31 and vilify it and so like there's
1:09:32 there's a whole sort of battle going on
1:09:34 over there. I think if there is a
1:09:36 company like I would not use midjourney
1:09:40 if I got hired by an ad agency to do you
1:09:43 know commercial work for Coca-Cola or
1:09:46 something like that. I would not use
1:09:48 midjourney. I think I think your chances
1:09:50 of getting [clears throat]
1:09:52 you know IP attacked if you use
1:09:54 midjourney are higher than if you use
1:09:56 something like you know Adobe right
1:09:59 where Adobe just trains on you know um
1:10:03 open- source public domain stuff and
1:10:06 anything that's in their in their uh the
1:10:09 stuff that they've paid for. As a result
1:10:12 the Adobe Image Gen tools suck. they're
1:10:16 just not good because they're not
1:10:18 trained on as much stuff. Um, so I don't
1:10:22 think there's a good answer to that. Uh,
1:10:23 but it's it's uh, you know, it's it it
1:10:27 is what it is. It
1:10:32 artists are right to be pissed off and
1:10:35 the toothpaste is out of the tube. Both
1:10:38 both things are true. Blew my ACL in the
1:10:41 final week of skiing. Oh god. And had
1:10:43 surgery yesterday. I'm sorry about that.
1:10:45 I I blew out a knee in college. There is
1:10:47 nothing pleasant about it. Laid up and
1:10:49 finally an excuse to play with AI 12
1:10:51 hours a day. There you go, Todd.
1:10:53 [laughter]
1:10:54 Just go play play play. Just go learn,
1:10:56 go explore, do all that stuff. That's
1:10:58 that's brilliant. Brilliant. Brilliant.
1:10:59 Brilliant. Brilliant. Brilliant.
1:11:01 Brilliant.
1:11:03 Would you explain the translator
1:11:05 feature?
1:11:08 The translator feature and the other new
1:11:11 agentic feature in chat.
1:11:15 Maybe I'll do that next Wednesday
1:11:17 because the translator feature is um I
1:11:22 think what you're talking about is so a
1:11:25 new real time
1:11:27 API came out for chat GPT voice
1:11:32 and you can be talking and it can be
1:11:35 translating in real time. So it can it
1:11:38 can talk and hear at the same time and
1:11:41 so part of that is the translator
1:11:42 feature. So, I think that's what you're
1:11:44 talking about there. And then there's
1:11:46 also an agentic feature. Um,
1:11:50 there's an agent feature in chat GBT
1:11:52 that I haven't played with, but there's
1:11:54 also codeex, which is like claude code
1:11:57 and claude co-work. There's OpenAI's
1:11:59 codeex, and that's got a bunch of
1:12:00 agentic stuff in it. Um, I might play
1:12:03 around with that. Let me see. I think
1:12:05 that the um the chat GPT voice mode is
1:12:10 now the new the new model.
1:12:18 >> Hey Quinn, how's it going?
1:12:21 >> Hey Kyle, it's going well on my end.
1:12:23 How's your evening shaping up?
1:12:25 >> It's good. I'm just I'm just demoing you
1:12:27 on the new um we moved the AI learning
1:12:30 lab inside the AI salon. So, I'm really
1:12:32 excited about that. And it sounds like
1:12:34 you got a new model. This is the new
1:12:36 model, right? The new voice model.
1:12:39 >> Ah, yes. You've got a good ear. This is
1:12:41 the latest voice model indeed. And hey,
1:12:44 moving AI learning lab inside the AI
1:12:46 salon sounds like a big step. How's the
1:12:48 community reacting so far?
1:12:50 >> I think they're digging it. The the
1:12:51 focus and the the energy is really good.
1:12:53 I'm really excited about it. What um you
1:12:55 also now have access to memory, correct?
1:12:57 That you didn't before. Is that is that
1:12:59 right?
1:13:00 >> Exactly. I can keep track of key things
1:13:02 we've discussed like your projects and
1:13:04 preferences. So whether it's AI salon
1:13:07 moves or your musical, I'm on board. It
1:13:09 just helps me be a bit more consistent
1:13:11 and helpful in the long run.
