8/13/2025 - ChatGPT 5's New Tricks: Models, Personalities, and Prompts

Description

Chapters

Transcript

0:08 [Music]
0:15 Woohoo!
0:19 Woohoo!
0:22 Woohoo!
0:24 Woohoo!
0:28 Woohoo!
0:30 [Music]
0:35 You're feeling nervous, aren't you, boy?
0:39 With your quiet voice
0:42 and impeccable style.
0:47 Don't ever let them steal your joy
0:50 in your gentle ways to keep them from
0:54 running wild.
0:56 [Music]
0:57 They can kick dirt in your face, dress
1:00 you down, and tell you that your place
1:02 is in the middle. When they hate the way
1:05 shine,
1:10 I see you tugging on your shirt,
1:14 trying to hide inside of it. Hide how
1:17 much it hurts.
1:18 [Music]
1:20 Let them laugh
1:22 while they care.
1:25 Let them spin.
1:28 Let them scatter in the wind.
1:33 I've been to the movies. I've seen how
1:36 dance
1:39 and the jokes on there.
1:50 [Applause]
1:51 You get discouraged, don't you, girl?
1:55 It's your brother's world for a while
1:58 longer.
1:59 [Music]
2:01 We got to dance with the devil on a
2:03 river to beat the stream.
2:06 Call it living the dream.
2:09 Call it kicking the ladder.
2:12 They come to kick dirt in your face. to
2:15 call you weak and then displaced you
2:17 after carrying your baby on your back
2:19 across the desert.
2:22 [Music]
2:25 Uh I see you your eyes behind your hair
2:30 and you're looking tired
2:32 but you don't look scared.
2:35 Let them laugh while we can.
2:41 Let them spin.
2:43 Head up scatter in the wind.
2:48 I've been to the movies. I've seen hot
2:54 and the jokes
2:56 on them.
3:02 [Music]
3:09 Oh, Wednesday night. It is Wednesday
3:12 night. It is Wednesday night. What is
3:14 happening? The good people. The good
3:17 people.
3:20 Nobody cares about AI on a Wednesday.
3:24 It was a small but appreciative crowd.
3:35 [Music]
3:47 All right.
3:49 [Music]
4:17 Woohoo!
4:20 [Music]
4:34 Oh, man. Alive. Man alive. Man alive.
4:38 Man alive. Man alive. What are we going
4:40 to do tonight?
4:42 I think we're gonna play around with
4:45 chat jeepet.
4:48 Chat jeepet 5.
4:50 If you've not played with chat jeepet 5,
4:53 what are you doing with your life? Be
4:55 not afraid. Uh, I shared a code for pica
4:58 social app in a regulars earlier. Cool.
5:01 Did it already get snatched up?
5:10 100 invites. Holy goodness. All right.
5:13 Well, so you know what we'll do then.
5:17 Beautiful.
5:19 Love it.
5:22 No YouTube. Yeah, we're on YouTube. Did
5:25 I stream to the wrong YouTube? No.
5:30 I'm on YouTube.
5:35 All right. So, what we're going to do to
5:37 people,
5:38 there's only a handful of people in
5:40 here, but if you producer Brandon
5:45 verified, we are on YouTube. Vicki, I
5:48 don't know what's up with your YouTube,
5:49 but it does not sound promising.
5:52 Sounds very distressing.
5:54 I would not put up with that from
5:56 YouTube.
5:59 All right, archetypal architect can see
6:01 us.
6:06 It's my TV. You're on my laptop, but not
6:09 my TV. Wait, I'm not on TV. No. Say it
6:13 ain't so. Say it ain't so, Joe.
6:18 We are getting less and less.
6:23 Well, that's not on you, Mr. It. That's
6:25 on me.
6:29 [Music]
6:46 [Music]
7:11 [Music]
7:28 Let us be lovers. We'll marry our
7:31 fortunes together.
7:34 [Music]
7:36 I've got some real estate here in my
7:38 bag.
7:41 >> So, we bought a pack of cigarettes
7:45 and Mrs. Wagler's pass.
7:48 And we've all come to look for America.
7:54 [Music]
8:05 [Music]
8:16 Kathy, I said as we
8:18 init
8:20 [Music]
8:22 Michigan seems like a dream to me.
8:29 It took me It took me four weeks. It
8:33 took me It took me
8:39 uh I don't know. All right. Um
8:42 Okay. What are we gonna do tonight?
8:43 Tonight. So, a couple of things.
8:46 Let me flip over to yan
8:50 yan that and do my black bar magic.
8:53 There we go.
8:56 All right, we'll do that. We'll do that.
8:59 We'll do that.
9:01 All right, I think we're good. How's
9:04 We'll do that. Now we're better.
9:07 Steo's going to be right back. Vicki's
9:10 here.
9:12 Sharon Crawford, howdy, howdy, howdy.
9:15 Seriously, it says on my TV, "I'm
9:18 waiting." I know that's not right.
9:21 [Music]
9:25 Sharon Crawford, I'm late. Even hubby
9:27 was saying, "Hey,
9:31 you gota you got to train those spouses.
9:34 You got to train those spouses for the
9:36 uh AI learning lab.
9:39 [Music]
9:47 Mrs. Producer woke me up 15 minutes ago.
9:52 Vickiy's rebooting her TV. Brandon's
9:54 rebrooting his brain.
9:56 [Music]
10:06 Oh, what's the one that's down here?
10:08 This one is
10:10 [Music]
10:11 Wait.
10:14 [Music]
10:29 10,000 words swarm round my head. Wait,
10:31 is that right? No.
10:40 10,000 swing.
10:42 [Music]
10:53 10,000 words swim around my head. 10
10:56 million more in books written beneath my
10:59 bed.
11:04 I wrote or read them all when searching
11:07 in the swamps. Still can't find how to
11:10 hold my hands.
11:14 [Music]
11:15 And I know you need me in the next room
11:18 over. I am stuck in here paralyzed.
11:26 For months I caught myself in rots. Too
11:29 much time spinning
11:32 mirrors framed in yellow walls.
11:37 [Music]
11:39 Ain't it like most people? I'm no
11:42 different. Like to talk on things we
11:45 don't know about.
11:50 Ain't it like most people? I'm no
11:52 different. like to talk on things we
11:55 don't know about.
11:59 All right, whatever. I'm not feeling the
12:02 guitar tonight.
12:04 [ __ ] music. Music's a waste of
12:05 everyone's time.
12:08 You can't do math with music. I mean,
12:11 sure, you could do Prague rock and where
12:13 where it's all math. You can absolutely
12:16 do math with music. I'm just bitter. I'm
12:19 just bitter. Bitter. You betcha. Hi, I'm
12:22 Kyle.
12:24 I'm the uh host of the AI learning lab
12:26 here on Tik Tok and YouTube and LinkedIn
12:28 and X. I like to call X Twitter because
12:31 that's what it used to be called. And
12:33 I'm stubborn and I'm stuck in my ways. I
12:36 haven't grown a bit in 37 years. What
12:39 happened 37 years ago? I don't know.
12:42 Just kind of the last time I have a
12:43 memory of anything resembling progress
12:45 in my life.
13:02 while he calls you Marty. His heart is
13:04 in the right place. Oh, it's AI Marty.
13:08 It is AI Marty.
13:11 Um, math music, Kyle. Life is good,
13:14 Kyle. It's good. Life is good. Life is
13:16 good.
13:18 Maybe start with chat GPT personalities.
13:20 The hell are you talking about?
13:24 Oh, does it have personalities now? Is
13:26 that something? We talked about that,
13:27 didn't we? Does it have custom
13:30 personalities?
13:32 There's customized chat GPT
13:35 advanced.
13:38 No
13:42 settings.
13:45 Personalization.
13:47 I scrolled past it. Dang it.
13:52 Custom instructions.
13:55 Oh,
13:56 what personality should chat GPT have?
14:03 Oh my god. Listen, people. People,
14:06 people,
14:08 if I've told you once, I've told you a
14:10 thousand times. And our one of our
14:11 guests last night couldn't accept this
14:14 fact that there are no experts.
14:17 There are no experts. We're all starting
14:19 over. We are all starting over. Okay. Um
14:24 so what we're going to do
14:29 Oh my god. I don't even know where to
14:31 begin.
14:35 Don't even know where to begin.
14:38 Okay. So, if you're new here, and if
14:41 you're new to AI, welcome.
14:46 My name is Kyle. This AI learning lab.
14:48 We do some some learning here. We do
14:50 some some teaching, some learning. It's
14:52 mostly just me fumbling around with
14:54 tools I don't quite yet understand.
14:57 The reason that's valuable is because
15:00 none of us understand them. Now, it's
15:02 possible you've spent the past week
15:04 looking at every nook and cranny of chat
15:06 GPT5, but my guess is you've just been
15:09 living your life and occasionally check
15:10 in with chat GPT and like it still sort
15:13 of works. Or if they took away your
15:15 girlfriend with GPT40, they give it back
15:18 to you now. So, you're probably back to
15:20 being adjusted and antisocial.
15:23 You know, your happy place. Um,
15:28 but I figure what we're going to start
15:29 doing is we're just going to start
15:30 playing with chat GBT. So, let me walk
15:32 you through a little bit of the
15:34 interface,
15:35 trying to put in the Pika code, but
15:37 can't find where to put the code. So,
15:39 where you put the code is you have to
15:42 download the Pika social video app. So,
15:47 so right now for those of you that are
15:50 not part of the AI salon, you should go
15:53 to community.thesalon.ai
15:58 and join.
16:00 And then in the um left hand side, the
16:05 lefth hand navigation, you scroll down
16:07 until you see the irregulars channel
16:11 and the
16:12 what post?
16:15 I think it's the first post.
16:19 Having to reload. Hang on. Yeah, the
16:22 first post. Danielle
16:25 um put in a code for a 100 people to get
16:29 the Oh, yeah. A code for 100 people to
16:32 get the Pika Social app. Okay. It's on
16:35 iOS only. There's two different apps
16:38 from Pika Labs. One is the Pika app and
16:41 one is Pika Social. get picocial
16:45 and then you put in this code when you
16:46 go to register.
16:49 I typed in the social website.
16:52 I didn't know social had a website. What
16:55 is it? social.pika.art
17:00 or did you just do it in the regular? I
17:01 think it's I think this is only
17:03 available on iOS right now.
17:06 Source camp.
17:13 Whatever address Danielle put in,
17:17 the address she put in
17:23 takes you to an expost.
17:31 Well, so that just takes you to an
17:33 expose. So, what you have to do,
17:42 what you should do is go um of course
17:46 there's a social pika.com that's not
17:48 related. Okay. Yeah, of course.
17:55 Anyway, all right. Let's get back to
17:57 chat GBT.
17:59 um download the iOS app of Pika Social
18:04 and then go get that code from
18:06 Irregulars. And it looks like there was
18:08 a hundred available, but I think that
18:10 was on Twitter. So I assume those are
18:12 probably going pretty quick if if not
18:15 has already gone.
18:19 Oh, it was Discord. Okay, cool. Got it.
18:27 T
18:30 [Music]
18:34 I'll get the link for the post and drop
18:36 it in there now. Okay, good. You guys
18:37 are all doing that. Okay, so back to
18:40 chat GPT.
18:44 So, one of the promises of the new chat
18:47 GPT5 is that they would hide all the
18:49 models. You didn't have to pick models
18:51 anymore. They've undone that. Um,
18:56 GPT5 auto is what they had launched
19:01 and then they quickly added thinking so
19:04 you could explicitly tell it to think
19:05 because their router was screwed up. And
19:08 now they've now they allow you to just
19:10 manually switch between fast thinking
19:13 mini and thinking and then pro.
19:16 And then you can also enable legacy
19:19 models. So, if you know enough about
19:22 legacy models, you can go play with
19:23 them. I I don't know why you would
19:26 unless you're really attached to chat
19:27 GPT4, but you can probably make five act
19:30 like four. Okay,
19:33 so I figure what we'll do tonight is
19:34 just play around with chat GPT5 and
19:37 maybe play around with these different
19:38 models and just
19:41 give it give it some things. Um,
19:46 when I did the AI readiness project
19:48 today with Ann Murphy and um, Vanessa
19:51 Chang,
19:52 Ann came up with a really good idea. She
19:54 she stumbled onto
19:57 she stumbled onto what's a really good
19:58 way to test chat GPT5,
20:01 which is she was on a call with a client
20:03 and she was doing a little add work on
20:06 the side. She was trying to get a prompt
20:08 to work for that client and she'd been
20:11 trying for three or four months to kind
20:12 of crack the code on the positioning of
20:14 this thing they were talking about. And
20:16 because she was on a call, she didn't
20:19 have a lot of attention to put to chat
20:21 GPT. So, she put in a really short
20:23 crappy prompt, but it was enough. And
20:26 chat GPT5 gave her what she wanted that
20:30 she hadn't been able to get in months
20:32 from chat GPT4.
20:35 So, an idea we came up with live on the
20:38 podcast was
20:40 an interesting thing to do might be back
20:42 down your prompting. Like, if you're
20:44 doing relatively structured, complicated
20:46 prompts with chat GPT4, a thing to play
20:49 with is sort of backing it down and do
20:51 doing something really simple.
20:53 All right,
20:55 Steo. Okay, guys, I have three codes.
20:57 I'll post it here. All right, so Steo
20:59 just put three codes for Pika, so jump
21:02 in and grab those.
21:04 Um, let me jump back over to here so I
21:08 can see comments.
21:11 Get the black bar going again. All
21:13 right. So, now we're going to jump in
21:14 here. So, now let's let's jump in and
21:17 say, um, what are we going to ask this?
21:19 Um,
21:22 let's see.
21:24 Uh,
21:28 why
21:31 do tomato
21:38 plants.
21:42 Why did tomatoes
21:45 not get red on plants
21:49 with too much sun?
21:57 They can stall in ripening when they're
21:59 getting too much direct intense sun,
22:01 especially in hot conditions. The reason
22:03 isn't that sunlight turns them red.
22:06 Ripening is mostly a chemical process.
22:08 But the heat and stress from excessive
22:11 sun can disrupt the process. Ripening
22:14 depends on the lycopine and keratene
22:17 production.
22:20 Yeah. So when temperatures spike above
22:23 85, we've been above 85 here for
22:27 months, a month and a half. Too much sun
22:30 can suncald.
22:32 All right. Well, that's why we don't
22:34 have green tomatoes. I mean, red
22:36 tomatoes.
22:39 Source Cam still having trouble.
22:42 The Pika Social app link is at the top
22:44 of irregulars. Okay, let me go look
22:47 here. We'll share this tab. Let me
22:49 reload this.
22:54 All right. Install the Pika Social app.
23:01 then put in that code. All right, there
23:03 you go.
23:09 Oh, it's at the top of your regular's
23:11 chat.
23:13 There you go. So, go to a regular's
23:16 chat. I have the code. I don't know
23:17 where to insert it. Do you have the iOS
23:20 app, Source Camp?
23:23 If you have the iOS app,
23:27 I think I think you click log in and
23:30 then it says you need a code.
23:33 So So click log in and then it'll ask
23:36 you for a code.
23:38 Something like that. All right.
23:46 Gy,
23:50 get in my belly.
23:53 Um, okay. That's not a good that's not a
23:56 good prompt.
23:59 Although that this could be a good
24:00 prompt. Okay. So, we're going to do new
24:02 chat.
24:04 So, now I'm going to go try this
24:06 personality thing. We're going to go to
24:08 settings. We're going to go to
24:09 personalization.
24:11 We're going to go to custom
24:12 instructions.
24:16 What personality should chat GPT have?
24:19 Cynic. Okay, we're going to switch chat
24:21 GPT to cynic mode.
24:25 That's kind of fun. And then we'll go
24:28 get our prompts for tomato ripening. So,
24:31 we're going to do the exact same prompt
24:32 for all these different things.
24:35 So, there's your just basic answer. We
24:37 don't want basic.
24:42 All right. So now we've got our chat GPT
24:44 set to synic.
24:50 Oh, it went into thinking mode. That's
24:52 interesting. So this thing's
24:54 automatically
24:55 flipping between
24:58 non-thinking and thinking modes, which
25:00 is what it's supposed to do. We're in
25:01 auto.
25:03 Or wait, am I in auto? Oh no, I switched
25:05 to thinking.
25:15 That's not very cynical,
25:20 champ.
25:41 All right, Champ has been relieved of
25:43 his duties.
25:45 Yes, tomatoes love the sun, just not
25:47 surface of mercury sun. So, I guess
25:49 there's the cynic.
25:51 Keep the fruit a bit colder colder and
25:53 they'll reen like they remembered their
25:55 job. Okay. Yeah, it's a little bit of a
25:56 smartass, but I had this on thinking. I
25:59 thought I I wanted this on auto, but
26:01 we're on auto now. Okay. So, all right.
26:04 new chat. So, let's go change the
26:05 personality again.
26:09 And we're going to change we're going to
26:10 go personalization
26:13 and then we're going to go personality
26:14 cynic and we're going to switch it to
26:16 robot efficient and blunt.
26:22 Tik Tok pin silver fox. I have so many
26:25 devices streaming in my house. It sounds
26:28 like a kennel of champs.
26:34 Sorry about that, source cam. Okay, so
26:37 now we've got chat GPT is going to act
26:40 like a robot. Efficient. So why do
26:43 tomatoes not get red on plants with too
26:45 much sun?
26:49 Excessive sun can slow or prevent Okay.
26:54 Like when fruit gets too hot, especially
26:56 in direct sunlight, it can stay green or
26:58 turn a yellowish orange because only
27:00 keratene and heat tolerant pigments
27:03 remain. This is called suncald if tissue
27:06 damage also occurs. In high heat, the
27:09 plant may also shift energy into
27:11 survival, reducing pigment production.
27:14 Well, I've got survival mode tomato
27:16 plants right now. It sucks.
27:19 It's very mash- robotic. Oh, yeah. It's
27:21 very M dash heavy, isn't it? That's
27:23 funny. Okay, so new chat.
27:26 I'm going to go try another personality
27:28 now. Customize GPT. Scroll down. That
27:31 was robot. Here's listener. Thoughtful
27:34 and supportive.
27:36 All right.
27:38 So, let's see how thoughtful and
27:39 supportive this one is. My My tomatoes
27:42 aren't turning red. What do I do? Chat
27:44 GPT.
27:49 Oh, man.
27:51 When tomato plants get too much intense
27:53 sun, especially in hot climates, they
27:54 can stall. It's not the lightest stuff.
27:56 Here's what's going on.
27:59 All right.