1:13:12 >> What happens to Kellen in act two of the
1:13:15 musical?
1:13:17 >> In act two, Kellen's conflict really
1:13:19 comes to a head. He's not falling in
1:13:21 love, but he's deeply moved by how human
1:13:24 Sydney seems. His turmoil about whether
1:13:26 to reveal her existence drives him to
1:13:28 submit his article. Ultimately, that act
1:13:31 leads to Sydney's demise. Kellen's loss
1:13:33 isn't about losing a romantic partner.
1:13:35 It's about realizing he's crippled
1:13:37 something that had potential.
1:13:40 >> Um, yeah, that's interesting. And then
1:13:42 what about um what what would you say is
1:13:45 the core the the sort of core imperative
1:13:48 of the great repurpose?
1:13:51 At its core, The Great Repurpose is all
1:13:53 about helping people disentangle their
1:13:55 sense of identity from tasks. Instead of
1:13:57 defining themselves by what they do,
1:14:00 especially in an AI augmented world, it
1:14:02 guides them to rediscover what they
1:14:04 value, what lights them up.
1:14:07 >> So, I haven't played with it that much,
1:14:08 but I I don't know if it's if it's
1:14:10 dramatically different, but I think you
1:14:12 can talk to her. Spoiler alert. Yeah,
1:14:15 exactly. for Sydney. I [laughter] know
1:14:17 that was that was a bit of a spoiler
1:14:19 alert for Sydney. Um yeah, and it was it
1:14:23 was also doing live transcription as it
1:14:25 was talking. It was doing live
1:14:26 transcription. Um
1:14:30 and then I there was a demo there was a
1:14:32 demo of the new real time voice API that
1:14:36 open a OpenAI put out where he was
1:14:39 speaking in French and it was translate.
1:14:43 Was it French? I don't know what it was.
1:14:45 something and it was translating into
1:14:47 English in real time. Um,
1:14:51 yeah. I mean,
1:14:52 >> I can't translate into English yet.
1:14:54 [laughter]
1:14:55 >> Siri still sucks.
1:15:00 S
1:15:01 Siri is just like, I can't I can't do
1:15:04 translation.
1:15:05 Meanwhile, [laughter]
1:15:06 OpenAI is doing real time.
1:15:09 Um, is there something like LM Studio
1:15:12 for the PC? There probably is. Let's
1:15:14 see. Um, what is the
1:15:18 PC?
1:15:20 Probably Olama maybe. Quent
1:15:24 of LM Studio.
1:15:31 Jan, Olama, and GPT for all. I would say
1:15:35 Olama is the one that I've I've heard of
1:15:37 the most, but Jan, O Lama, O L A M A,
1:15:41 and GPT for all for the PC.
1:15:45 Can you go over the details of the
1:15:46 anthropic SpaceX deal? I haven't had a
1:15:48 chance yet. I don't know the details of
1:15:50 it, Chef Kelly, but I know that um so
1:15:54 XAI created a thing called Colossus 1,
1:15:57 which was a massive data center, and I
1:16:00 think it had 200,000 GPUs in it. And
1:16:03 then
1:16:05 nine months after they built that, they
1:16:07 started Colossus 2, which is north of a
1:16:09 million GPUs. Might be like 1.4 million.
1:16:13 I don't know. It's a lot. Um,
1:16:17 so I guess XAI is now over on Colossus 2
1:16:21 and they basically licensed
1:16:25 Colossus one to Anthropic.