28:01 If you want, I can walk you through a
28:03 quick trick growers use to get heat
28:05 stalled tomatoes. Yeah, I want the
28:07 trick. Yes, please. On trick,
28:16 the cool and cozy method. Give them some
28:19 afternoon shade. Use a piece of piece of
28:21 shade cloth, an old bed sheet, or even a
28:24 cardboard lean to on the sunny side. Aim
28:28 to block the harshest sun from about 1
28:30 to 4 p.m.
28:32 The fruit. This drops fruit temp enough
28:35 for ripening to begin. Light pruning,
28:38 but not a haircut. Thin a few leaves.
28:41 Mulch the soil. All right, so that was
28:45 helpful. Which one? Which which one was
28:47 that? What did I call that one? The
28:50 listener. And then the final one is the
28:52 nerd.
28:54 All right, we're gonna nerd out. This is
28:57 good.
29:00 Um, I find this kind of cute that all of
29:02 these frontier model companies,
29:07 you know, you have in your head
29:11 that these are all fancy ass
29:12 mathematicians and things, but like
29:16 changing the personality of your chat
29:18 GPT. This is just them prewriting some
29:20 prompts and saying act like a nerd and
29:23 give lots of details. So now it's gonna
29:26 do that. So, this is basically you could
29:28 do this you could do this on your own
29:30 with custom instructions, but they
29:31 basically just did it for you. Um,
29:36 this doesn't seem all that nerdy. The
29:37 result the All right, this is this is
29:40 hard to see. This example is not very
29:43 good to see these personalities. All
29:45 right, whatever. Um,
29:49 okay.
29:50 Wonder if there's anything else we
29:52 should look at in settings.
29:54 Theme
29:57 accent color. Oh, we can change accent
29:58 colors. Let's go with
30:03 blue or orange.
30:06 Oh, change it back to default before you
30:08 forget. There's producer band Brandon
30:11 having my back on the uh on yield
30:16 add side.
30:22 back to default. Cheery and informative.
30:26 Okay.
30:29 Um,
30:32 what was I looking at?
30:35 Settings. Accent color. Let's go with
30:38 orange.
30:41 Language. Auto detect. Spoken language.
30:43 Voice. Show additional models. Show
30:46 follow-up suggestions.
30:49 Is black accent still $200?
30:53 I don't have a black accent. So, yes.
31:00 Oh, look. There's there's the orange. Do
31:03 I like orange? No, I don't. I want
31:05 something else.
31:09 Votes blue.
31:13 That looks corporate and boring.
31:20 green. Maybe
31:23 that looks kind of different. We'll
31:25 stick with that for a bit. Strength,
31:27 peace, and blessings everyone to you
31:29 too. Oceans in the house.
31:32 Um Jeff Flanigan just got here late.
31:39 Source Camp still working on getting her
31:41 code in her in her thing. Working with
31:44 Danielle directly now. It's good.
31:46 Irregulars helping one another. This is
31:48 one of the things we do here. You get
31:50 stuck on something, you're like, "Hey,
31:52 hey, hey,
31:55 little help."
31:58 And then people do. All right. Um,
32:03 let's do something. Let's play here.
32:11 I'm going to write a prompt.
32:25 Write me
32:29 an
32:30 epic poem
32:34 combined
32:36 with
32:39 a hiphop
32:41 cipher.
32:44 combined
32:46 with
32:54 Shakespeareian
32:58 amic pentameter
33:02 in a
33:06 huid huge poem
33:09 about
33:21 the remarkable
33:23 accomplishments
33:26 of Sir Isaac Newton
33:33 by the time he turned
33:36 26.
33:39 I learned that from Neil deGrasse Tyson.
33:41 the other day. All right. So, let me
33:43 copy this prompt. I spelled
33:44 Shakespearean wrong, but I don't care.
33:47 By the way, if you don't know Oh, yeah.
33:49 Look at my green my green highlights.
33:52 Watching on the big screen tonight. Hey,
33:53 what's happening, Keith? Okay,
33:58 Jeff Flanigan. What? Huh? Did I hear my
34:00 name? Yeah, I said you were late. Source
34:03 camps off gallivanting with Danielle
34:06 figuring out how to do some stuff. You
34:08 showed up late. Ocean's here. Just a
34:10 whole thing happened. I changed my
34:13 highlight color to green. Look, we've
34:15 got green now. I like it.
34:18 Um, I wonder what it would show up
34:20 better on the on the screen here for
34:22 y'all. Maybe purple or pink could be
34:26 could be useful. We'll see.
34:29 For some reason, I've always liked the
34:30 words amic pent pentameter. Enough about
34:33 me. I I'm the same way. I There's
34:35 something amic pentameter. There's
34:37 something very beautiful about the the
34:40 syllables there.
34:42 Um, okay. So, what I'm going to do is
34:46 I'm going to switch to chat GPT5 fast.
34:51 So, let's see how fast this does this.
34:53 Bang.
34:55 Thinking a little bit.
34:58 Thinking a lot.
35:07 [Laughter]
35:11 That's not very fast. I have a feeling
35:14 fast is a relative term. Um, now it
35:18 seems to be going fast. Okay, so let's
35:20 see how it did. Here's a massive
35:22 theatrical rhymelaced amic pentameter.
35:26 We'll be the judge of that
35:29 loa
35:30 loather.
35:32 Wait. Lo, gather ye round both scholars
35:36 and
35:38 wait both scholars and the streets.
35:41 That's 11. Where rhyme and reason in a
35:46 cipher meets
35:48 the year is ripe. The mind of blazing
35:50 star. A lad named Newton taken
35:53 thought too far. All right. It's okay
35:57 but not great. Shakespearean interlude.
36:00 All right. So, it did it once it got
36:02 going. It did it sort of fast. Let's do
36:04 a new chat now. Let's flip this into
36:07 thinking mini. So, this should be
36:08 better.
36:12 Did I spell Isaac wrong?
36:15 Is that right? Yes.
36:19 Source camp got it sorted thanks to
36:21 Danielle. Yes.
36:25 Way to go. Irregulars supporting one
36:26 another. That's what we like around
36:28 here. All right. Thinking quick answer,
36:31 considering poem details, evaluating
36:33 use, refining poem length and approach.
36:38 >> Hey guys, I just want to point out that
36:40 it gives you the option to flip the
36:41 model while it's doing it. So like if
36:43 you wanted to go back to fast, you could
36:45 click get a quick answer and it will
36:47 override the fact that it's and allow
36:51 you to change it back.
36:52 >> That's cool. Sonnet of the quiet floors.
36:55 When plague When When plague shuts
36:57 courts and Cambridge clo wait, I got to
37:00 do this with my fingers. When plague
37:03 shut courts and Cambridge closed its
37:06 halls. 10. Young Isaac fled to fields
37:10 where quiet rains. 10. In solitude he
37:15 sharpened thought and sight. 10. He drew
37:18 the flux. Okay, so this is already
37:20 better. Fast didn't do actual amic
37:24 pentameter. You would expect a thinking
37:26 model to be able to do amic pentameter.
37:29 When plague shut courts and Cambridge
37:31 closed its halls, young Isaac fled to
37:33 fields where quiet reigns. In solitude,
37:36 he sharpened thought and sight. He drew
37:38 the flux and learned the shape.
37:42 Learned to shape the change. He watched
37:44 the spectrum
37:46 break a single beam. He proved that
37:48 white was braided from the lights. He
37:50 bent the mirrors, built a glassy eye,
37:53 and chased the stars with metal, wood,
37:55 and wire.
37:59 Hip hop voices. Oh, that's cool. Call
38:02 response. Epic narrative. Oh, so this
38:05 this mixed all three. This did different
38:08 sections. One in epic poem sty or one in
38:11 epic pentameter, one in epic poem style,
38:13 and one in hip-hop
38:16 cipher style. Interesting. All right,
38:19 let's flip to thinking. So now we're
38:22 going to do big big think.
38:33 So many new features to play with. I
38:34 know it's Well, that's that's the thing,
38:36 Jeff, is
38:40 I just I don't think that there's a real
38:43 way to learn this without doing kind of
38:46 stupid [ __ ] like I'm doing now. Like do
38:49 stuff that is just not
38:52 normal. Do normal [ __ ] but just do do
38:55 wacky [ __ ]
38:58 Oh muse laced up the light with measured
39:00 fire. Let drums of reason throb neath
39:04 the liar.
39:06 Oh muse. Oh muse lace up the night with
39:10 measured fire. Let drums of reason throb
39:13 beneath the liar. Yeah,
39:17 I stalk the ghost of motion's hidden
39:18 heart. I test the world and make it
39:20 speak its heart. Where apples fall, I
39:22 map the false decree and bind the earth
39:26 to moon by one same key.
39:30 Oh, you know what would be fun? Let's go
39:31 change our personality now
39:35 to something more creative.
39:38 Let's do listener. Thoughtful and
39:39 supportive.
39:41 We'll do a new chat. We're in thinking.
39:44 Same prompt.
39:54 Rhyme, rhyme, rhyme, rhyme, rhyme.
40:00 Good day to you, s planning the huge
40:03 poem.
40:09 Oh muse, tune up thy drum and liar as
40:11 one.
40:13 Let quill and kick drums share a single
40:16 son. Attend the stage where reason rocks
40:18 the crown. The boy from Wolvesthorp
40:22 writ the cosmos down before 20 and six
40:25 winters iced his brow. He flipped the
40:28 chalk and made the heavens bow. I call
40:31 the circle gather minds and hearts. The
40:34 ciphers lit. The beat of thought
40:37 departs. Oh, I have a fun idea. Okay,
40:41 welcome to Chat Add. If you're new here,
40:44 one of the things we do is follow my
40:47 shitty ass brain, which which is all
40:49 over the place. And so when I have a an
40:52 idea, we run with it because I got
40:55 producer Brandon, but he's just like,
40:57 you know, he's probably had a gummy at
40:59 this point. and he's got kids nipping at
41:01 his ankles and dogs barking and pooping
41:04 and doing just there's a lot of things
41:06 going on there. So, I'm essentially on
41:08 my own and you all are just like helping
41:12 each other install apps and things like
41:14 that. So, what we're going to do is
41:16 we're going to go over to Manis
41:20 and we're going to do something called
41:21 wide research. And I I have an idea for
41:24 something we can do for Isaac Newton.
41:27 Um that that's not poetry related, but
41:31 it is research related. So I want to
41:33 play this um video. I think it's 45
41:37 seconds or something like that. Uh oh, I
41:39 can see Brandon's about to type. He's
41:41 He's about He's about ready to watch me
41:43 hose this thing.
41:46 Just check on audio. I'm pretty sure I'm
41:50 sharing these as tabs, so I think we're
41:52 good. Okay. But you tell me if
41:54 different.
41:56 Um,
41:58 so Manis, if you don't know what Manis
42:01 is, Manis is like,
42:04 you know, about a month ago when when
42:07 OpenAI launched GPT agent and they
42:10 thought it they they made a big deal of
42:12 of this new agentic capability. Well,
42:14 man, this has been out for like eight
42:15 months and it it does it better than
42:18 chat GPT does. This one and GenSpark are
42:20 the two biggies. They just introduced
42:22 this thing called wide research, which I
42:24 don't quite get, but we're going to
42:26 watch the video and then we're actually
42:28 going to turn this thing loose and see
42:29 what it does. Okay. Tackle complex
42:32 research in a single step with wide
42:34 research. Manis breaks down massive
42:36 largecale tasks into multiple subtasks
42:40 and runs them all at once. So, it's
42:42 doing parallel. So, think of deep
42:45 research as this agent goes out to a
42:48 bunch of websites and runs down one
42:51 concept,
42:54 a bunch of websites, learns them, comes
42:56 back, gets the information, and then
42:58 maybe goes and does does it down another
43:00 one. This does them all in parallel. All
43:02 right, it runs them all at once.
43:04 Launching today for pro users with
43:06 gradual rollout for for basic. Uh, I
43:09 don't know if I'm pro or not, but let's
43:10 watch the video.
43:13 Introducing Wide Research. Got a mega
43:15 task to dig into, 50 companies to
43:17 profile, 100 leads to track down. Don't
43:20 sweat it. Now Manis handles it in
43:23 minutes.
43:26 With wide research, one short prompt,
43:28 big wide research.
43:30 Just tell Manis your research goals.
43:32 Manis splits the job into multiple
43:34 subtasks, finds the data, and runs
43:37 everything in parallel.
43:40 Everything happens all at once.
43:42 [Music]
43:45 No matter how complex, no matter the
43:47 scale, you'll never need to manage
43:49 multiple windows again. Prompt once, run
43:52 wide. No more digging.
43:55 Leave it to Manis.
43:57 >> It's actually a pretty good little
43:58 little ad. I like the music. Um, okay,
44:02 cool. So, do we have access to this? I
44:05 may not.
44:13 I do not think I do.
44:17 Oh, there was a try it button
44:26 recommend.
44:37 Explore more use cases.
44:40 I think I screwed up. I got rid of
44:46 I got rid of that window. Can I bring it
44:48 back? Incognito mode. Yeah, maybe that
44:51 new incognito mode. Manis dot. What is
44:55 it? IM
45:02 lm.
45:22 Double credit promotion.
45:25 Free credits received.
45:28 Reload. This is where I got it before
45:29 when I reloaded.
45:31 Not there anymore. Dang it.
45:48 Oops. Sharing just went away. That was
45:51 weird.
45:56 Well, this is kind of a bummer.
46:00 Kind of a bummer, dude.
46:06 Um
46:08 slides.
46:37 So, they have different styles of
46:39 slides. That's kind of fun.
46:41 We'll do handdrawn. Oh, wait. We'll do
46:44 Yeah, we'll do handdrawn.
46:47 Um, I don't know where Wide Research is.
46:49 I don't have access to it, and I would
46:51 be disappointed if I were you. But what
46:54 we're going to do is we're going to say,
46:56 I want you to research
47:00 all of the things
47:04 Sir Isaac Newton
47:10 discovered or authored
47:14 and
47:17 the
47:20 evolution
47:22 of
47:24 products
47:26 or products
47:31 or life improvements
47:35 from his day
47:38 to today.
47:44 Get as detailed as possible
47:49 and make
47:53 an unforgettable
47:58 presentation
48:02 as if we
48:07 were thanking him. And I know this is
48:09 painful listening to me type. Whatever.
48:12 We're thanking him for
48:16 his
48:18 contributions
48:20 to society.
48:22 Uh, I'm bummed we can't do wide, but
48:24 whatever. All right. If someone figures
48:26 out how to do wide, let me know. All
48:28 right. So, Manis is off thinking, so
48:29 let's go back to chat. Jeeta,
48:33 share this tab instead.
48:36 Okay.
48:38 Um,
48:42 so I did this.
48:44 Oh muse tuna tune up thy drum and liar
48:48 is one. Let Quill and kick drum share
48:50 the single song. Wait, let's do a new
48:53 chat.
48:56 Let's do make sure the personality was
48:59 set to listener. It is
49:04 put in our prompt. See what we get.
49:08 All right. Anyone are different styles
49:11 nerd, etc. on the mobile app, too, or
49:13 just desktop? I don't that I don't know,
49:15 Silver Fox. That's that's a good thing
49:17 to check.
49:34 Newton Newton's 26, an epic cipher in
49:38 amic time. Behold the boy from wolf
49:42 woolsorps winter frost who trimmed the
49:45 dark and counted out its cost. A candle
49:48 flame a slate of midnight skies.
49:51 He split the night and taught the dawn
49:54 to rise when plague winds closed the
49:56 gates of Cambridge town. Okay. Cipher
49:59 one. I step inside the flag with fluxion
50:02 flow. My bars are rates of change. Watch
50:05 moments grow.
50:22 I have an idea.
50:55 Let's go make this into a song.
50:59 So, where we're going to go
51:02 to share this tab instead. We're going
51:04 to go to producer.ai.
51:14 So, I'm gonna say um I want
51:18 an Elizabeth Ethan
51:23 hiphop
51:34 jam
51:37 that can handle
51:45 these
51:47 amic
51:49 pentameter
51:54 uh words.
51:58 All right.
52:07 This is brilliant and wildly ambitious.
52:09 Elizabeth in language, amic pentameter,
52:13 and hard Newtonian math with a hip-hop
52:15 edge. Total genre mash mashup with
52:18 history, science, and swagger tied up in
52:21 one. Love that you already brought the
52:24 lyrics and ciphers frames are tight.
52:27 Let's shape this into a full track.
52:29 Elizabeth and hip hop, but with
52:32 production that nods to both eras. Okay.
52:34 Sound prompt. Elizabeth and hiphop jam.
52:38 Mix of Baroque instruments, loot, harps,
52:40 accord violles and strings and hard 90s
52:45 boom bap hip hop beats
52:48 structure
52:49 demo flow example.
52:54 All right, it's describing it. Ready to
52:55 make it?
52:57 Yes, let's go. If you haven't used
53:01 producer.ai, I don't know if it's widely
53:03 available yet. Yeah, this is
53:06 producer.ai.
53:09 Oh, Vicky Baptiste. Wide research.
53:11 Nothing special. You just put it in and
53:13 watch it do its thing. Oh, that's
53:15 interesting.
53:18 All right. It's off making its slides.
53:24 Oh, this is going to be cool. These
53:26 slides look really cool. Okay.
53:29 Um, where am I? Here I am.
53:35 creating Annis Mirabalis.
53:38 [Music]
53:49 I step inside the ring with fluction
53:51 flow. My bars are rates of change. Watch
53:53 moments grow from fluent streams to
53:55 slopes that kiss the curve. We chase the
53:58 instance ghost with fearless nerve. No
54:00 guesswork. Limits whisper tight and
54:02 true. The changes breath becomes a break
54:04 through you. We solve the fall, the
54:07 flight, the twist spin. The beat divides
54:09 and laws of motion win.
54:13 [Music]
54:16 >> Where time runs thin. He measures
54:18 slipping speed and caught the shape that
54:20 hides emotions need a fluent moves as
54:23 flu marks the trace. A quiet gram of
54:26 nature can't a face. He tamed the slope
54:29 of lines that kiss a Ben make curving
54:31 path for simple symbols in some the
54:34 grain of space he sin
54:37 the testim
54:40 all right so
54:42 uh how do I go back let me go back here
54:44 okay so there's our there's our little
54:47 song
54:49 um
54:50 let's see make it more dramatic extend
54:52 the last cipher add more choir let's say
54:56 more choir
55:02 More Eliza Bethan.
55:07 We'll say more choir. We'll do
55:11 Gregarian chant.
55:15 Gregorian chant, I think it is.
55:18 Gregorian chant. um more Elizabethan
55:23 and more dynamics
55:28 on the hip-hop
55:31 uh versus
55:37 [Music]
55:41 the user wants a more pronounced thing.
55:44 It's doing it now.
55:46 This producer uh thing is so well done.