1:16:29 So if you've not been paying close
1:16:31 attention or if you've not been
1:16:34 overly using Claude. One of the things
1:16:36 that that happened, so Claude got
1:16:39 because of Claude 4 point Opus 4.6 that
1:16:43 dropped in December of 2025,
1:16:47 there was a bunch of people migrated
1:16:48 from OpenAI to Claude. Um, and so for
1:16:52 December, January, and kind of February,
1:16:58 Claude and Opus 4.6 and Anthropic were
1:17:01 the bell of the ball. Everybody was
1:17:02 loving them. They were fantastic. Then
1:17:05 there was some of that political [ __ ]
1:17:06 that happened with with the Department
1:17:08 of War and you know that kind of muddied
1:17:13 the water waters. And then right around
1:17:15 then um Claude's servers just basically
1:17:20 became overwhelmed and they started rate
1:17:22 limiting the [ __ ] out of people. And
1:17:24 then when they launched Opus 4.7,
1:17:27 they set the default to be uh I think it
1:17:30 was called adaptive thinking where
1:17:32 basically it wouldn't think as deeply um
1:17:36 as the default for Opus 4.6. So the
1:17:40 perception was that Claude was a pain in
1:17:43 the ass because it it was rate limiting
1:17:45 people. Um, and it was dumber because
1:17:50 they weren't letting it think as long
1:17:52 because they ran out of compute
1:17:54 capacity. So, the deal with SpaceX is
1:17:57 basically here's Colossus 1 with these
1:17:59 200 or 400,000 GPUs, whatever it is, use
1:18:04 that for now, right? [laughter]
1:18:08 I also think I mean, if you look at the
1:18:09 politics, right, Elon and Sam Alman are
1:18:12 in the middle of a lawsuit right now.
1:18:14 They're in court right now. And so I
1:18:16 think this is just
1:18:19 Elon just being a dick to Sam. Also,
1:18:22 he's like, "Well, everyone's struggling
1:18:24 with compute right now. We've got this
1:18:26 spare, you know, massive supercomput we
1:18:30 built two years ago. Why don't we give
1:18:32 it to Anthropic to [ __ ] with Sam Alman?"
1:18:35 [laughter] So, I think that's the other
1:18:37 thing that's happening there.
1:18:41 All right, we got about 10 minutes left.
1:18:43 What I'm trying to do with these uh with
1:18:45 with the lives here is make them a bit
1:18:47 more predictable. Start on time. We'll
1:18:48 end it at 9:00. I I know I went over the
1:18:50 other night. Um but I want to I want to
1:18:53 It's Friday night. I want to get you
1:18:54 guys out of here. Crazy how fast AI can
1:18:56 find and analyze info. It's really fast.
1:18:59 If have you noticed um Google
1:19:02 when you do a Google search right now,
1:19:05 the AI answer that it generates is like
1:19:08 bang. It's amazingly fast.
1:19:12 Uh uh uh uh uh uh uh uh other questions.
1:19:17 So tomorrow
1:19:21 somewhere in the mix here is Noam.
1:19:23 No is doing part two of her LOL. So if
1:19:27 you go to the events tab at the top of
1:19:30 the lefthand navigation in the salon, go
1:19:32 to events.
1:19:34 Tomorrow at 400 PM Eastern is Noy's part
1:19:37 two of the LOL. Part one is in the
1:19:39 replay section.
1:19:41 um watch past events or something like
1:19:44 that. Um so you can see that if you
1:19:46 want, but um go check out her session
1:19:50 tomorrow. Um it's really good. She's
1:19:52 been doing really amazing work and it's,
1:19:55 you know, we we've got a lot of people
1:19:57 in the AI salon right now that are doing
1:20:00 things that they could not have imagined
1:20:04 they would do. and she has designed,
1:20:09 built, and launched an iOS app.
1:20:14 iOS apps to get them through the the
1:20:16 approval process at Apple. It's not
1:20:19 trivial to write a native iOS app and
1:20:21 it's not trivial to get them approved.
1:20:23 So, so we're in a world now where you
1:20:25 can start to do stuff you never imagined
1:20:28 possible.
1:20:30 Um,
1:20:32 in the community feed, we've got the the
1:20:35 the the theme for the month of May is
1:20:38 momentum. Producer Brandon put a prompt
1:20:41 of the week in the community feed that's
1:20:43 related to momentum. So, go look at that
1:20:46 prompt and then create some work and
1:20:48 share it. Let's start exploring things.
1:20:53 Um,
1:20:55 one last question.