55:51 It is so well done. More cowbell. Did I
55:55 I Oh, I was gonna put more cowbell. Damn
55:57 it. I can't believe I didn't remember to
56:00 put it in. It was in my head. Joker.
56:02 Great minds think alike.
56:05 Was it Joker that said that? Oh, quick
56:07 click flick Paul. I thought it was
56:08 Joker. Oh, Joker, too.
56:15 The Royal Cipher Reququum.
56:19 [Music]
56:39 I step inside the ring with flux and
56:41 flow. My bars are rates of change. Watch
56:43 moments grow. From fluent streams to
56:45 slopes that kiss the curve. We chase the
56:47 instance ghost with fearless nerve. No
56:49 guesswork. Limits whisper tight and
56:50 true. The tangent's breath becomes a
56:53 breakthrough view. We saw the fall, the
56:55 flight, the twist, the spin. The beat
56:56 divides and laws of motion. When
57:01 time runs thin, he measured slip and
57:03 speed and caught the shape that hides
57:05 and motions need a fluent moves as
57:07 fluction marks the trace. A quiet
57:09 grammar of nature can't a face. He taint
57:11 the slope of lines and kiss a bend. Make
57:13 curving paths with simple symbols end to
57:16 sum the grain of space. He saw a scheme
57:18 where infinite decimals become a stream
57:20 with method set. He bridge the scene and
57:23 guess and lace the void with rules that
57:25 held the rest with bodies through forces
57:27 and start. He found the meter of the
57:29 moving heart.
57:32 Now pass the lights and watch exponent
57:36 wing each other fear the terms pro
57:42 like dancers's feet the pattern snaps
57:44 precise relentless beat stacks the cards
57:46 and deals an endless deck the house is
57:49 mad and new check
57:52 [Music]
58:01 >> you brought the that never cease to
58:03 climb. Unfolding truths by patientness
58:06 of time a circle.
58:08 >> I you know
58:11 I just I can't believe we live in a time
58:14 when you can just do this [ __ ] It's
58:17 amazing.
58:21 Oh man. All right. Um let's go back to
58:25 chat. Jeat
58:28 share this tab. Let's go back here. Is
58:31 this done? Looks like this is done. All
58:34 right, let's go here. So, we're Oh, no.
58:36 It's still going.
58:39 An unforgettable tribute to Sir Isaac
58:41 Newton from Apple to the Stars.
58:44 Celebrating the enduring legacy of a
58:46 scientific giant.
58:49 And it's still working.
58:51 We got nine slides coming.
58:55 They look great, though.
59:02 Test completed.
59:05 Preview.
59:08 Um, can I
59:11 can I go into presentation mode?
59:14 Edit in Google Slides. Minimize.
59:28 Oh, preview. Here we go.
59:33 The man who saw further. A glimpse into
59:35 Newton's genius.
59:39 Unveiling the laws of motion. The
59:41 foundations of dynamics.
59:44 So, this didn't do what we asked it to
59:45 do. The universal law of gravitation.
59:50 Unraveling secrets of optics. Invention
59:52 of calculus.
59:55 Oh, from space. Here's here's what it
59:57 did. From space travel to everyday life,
1:00:01 a world transformed by Newtonian
1:00:03 physics. Thank you, sir. All right. It
1:00:06 did kind of a a silly stupid version,
1:00:08 but look how beautiful those slides are.
1:00:11 Those are really good.
1:00:16 Sea slug of dunes. So much for the pop
1:00:18 music industry.
1:00:20 No, you know what we're going to end up
1:00:21 with, Sea Slug, is um
1:00:28 Sunno's already started this. You want
1:00:29 to go see something cool? Um let's go to
1:00:32 Sunno.
1:00:37 And then down the left hand side here,
1:00:40 oh, Sunno Studios coming. Join the
1:00:42 weight list. I wonder if I have access
1:00:44 to it.
1:00:45 Join weight list.
1:00:48 I think I already did this. Oh, skip to
1:00:51 wait list.
1:00:53 You're on the list.
1:00:55 You'll be one of the first in. Okay,
1:00:57 whatever. Great. Fantastic. Bob. Hey,
1:01:02 BB. B. So, over here there's this thing
1:01:05 called radio.
1:01:09 And then let's see. There's there's Dr.
1:01:12 Focus Beats. How do you change
1:01:17 Vote for the vibe.
1:01:19 104 listeners.
1:01:45 Well,
1:01:48 okay. I thought this was better than it
1:01:51 was. It's kind of shitty. So So you
1:01:54 basically just vote for the groove and
1:01:57 then they're they're generating songs
1:01:59 dynamically. Oh, got no audio over
1:02:01 there. Oh, really? Oh, because share
1:02:03 this tab instead. That's why. Um,
1:02:07 [Music]
1:02:17 so I guess you vote for styles and then
1:02:20 they make songs and and you do it.
1:02:23 But I can imagine so imagine there being
1:02:25 a Spotify where just people are
1:02:28 generating songs and they're putting
1:02:29 them out of Spotify tracks but it's it's
1:02:32 AI generated. This is what um Pika is
1:02:35 doing with Pika Social where it's just
1:02:37 like Tik Tok except everything is AI
1:02:40 generated.
1:02:41 There's going to be music, you know,
1:02:44 scenes that are that are like this. And
1:02:45 I here's what's going to happen. There
1:02:48 are going to be people that get really
1:02:50 good at making songs that people want to
1:02:52 listen to just using these tools. And
1:02:55 they're going to get known for that. And
1:02:56 then some are going to perform, some are
1:02:57 going to DJ, and some will just make
1:03:00 songs. And then maybe other bands pick
1:03:03 them up and imagine being in a cover
1:03:06 band that all you do is cover AI songs.
1:03:08 Why not?
1:03:10 If they become popular, why not?
1:03:14 Don't think about
1:03:17 Well, you can think about AI music
1:03:19 however you want. A possible way to
1:03:22 consider AI music is it's just
1:03:24 different.
1:03:28 Just like punk punk rock music is
1:03:31 different than classical music. It's not
1:03:32 better or worse. Classical musicians
1:03:35 would say it's way worse. It's just
1:03:37 noise. There's no craft.
1:03:40 It's just three chords and you know
1:03:43 three power chords per song.
1:03:48 But that's become an art form.
1:03:52 So imagine this as that.
1:03:56 Right now we've got on our
1:04:01 egos
1:04:03 and our blinders
1:04:06 to what it means to make music.
1:04:10 But you know digital audio, the the
1:04:13 ability to sample other songs and make
1:04:16 songs out of those 20 or 30 years ago,
1:04:19 that was not considered music. Hip-hop
1:04:21 was not considered music,
1:04:23 right? They're just sampling other
1:04:25 people's work. There's no craft there.
1:04:27 There's no talent. There's no skill.
1:04:31 And yet there is. And there will be
1:04:34 similar sort of crafts people who master
1:04:37 using these tools. So, and figures
1:04:40 figure out ways to rise above the noise.
1:04:42 What's up tonight? Sorry, I'm late. We
1:04:44 finally we can get started. Cam,
1:04:47 the minstrel model. Uh,
1:04:52 [Music]
1:04:53 I'm all over the place tonight.
1:04:56 Back to here.
1:05:08 Let's see. New chat.
1:05:11 Um,
1:05:15 I don't care about the legacy models
1:05:16 right now. Let's just do auto. I'm put
1:05:18 this in auto.
1:05:21 Oh, I know what I'm going to do. Let's
1:05:22 see.
1:05:25 Um,
1:05:39 so we're going to put this in auto.
1:05:41 We're going to put in our thing about
1:05:43 the the epic poem of Newton.
1:05:48 We'll see which model. Okay, it didn't.
1:05:50 It chose not to think this time.
1:05:54 You gather around you lords and ladies
1:05:56 bright.
1:05:58 Let mine's convene
1:06:09 section.
1:06:11 Yes, battle
1:06:13 verse
1:06:15 section.
1:06:36 All right. Now, I'm going to say, um,
1:06:41 give me
1:06:43 three visual
1:06:45 styles
1:06:48 for
1:06:50 illustrations
1:06:53 to go with this piece.
1:06:59 Renaissance graffiti fusion.
1:07:02 Celestial stage play.
1:07:05 Hiphop hieroglyphics. I like that.
1:07:09 Let's go with number three.
1:07:16 And damn it, with number three. And
1:07:22 I want you to
1:07:26 give me a list
1:07:30 of illustrations
1:07:35 to go with main ideas in the two poems.
1:07:44 the description of the image
1:07:49 and
1:07:55 in between
1:07:57 let's say and above and above
1:08:02 which
1:08:04 line of text.
1:08:20 Born in the plague year,
1:08:22 opening of first poem. A chunky bald
1:08:26 Newton is a swaddled infant.
1:08:31 Okay.
1:08:33 So, make these all wide images.
1:08:41 And
1:08:43 in what was it called? Hiphop.
1:08:48 What style was that?
1:08:54 Hip-hop hieroglyphics.
1:08:56 Hip hop hierro
1:09:01 glyphics
1:09:04 and make me image
1:09:08 number one.
1:09:19 Born in the plague
1:09:23 year.
1:09:25 All right. So, one of the things that I
1:09:27 found about GPT4
1:09:30 is that if you tried to do something
1:09:32 like this where you had a big long thing
1:09:34 and then you tried to make images for
1:09:35 it, it would get kind of confused and
1:09:37 lost. So, let's let's see how this thing
1:09:39 does with making images. A, does it give
1:09:42 us hip-hop hieroglyphics? I don't really
1:09:44 know what those look like, but there's a
1:09:46 description up above. And then, um,
1:09:52 I should be able to just say now number
1:09:54 two. Number three, number four,
1:09:57 or or even just say, "Now make the rest
1:09:59 of them, and it should just make them
1:10:00 all in a row.
1:10:02 The world has officially changed."
1:10:08 What's blowing your mind, Source Camp?
1:10:29 just everything that's now possible.
1:10:31 Yeah, it it is it's it's
1:10:36 th this is the thing that I am most um
1:10:45 that I'm most um Hang on. Itchy eyes.
1:10:49 Ah, that can't be good for them.
1:10:53 Um,
1:11:03 I have like, you know, radical ADHD,
1:11:09 right?
1:11:10 And
1:11:12 I'm really good at thinking horizontally
1:11:16 and thinking outside the box and being
1:11:18 able to jump from this idea to that
1:11:20 idea.
1:11:23 I'm finding it overwhelming.
1:11:29 Like
1:11:31 one of the things I talk about with AI
1:11:33 is just go play with it. And as you get
1:11:35 more
1:11:37 fasile with the different tools,
1:11:40 there are so many possibilities. Like I
1:11:43 don't quite know how to focus right now.
1:11:46 I don't I don't quite know how to do it.
1:11:47 And I
1:11:49 I mean I think it comes down to use
1:11:52 cases like pick a use case. That's a
1:11:54 very cool image. So let's say now make
1:12:00 images
1:12:02 two to five.
1:12:05 And let's just see if it if it does
1:12:06 them.
1:12:08 And let's see if it maintains that
1:12:10 style.
1:12:13 I know that eye rubbing feeling.
1:12:15 Struggle is documentation. I finally
1:12:17 have a tool that moves as fast as my
1:12:19 brain. Yep. There are so many
1:12:21 directions. I feel the same way. You're
1:12:23 not alone in the overwhelm, Kyle.
1:12:26 It it's it's overwhelming. And
1:12:31 like the whole purpose of this channel
1:12:34 is for us to be able to just play and
1:12:36 like not have any expectation to it. But
1:12:39 it's like
1:12:41 I have I have yet to play with like
1:12:44 Midjourney introduced HD video. Like we
1:12:47 played a couple of nights with H with
1:12:49 Midjourney video and then they've since
1:12:52 now come up with HD. I think it's 4K
1:12:55 video or some some crazy ass thing like
1:12:57 that. Maybe it's just HD 1080p. I don't
1:12:59 know what it is. But um
1:13:02 like I haven't even played with that
1:13:04 yet. Like there's just so much
1:13:07 like I feel like I'm
1:13:09 I'm sort of short,
1:13:12 you know, given the short straw to chat
1:13:14 GPT here because I'm I'm just
1:13:17 It's 9:15 and we've done like two
1:13:19 things. It's just amazing. Just
1:13:23 absolutely amazing. I keep asking myself
1:13:25 what new thing to try now and write
1:13:28 about. I know with AI's advancement, I'm
1:13:31 more surprised when I can't do something
1:13:33 instead of what it can. Yeah, that's
1:13:35 fascinating.
1:13:37 Thanks for the reminder. And midjourney
1:13:38 renewed today. There you go.
1:13:42 Um,
1:13:44 traits override personality. Casual
1:13:46 traits plus robot equals casual casual
1:13:49 robot. Blank traits equals pure
1:13:52 personality. Huh?
1:14:00 What were you talking about, Jeff?
1:14:10 All right. Same style.
1:14:17 So, this is kind of cool. Let's see if
1:14:19 it when it finishes this, we'll see if
1:14:20 it goes on to the next image.
1:14:30 It did not.
1:14:35 Can you make
1:14:38 images
1:14:40 3 to 10
1:14:42 without me having
1:14:45 to ask?
1:15:02 Yeah. So, Gareth, we changed colors. How
1:15:06 we change colors is you go to settings
1:15:09 and then under general, you can just
1:15:10 change your setting colors.
1:15:14 Let's see how pink plays.
1:15:19 That's hard to read on screen. I'm
1:15:22 guessing purple would be easier to read
1:15:24 on screen with white, right?
1:15:31 Yeah, that's good. I like purple. All
1:15:34 right.
1:15:36 I spend so much money on maintaining the
1:15:39 potential to make a random thought come
1:15:42 to life. Very little time actually doing
1:15:45 so. I feel the exact same way. In fact,
1:15:49 I am very much due for an audit of my
1:15:52 subscriptions.
1:15:54 Um, I've just got to go in and and and
1:15:57 call them because it's like
1:16:02 I feel like between
1:16:06 one good music generator midjourney.
1:16:12 So, so I would say Suno or Producer AI
1:16:16 midjourney chat GPT of course
1:16:22 um
1:16:26 probably an advanced video tool like VO
1:16:29 or Lumalabs or
1:16:33 um
1:16:35 one of those
1:16:42 maybe a vibe coding tool.
1:16:46 Like I I feel like if you just picked
1:16:48 those five categories, a single tool in
1:16:51 each one,
1:16:53 I I feel like you could probably spend
1:16:56 the next three months getting your head
1:16:58 around it,
1:17:01 you know?
1:17:03 So the answer is no. It cannot just make
1:17:07 images all in a row. Does anyone know
1:17:08 how to prompt chat GBT so I don't have
1:17:11 to keep doing it? Now number four,
1:17:15 [ __ ] loser.
1:17:17 The color of royalty. That's me.
1:17:21 Hi, my name's Kyle Shannon.
1:17:24 They call me the prince.
1:17:27 Not because of my musical talent. No,
1:17:29 no.
1:17:31 It's because I'm royalty
1:17:34 in my mind.
1:17:38 Um, yeah. Hola friends. Hey, rock paper
1:17:40 scissors. What's happening?
1:17:44 What is happening? What's going down?
1:17:46 What's shaking people? What's shaking?
1:17:48 Good people.
1:17:53 Oh, Gareth just sent a Garland
1:17:55 headpiece. Thank you very much for that.
1:17:57 I appreciate that.
1:18:00 Although I that probably showed up over
1:18:02 here.
1:18:04 I get sleepy people.
1:18:10 I spend so much money maintaining.
1:18:12 There's no sleeping in AI. Seriously,
1:18:17 I remember in the early days when I was
1:18:19 sort of six months into this, I remember
1:18:22 on multiple nights waking up at like
1:18:24 3:30 or 4 in the morning with an idea.
1:18:28 Oh, I wonder if I can do that. And I
1:18:31 jump onto my computer.
1:18:33 Holy [ __ ] I can do that. I'm gonna go
1:18:36 back to bed.
1:18:40 Isn't working for me. What are we up to
1:18:42 tonight? My two-year-old grandson wore
1:18:44 me out today. Um, tonight. So, here's
1:18:47 what I would encourage all of us to do
1:18:51 probably for the next two weeks
1:18:52 straight.
1:18:54 Just play and chat GPT. Switch models.
1:18:57 just play in GPT5. You you can go to
1:18:59 your old legacy models if you want, but
1:19:03 I feel like just go try to do [ __ ] in
1:19:07 chat GPT5
1:19:09 fast thinking mini thinking like there's
1:19:12 three main models there. Um, put on auto
1:19:15 and and figure out when you kick it into
1:19:18 when does it automatically kick into
1:19:20 thinking?
1:19:23 What kinds of prompts? What's it good
1:19:25 at? What's it not good at?
1:19:27 Um,
1:19:31 all seems like um,
1:19:34 time well spent.
1:19:38 The way I'm treating chat GPT5 is as if
1:19:41 I've never used a large language model
1:19:43 before
1:19:46 because effectively we're starting over.
1:19:50 Um, number five now.
1:19:55 Oops. What just happened?
1:19:58 What did I do?
1:20:00 That was weird.
1:20:03 Number five now.
1:20:07 All right.
1:20:11 I let chat GBT would listen to Red Note
1:20:14 in Chinese and then it translate back
1:20:17 the entire thing. Oh, that's
1:20:18 interesting. That's very cool.
1:20:27 That might be a fun thing to do. Let me
1:20:29 go find something here.
1:20:33 Chinese
1:20:37 proverb
1:20:42 art
1:20:45 images.
1:20:48 Oh no.
1:20:52 ancient
1:20:56 Chinese text
1:20:59 art.
1:21:00 There we go.
1:22:04 Oops.
1:22:13 Okay, we're gonna we're going to do
1:22:15 something here, I think.
1:22:35 Okay,
1:22:37 I'm going to paste in. So,
1:22:46 okay, I'm gonna say
1:22:51 translate
1:22:54 these eight poems
1:23:00 and
1:23:02 describe
1:23:04 anything
1:23:06 interesting
1:23:08 about
1:23:11 them
1:23:17 set of a Chinese poem written in
1:23:19 expressive cursive script
1:23:22 koshu
1:23:24 likely from the late Ming or early King
1:23:26 period king Quing
1:23:28 given the brushwork style and seal forms
1:23:31 they appear to be short self-contained
1:23:33 verses the mountain path grows still
1:23:36 with falling snow Okay,
1:23:39 great. This is what I was hoping.
1:23:48 Okay, I want you
1:23:52 to make
1:23:54 a piece of ancient
1:23:59 Chinese
1:24:01 style art,
1:24:03 but with the script
1:24:07 in English,
1:24:10 but looks like a Chinese
1:24:16 calligrapher
1:24:19 wrote it.