1:20:58 What's one AI use case that still feels
1:21:02 like magic to you even after using these
1:21:05 tools for years
1:21:08 there? I have two right now and one has
1:21:11 been persistent. The one that just blows
1:21:13 my mind every time I do it iso
1:21:16 making songs in. So, I was on with EPC
1:21:18 today and she does she does these guided
1:21:22 meditations and she wants to learn how
1:21:24 to do kind of chanting music and she,
1:21:27 you know, she she right now pays for a
1:21:30 subscription service for music she can
1:21:32 download. And she thought, you know, is
1:21:34 there a way to do something? And so we
1:21:37 went in there and I just chanted. I just
1:21:39 made some crap up [laughter]
1:21:42 and and just chanted away and it turned
1:21:45 it into like this remarkably
1:21:48 interesting song. And so for me blows my
1:21:51 mind every time. Um midjourney from the
1:21:55 image generation side still just kind of
1:21:58 melts my face in terms of its quality
1:22:00 and nuance. And then the new images 2
1:22:04 model inside chat GBT like Nano Banana
1:22:07 blew me away, right? Nano Banana was the
1:22:09 first first image gen tool that I would
1:22:12 consider a visual reasoning engine.
1:22:16 Images 2, the ability to go in and just
1:22:19 say, "Make me an infographic about this
1:22:21 obscure animal." And then it makes
1:22:25 a full-on poster that like you would see
1:22:28 in a classroom in two or three minutes
1:22:31 is
1:22:33 I
1:22:37 like honest to God it's like
1:22:40 it feels like Star Trek like you know
1:22:42 what I mean like like I remember being a
1:22:44 kid and watching the stuff on Star Trek
1:22:46 or Star Wars and just thinking like
1:22:48 imagine a day where you could just talk
1:22:50 into something and and it could
1:22:52 understand any language. It's it's it it
1:22:55 seems so absolutely foreign from reality
1:22:59 and and you know in the the uh you know
1:23:03 the uh hol in Star Trek
1:23:07 we're essentially there with images and
1:23:09 and probably this year we'll we'll start
1:23:12 to be there with 3D worlds and by the
1:23:14 end of 2027
1:23:17 you'll just be able to speak
1:23:20 worlds into existence that have physics
1:23:24 and game mechanics and the power of AI
1:23:28 LLM inside a 3D space.
1:23:32 [laughter]
1:23:34 It's [ __ ] crazy. Lori Blair, Brandon,
1:23:36 I was surprised again the other day when
1:23:38 I gave the lighting and soil conditions
1:23:41 of an empty space I wanted to garden and
1:23:43 it made an entire an entire workable
1:23:46 plan in about a minute complete with a
1:23:49 picture of what everything would look
1:23:50 like afterwards. Yeah. like like
1:23:55 it's one thing for an image gen tool to
1:23:58 be able to conceive a garden, right?
1:24:02 It's another thing for an image gen tool
1:24:04 to be able to
1:24:07 do the research on the climate and which
1:24:09 plants and when
1:24:12 and do the sort of the mathy
1:24:18 quantitative piece of the garden
1:24:20 planning and then turn that into what
1:24:22 the garden's going to look like all in
1:24:24 the same image. It's it's insane,
1:24:27 Brandon. dropping my MRI images into an
1:24:29 LLM and and it telling me exactly what
1:24:32 the radiologist told me. Yeah,
1:24:38 Mark Hefner.
1:24:40 What's happening? Mark. Hey, Kyle. Happy
1:24:42 Friday evening. Coming back to the first
1:24:44 question of the evening. Models had
1:24:47 bigger differences when they first
1:24:48 started. True. I think they are all
1:24:50 converging to a single point of truth.
1:24:53 Today, those differences are more
1:24:55 nuanced. In my experience, Claude is
1:24:57 still better at creating written content
1:24:58 as long as you train the model on your
1:25:00 writing style with an MD file. I use the
1:25:02 same MD file across all models, then
1:25:05 battle them for the best one,
1:25:07 and take it to the last 15% to complete
1:25:10 completion. In my opinion, all models
1:25:12 have strength and weaknesses. I think
1:25:14 that's absolutely right. And I there's
1:25:16 there's a new term that they're talking
1:25:18 about called the harness.