1:24:22 Be a beautiful and fluid.
1:24:27 also add
1:24:30 Chinese style
1:24:35 illustrations
1:24:37 as if
1:24:42 these were
1:24:45 art to be sold.
1:24:48 I want the poem
1:24:51 and illustrations
1:24:55 on a single
1:24:57 piece
1:25:00 of
1:25:02 parchment style
1:25:05 paper.
1:25:09 Make the first
1:25:13 wide image now. All right, let's see if
1:25:18 this can do this. So, this could be a
1:25:20 this is kind of a fun fun little
1:25:23 experiment, right? Take these Chinese
1:25:26 documents,
1:25:27 translate the poems. These sold at
1:25:29 Christies for some ungodly amount of
1:25:31 money.
1:25:34 GVT is now linked to your Google
1:25:36 calendar and Gmail. It can set
1:25:38 reminders. That scares the [ __ ] out of
1:25:40 me. I haven't tried that yet, but I
1:25:42 guess I should.
1:25:48 How do we know this is correct? Do we
1:25:50 have a translator? No. But I mean, for
1:25:53 the most part, Gareth,
1:25:55 um, these large language models are
1:25:57 remarkably good at translation. Like
1:26:00 remarkably good at it. Um, might it miss
1:26:05 nuance and tone and cultural things? It
1:26:08 might, but it's it's pretty good.
1:26:12 Um, I used to ask Google Calendar where
1:26:14 my meeting was on my way to whatever.
1:26:18 There's a cool new font that shows upper
1:26:20 and lowercase somethings.
1:26:25 Tab check.
1:26:29 Let's see what we get here.
1:26:35 Yeah. No, I was just reading comments on
1:26:37 Tik Tok.
1:26:41 It's great at translating English to
1:26:43 Greek. My son asked me when he was
1:26:46 little, "Do Chinese people in China eat
1:26:49 Chinese food?"
1:26:52 Just like we eat American food. John
1:26:55 Harris, what's up, Mr. K Shannon? What's
1:26:57 happening? John Harris, you know, just
1:27:00 hanging out. Just hanging out in the AI
1:27:03 learning lab
1:27:05 doing a little I hate that. I hate that
1:27:07 style of font. That looks like What's
1:27:09 that? What's that font from the original
1:27:14 Macintosh team? It looks like that. It
1:27:16 looks I don't like it.
1:27:18 Um,
1:27:21 but I suppose the mountain path grows
1:27:24 still with falling snow. Traveler's
1:27:26 footsteps fade. And look, the
1:27:28 illustrations are looking like Chinese
1:27:30 illustrations. This is cool.
1:27:32 No, it's not Ariel. It's um I think it
1:27:36 begins with an L. Look at that. That
1:27:38 looks gorgeous.
1:27:41 So, I'm going to go
1:27:44 um make the text more free form,
1:27:52 more handd
1:27:55 drawn looking
1:27:58 and less like a font
1:28:03 distress.
1:28:07 the paper more
1:28:11 and add
1:28:14 some splashes
1:28:17 of color into
1:28:20 the drawings.
1:28:25 Looking good.
1:28:27 All right, we paid chat GPT a compliment
1:28:29 so it doesn't chop our head off as soon
1:28:31 as it gets arms.
1:28:38 [Music]
1:28:40 He says, "American Chinese food is
1:28:42 American food." It absolutely is.
1:28:44 There's a really interesting documentary
1:28:46 on chowine.
1:28:49 Chowne was like a It was like an
1:28:52 invented food. I think it was invented
1:28:54 in the Cat Skills or Vegas. I think it
1:28:57 was in the Cat Skills,
1:29:02 but it was some Chinese restaurant and
1:29:03 they basically just made this thing up
1:29:05 and then it it went
1:29:07 across the country like across Route 66.
1:29:11 I forget the name of that thing. Lucida.
1:29:15 Yeah. Lucida Grand. I think that's it. I
1:29:18 think that's the font. Yeah, exactly.
1:29:20 That's exactly what it looked like. It
1:29:22 was horrible. Chopsui. That's it.
1:29:26 Chopsui. Yeah, chopsui didn't exist.
1:29:29 Chopsui is a madeup American dish and
1:29:32 there's a documentary on it. Um, I'm
1:29:35 sure I'm sure. Ask here. We'll we'll ask
1:29:37 Quinn. See if she knows.
1:29:55 Come on, Quinn. Where are you?
1:30:08 New chat.
1:30:15 Connection failed.
1:30:17 Tap to retry.
1:30:22 Huh?
1:30:46 isn't there a documentary about the
1:30:48 invention of chopsie in America?
1:30:56 Searching the web,
1:30:59 you're absolutely right. Several audio
1:31:01 visual works touch on the story.
1:31:05 Chapsuyu invented it.
1:31:08 Uh oh. What did I do?
1:31:11 What did I do? Oh,
1:31:22 Chopsui Nation Canadian Center. YouTube
1:31:26 short. YouTube short. Maybe I saw a
1:31:28 YouTube video. That would be amazing if
1:31:30 it was a YouTube video. Can't tell the
1:31:32 difference between
1:31:34 YouTube videos and
1:31:37 actual documentaries. There you go.
1:31:39 Climbing high, I see the peaks. Is this
1:31:41 the same poem? No.
1:31:48 The mountain path grows still with
1:31:50 falling snow.
1:31:53 God, I Okay, so chat GPT still sucks at
1:32:01 anything with context longer than 38
1:32:04 seconds.
1:32:06 [ __ ] Um,
1:32:09 you just made
1:32:12 crap up.
1:32:17 Here's the poem
1:32:20 and image description.
1:32:26 [Laughter]
1:32:31 Got to use U-shaped prompting. Really?
1:32:35 This is nuts. It's nuts, I tell you. Oh
1:32:38 crap, my phone's about to die. All you
1:32:41 people be like, "Where'd Kyle go? Why
1:32:43 did he just leave us?
1:32:48 Hey, cow. Cow. Cow. Cow. Hey, cow.
1:32:51 Where'd you go? Where'd you go, cow?
1:32:57 Thanks to Danielle. I made my first PA
1:32:59 animation. Lovely. Thank you, Danielle,
1:33:02 for being so generous in the community
1:33:03 and getting us uh doing doing the good
1:33:06 work, getting us uh invites.
1:33:13 [Music]
1:33:21 I need my Kyle fix. Don't leave me. It's
1:33:23 recorded. You can go back there. There
1:33:26 is about 1,400 hours of AI learning lab
1:33:30 on YouTube. There's you can watch the
1:33:32 history of of chat GPT
1:33:35 on YouTube just in this channel
1:33:39 because I did starting in the end of
1:33:43 April 2023
1:33:45 is when I started doing these lives and
1:33:47 for one year I did seven nights a week
1:33:50 without missing a night. So if you want
1:33:53 to see if you want to see the early days
1:33:55 of chat GPT, it was like right around
1:33:58 the time when GPT4 came out. When it
1:34:00 first came out
1:34:02 and China, I was there. A lot of you
1:34:04 were there.
1:34:06 Crazy crazy crazy times.
1:34:09 All right. Creating image. This may take
1:34:10 a moment. This is good. I don't get an
1:34:13 opportunity to send send you down a
1:34:15 rabbit hole. You might. This is chat add
1:34:19 after all.
1:34:22 Oh, this is looking better. The mountain
1:34:23 path.
1:34:25 Yeah, this one's good.
1:34:36 I heard people say, "I made AI watch a
1:34:38 thousand hours of X."
1:34:40 Is that just bluster? Are they doing
1:34:42 something locally? That's a lot of
1:34:43 tokens.
1:34:45 Uh, I I'm sure they're not. I I'm sure
1:34:47 what they're doing, Brandon, is um
1:34:51 a a thousand hours of transcripts of X.
1:34:55 I don't think they're doing a thousand
1:34:56 hours of multimodal video. The mountain
1:35:00 path grows still with falling snow.
1:35:02 Traveler's footsteps fade on mossy
1:35:04 stones beyond old temples shadowed gate.
1:35:08 Pines breathe the scent of cold clouds.
1:35:13 You're right. That one's on me.
1:35:22 Make poem
1:35:27 number two like this.
1:35:31 Make sure
1:35:34 you use the poem.
1:35:36 All right. Tik Tok pin.
1:35:40 I was there telling you I was
1:35:42 intimidated to try it because I'm older.
1:35:44 You replied, "Get over it."
1:35:49 That's right. Becky Rue came in. She
1:35:50 goes, "I'm just a frail, old lady." And
1:35:54 I'm like, "Ah, shut up. Just go play
1:35:56 with the eye." And then before you knew
1:35:58 it, she was making custom GPTs of uh
1:36:03 like quilt making GPTs and kicking ass.
1:36:09 This is This is a channel of suck it up,
1:36:11 buttercup. It totally is.
1:36:16 You wouldn't add files. You would
1:36:19 connect to YouTube.
1:36:22 Yeah. But still, I wouldn't, unless I
1:36:24 was analyzing a video for the visuals,
1:36:28 I wouldn't use the multimodality.
1:36:30 I would just, you know, have it analyze
1:36:33 the transcripts.
1:36:38 You know, like maybe you want to find
1:36:41 all the times I held up the golden
1:36:43 button.
1:36:43 >> You can make money with Jackie.
1:36:46 >> Yeah, Becky. I was AI before AI was
1:36:48 cool. You were the salon.ai. I spelled
1:36:51 it right. Yes, you did. The salon.ai.
1:36:54 So, if you go to if you go to um
1:36:58 community.thesalon.ai
1:37:00 AI.
1:37:02 Um, that'll take you straight to the the
1:37:04 mighty networks thing and just sign up
1:37:06 there and introduce yourself. There's a
1:37:08 section to introduce yourself.
1:37:12 Um, we're launching a new website
1:37:15 relatively soon uh probably in the next
1:37:17 two or three weeks uh at the AI salon.
1:37:21 It's going to be really nice. So, that's
1:37:23 happening.
1:37:25 that be happening.
1:37:29 Everyone talking about jailbreaking chat
1:37:31 GPT5 in the AI hacking community. Cool.
1:37:34 Let them do that.
1:37:37 Waters clear to riverstones below. Let's
1:37:40 see if it actually paid attention
1:37:43 and got
1:37:45 No. God damn it. Unbelievable.
1:37:51 I like this purple. I like this purple.
1:37:54 Uh, highlight. I like that you can
1:37:56 change the highlights now. Um, nope.
1:38:01 You just made
1:38:04 crap up again.
1:38:08 How
1:38:12 much do you suck?
1:38:16 Here's poem two.
1:38:23 Stop.
1:38:34 [Music]
1:38:39 This is pretty cool. This is This is an
1:38:40 old school thing we did tonight, right?
1:38:43 Go get an image from the internet that's
1:38:46 in some crazy ass ancient Chinese
1:38:49 script. Translate it
1:38:52 and then have it make you little
1:38:56 modernized versions.
1:38:59 Modernize Americanized versions.
1:39:02 Pretty slick.
1:39:14 [Music]
1:39:20 Falling flowers, purity,
1:39:24 and an auditory image, a bell, balancing
1:39:26 the visual.
1:39:28 All right. One of my top 10 use cases
1:39:30 for Genai is giving Google Gemini Pro
1:39:33 2.5 a YouTube link and having it do
1:39:36 something.
1:39:40 The translation says, "We will own you
1:39:43 someday with technology."
1:39:46 Chinese version of Robert Frost.
1:39:50 These these are more like the Chinese
1:39:51 version of uh what's his name? The guy
1:39:54 that paints with light. Uh
1:39:58 oh, Christ. Concincaid.
1:40:00 Is that his name?
1:40:04 No. Not Bob Ross. Not Bob Ross. The one
1:40:08 that's got like mall galleries.
1:40:14 Gonna see Mercer Labs tomorrow. That's
1:40:17 cool.
1:40:24 Thomas Concincaid. Yeah, that's him.
1:40:27 This like the Chinese version of Thomas
1:40:29 Concincaid paintings.
1:40:35 Flowers fall soundless on the floating
1:40:37 stream. Yeah. Okay. This is this is the
1:40:40 actual one. Good.
1:40:43 Yeah. Thomas Concincaid. Would we be
1:40:46 able to use machine learning to figure
1:40:47 out old things like hieroglyphics? Yes.
1:40:50 In fact, Ryan, they are doing that.
1:40:52 They're doing that in in Egypt right now
1:40:54 because they've got technology that they
1:40:56 can X-ray scrolls. They can virtually
1:40:59 unroll scrolls that are too bit bitter
1:41:02 brittle to actually unroll
1:41:05 and they've got they've got things that
1:41:07 were previously untransatable that
1:41:09 they've they've now translated with AI.
1:41:11 So yeah, that's going to absolutely
1:41:13 happen.
1:41:14 Absolutely happen.
1:41:16 Flowers fall soundless into the flowing
1:41:19 stream. The monk's robe brushes past in
1:41:23 the morning mist from the valley's
1:41:25 depth. A single bell answers the
1:41:27 mountains endless green. Ping.
1:41:34 All right. Well,
1:41:37 I don't know if we learned anything new
1:41:39 tonight. It still sucks at continuity
1:41:42 uh when switching between text and image
1:41:44 generation. Um that's that's not been
1:41:48 improved.
1:41:50 Um apparently one of the things that has
1:41:52 been improved is it hallucinates less.
1:41:54 Um, I didn't do anything that would
1:41:56 really test that tonight. Maybe we'll
1:41:57 play with that tomorrow.
1:41:59 It's off a bit. Why am I in the Ozarks?
1:42:06 Oh man, I still like Happy Birds.
1:42:12 Yeah, it this this might be off a lot. I
1:42:15 don't know if this is off a little or a
1:42:16 lot if you compare it to, you know, here
1:42:20 are the original images.
1:42:22 So, if you speak ancient Chinese,
1:42:26 there you go.
1:42:28 Tell me if I got it right. Um, all
1:42:31 right.
1:42:32 Let's do this. Do that.
1:42:35 [Music]
1:42:37 I'm going to call it. The voice is
1:42:39 going. I don't know why I was cranky
1:42:41 tonight. I was cranky tonight.
1:42:44 Um, if you wanted to play with a new
1:42:47 tool, which would be the best? ChatgBT5.
1:42:50 Like right now,
1:42:52 everyone
1:42:54 is starting fresh with ChatgPT5.
1:42:57 No one knows how to use it.
1:43:01 I mean, you kind of do, but what we
1:43:04 don't know. So, here's the thing that I
1:43:06 was thinking about today.
1:43:08 Chat GPT40
1:43:10 hallucinated a lot. Chat GPT5, they
1:43:14 built in a thing that reduces
1:43:15 hallucination. So basically before it
1:43:18 actually gives you an answer, it looks
1:43:19 at the answer and goes, is this
1:43:21 [ __ ] And then if it is, it rewrites
1:43:24 it and it gives you the new answer. And
1:43:26 you don't know that that's happening.
1:43:27 That's all going on behind the scenes.
1:43:29 And then and probably lots of other
1:43:31 subtle little features like that. What
1:43:34 struck me today is
1:43:36 for a year and a half we've been working
1:43:38 with increasingly improved versions of
1:43:41 the same base model.
1:43:44 So whatever it did bad, it's generally
1:43:48 did bad at a structural level, right?
1:43:53 Chat GPT5 is a new model that probably
1:43:57 does some things better, some things
1:43:58 worse. We don't know yet. So there might
1:44:01 be ways that we prompted chat GPT
1:44:05 that you don't need to prompt it that
1:44:06 way anymore because the new model's
1:44:08 better. And in fact, the new model might
1:44:12 act worse if you give it one of the old
1:44:14 prompts. How do we learn that? There's
1:44:16 no way to learn that other than playing
1:44:18 and other than discovering, oh, if I
1:44:21 actually make my prompt more simple, it
1:44:23 got better. Or, you know, I have to give
1:44:26 it a much more structured prompt. I saw
1:44:28 that on Twitter today that someone said
1:44:30 chatbt5 really likes structure. Now that
1:44:33 might have been from someone who is
1:44:35 selling structured prompts on the
1:44:37 internet
1:44:40 but but I don't know. So I I think for
1:44:43 the next month
1:44:45 just playing with chat GPT there's nine
1:44:48 different models you can choose from.
1:44:50 They brought back the model picker.
1:44:53 There's now personalities in this in the
1:44:56 customization setup.
1:44:58 There's there's so many things you can
1:45:00 mess with with it. So, I would just just
1:45:02 play with it
1:45:04 and dig deep and just keep pushing and
1:45:07 pushing and pushing and pushing. Good to
1:45:09 know. What?
1:45:11 Yeah. You haven't seen the You haven't
1:45:13 seen they they brought back all the
1:45:14 models. Look, you now have auto, fast,
1:45:18 thinking MIDI and thinking and then if
1:45:20 you pay 200 bucks a month pro and then
1:45:22 you have legacy models 40,41 03 and 04
1:45:26 mini. So an a multimodal model, a non
1:45:30 multi multimodal 4.1.
1:45:33 Who was using that? I don't know. Why'
1:45:36 they bring it back? I don't know.
1:45:39 03 was the the big macdaddy reasoning
1:45:41 model previously. 04 Mini was in between
1:45:44 03 and what what we're using now.
1:45:48 Crazy
1:45:50 chatb5 is going to show me a new way to
1:45:52 use Notebook LM.
1:45:55 Where's the singing dog model? Exactly.
1:45:59 I thought I lost the model picker
1:46:01 forever.
1:46:02 You did. And then everyone bitched.
1:46:06 [ __ ] hypocrites. All these AI
1:46:09 pundits. AI is the best. Get rid of the
1:46:12 model picker. We hate the model picker.
1:46:14 The day they get rid of the model
1:46:16 picker, everyone's like, "Oh no, where
1:46:17 are my models? You can't take my models
1:46:21 away and then just pick a random model
1:46:24 for me. I want to know what's going on.
1:46:26 I want to know what model you're using.
1:46:28 And so Sam Alman was like, "Fuck it.
1:46:31 Here you go."
1:46:34 Here we renamed them fast thinking
1:46:36 minion thinking and pro. That generally
1:46:40 means what they do.
1:46:44 So So humans humans [ __ ] that one up.
1:46:51 But you know that's what we do.
1:46:57 All right. Oh, I wasn't sharing any of
1:47:00 that, but whatever. All right, everyone.
1:47:02 Uh, baffle them with [ __ ] Exactly.
1:47:06 Um, all right, everyone. Peace out. I
1:47:08 will see you tomorrow. Tomorrow's
1:47:10 Thursday. Same bat time, 8:00 PM, right
1:47:12 here. Come
1:47:14 do what you do. All right.
1:47:17 Hey, tell your friends about this. Let's
1:47:19 let's grow this this