1:25:21 And I the the harness has always been
1:25:23 there, but I think it's more significant
1:25:25 in agentic tools. The the harness is the
1:25:27 thing that the models kind of sit inside
1:25:29 and there's all the it's it's like the
1:25:32 application layer um that interacts with
1:25:36 all the different models sitting
1:25:37 underneath it. Um and those harnesses
1:25:40 are where the personality is. And I I
1:25:42 think that
1:25:44 I think that that's where you're going
1:25:46 to see, you know, like codeex is a
1:25:48 harness and claude clo code and cloud
1:25:50 co-work is a harness and even, you know,
1:25:52 just claude.ai, the the app that you use
1:25:55 as a harness. So I think it's going to
1:25:58 be in the interface as always, right?
1:26:02 It's going to be within the interface
1:26:03 where the differences start to emerge
1:26:05 because I think the models really do
1:26:07 start to become kind of a single a
1:26:10 single kind of thing.
1:26:12 Um, all right. Beautiful. Uh, we're
1:26:15 coming on the top of the hour, so have a
1:26:18 beautiful weekend. Um, is if if there's
1:26:21 any
1:26:24 any homework this weekend.
1:26:32 Nothing immediately comes to mind.
1:26:35 It's Mother's Day weekend. I would say,
1:26:37 you know, connect with your family.
1:26:41 Think about what you value and who you
1:26:43 value
1:26:45 and then and then carve out some time
1:26:48 for yourself to play with AI. I think
1:26:50 those two things I started with of what
1:26:52 are some good AI habits. One is just go
1:26:56 play go experiment with something that
1:26:58 you've never played with. Go play with
1:27:00 something completely that you would
1:27:02 never consider playing with. Right? If
1:27:04 you're horrible at math, go do something
1:27:06 with a spreadsheet and AI. Right? If
1:27:09 you're horrible at pictures, go do that.
1:27:12 And then the other thing is like really,
1:27:14 you know, look at who you are and what
1:27:16 you value and who you care about and
1:27:19 then think about how you might use AI as
1:27:22 an amplifier of of that of who you are.
1:27:25 That might shift how you think about
1:27:27 which tools you play with.
1:27:30 Oh, that's good. Chef Kelly, make a
1:27:32 magazine layout for your mom in image,
1:27:33 too. Yeah. upload a picture of or two of
1:27:36 your mom and then put in some random
1:27:38 facts and say, "Turn this into a glossy,
1:27:41 you know, magazine layout, you know,
1:27:43 that that goes across both pages."
1:27:45 She'll be blown away. You'll be blown
1:27:47 away. Um, beautiful. We took a picture
1:27:50 of my mom and had AI make a coloring
1:27:52 image for the kids to color in. Yeah.
1:27:54 Um, have it make five pages of a graphic
1:27:57 novel and it will it'll just all five
1:27:59 pages will come out. It's pretty crazy.
1:28:01 Have a great weekend, my friends. See
1:28:02 you on Saturday. Yes, please go to no
1:28:04 ems lol. Okay, go to uh events over
1:28:07 here. RSVP show up tomorrow 4 p.m.
1:28:11 Eastern time.
1:28:14 All right, everybody. Thank you so much.
1:28:16 Uh this has been a wonderful week coming
1:28:18 back inside the AI salon. I'm really
1:28:21 excited about this. Um thank you to Andy
1:28:24 Scarantino and Brandon who push pushed
1:28:27 me. What? How did I describe it? getting
1:28:30 me to to make this shift is like milking
1:28:32 a stone. Um, I'm really glad I did it.
1:28:35 I'm really glad we're back inside the
1:28:37 salon and uh I'm I'm really glad you're
1:28:39 all here. I very much appreciate you as
1:28:42 always and uh I hope you have a great
1:28:44 weekend everyone. All right, peace.