1/27/2025 - AI Reasoning Engines: Deep Dive & Comparison of DeepSeek, Gemini, and GPT-o1

Description

Chapters

Transcript

0:04 but it's
0:05 true all right
0:23 [Music]
0:37 [Music]
0:42 [Music]
0:55 is a place I can rest my head
0:59 [Music]
1:16 I'm singing like Timothy sham on
1:17 Saturday Night
1:20 [Laughter]
1:23 [Music]
1:25 Live yeah
1:39 [Music]
1:56 [Music]
1:58 yeah should FL
2:01 [Music]
2:09 [Music]
2:30 good the good
2:31 people you give this AI thing three days
2:35 and someone comes along with a little
2:37 model that just blows the whole damn
2:38 world up welcome to
2:41 AI it's gonna be fantastic tonight Bob
2:43 tell him what he's
2:44 won oh hasn't still wonder thing this is
2:49 not a game show right carrying
2:56 on um how is everyone tonight my name is
3:00 Kyle Shannon this is the AI learning lab
3:02 we got some people filtering in uh
3:05 they're watching me sing and they're
3:07 filtering right back out
3:15 [Music]
3:27 [Music]
3:30 every time I see you now get that look
3:34 in
3:36 my every time see your mouth I hear that
3:42 smile New York state was a rolling
3:46 Breeze and a sunshine with a blue sky
3:49 falling J of old September
3:54 [Music]
3:57 prep you
4:01 again
4:04 today you convince
4:07 me again
4:10 today you leaving this hotel looking for
4:14 someone else's cold and
4:19 Rain
4:21 say so long
4:25 [Music]
4:27 suzan hush now don't you cry
4:32 [Music]
4:36 so long
4:38 [Music]
4:44 suzan don't you cry for me
4:47 [Music]
4:58 [Music]
5:02 that is Champ Shannon on the
5:05 vocals I got uh got Timmy back here on
5:08 the
5:11 drums Jasper on the bass my name's Kyle
5:15 Shannon I'm on the lead guitar and
5:18 slightly flat vocals
5:21 [Music]
5:48 yeah deep sea kind of blew up over the
5:50 weekend that's an understatement what's
5:51 what's amazing to me
5:54 is like one of the ways that I kind of
5:58 follow the news
6:00 is you know someone will launch
6:02 something and you look at it and you're
6:04 like oh okay all right I all and then
6:08 you'll see some chatter about it
6:10 BL all right all right and then with
6:14 deep seek it was like they launched it
6:17 and then there was chatter about it and
6:19 then the chatter was
6:21 like the big research company should be
6:24 scared and then the chatter was like oh
6:28 the big researchers are in panic mode I
6:31 heard it about meta
6:34 first and I'm like they can't be in
6:36 panic mode it's just another [ __ ]
6:39 [Music]
6:42 model but it's how they it's how it's
6:45 How It Was produced assuming we can
6:48 believe it was produced how they say it
6:50 was
6:50 produced which that I don't know if we
6:55 can but they apparently produced it with
6:58 248 gpus two months so about $6 Million
7:02 worth of worth of
7:07 training as opposed to you know hundreds
7:09 of millions billions of dollars from the
7:11 big the big companies and they're and
7:13 they're kicking their butts they also um
7:15 today so it was like five days ago deep
7:20 seek launched
7:21 V1 three days ago four days ago it was
7:24 R1 the reasoning engine today they
7:27 launched a multimodal engine I hear that
7:30 they've got a voice engine
7:32 coming and then they've got apparently
7:35 and this could just be rumor but rumor
7:37 is chat TMZ style that they've also got
7:41 a
7:43 uh like operator like a a web browsing
7:46 agent so appar apparently they've got
7:49 they've got all of that coming
7:52 so yeah Nvidia stock tanked like 20%
7:55 today
8:02 it's fun time being
8:05 [Music]
8:09 a how how are we accessing deep seek so
8:13 we're gonna play with that tonight let
8:14 me let me switch this because it's hard
8:15 to read the letters and I got to put up
8:17 my Black Box here for the people for the
8:20 people there you go people um if you go
8:24 to chat. deep
8:27 seek.com um it's just like chat GPT it's
8:29 free
8:31 you can also if you want to have your
8:32 mind [ __ ] blown oh that's another
8:34 thing I should put up here hang on a sec
8:37 I'll put up the mind blowing
8:39 option go to
8:42 gq.com
8:44 gro.com
8:52 and you can flip that bad boy into deep
8:56 seek R1 if you want to
8:58 see if you want to see reasoning a
9:01 reasoning engine at the speed of
9:04 thought uh go to gq.com and choose the
9:08 R1 reasoning
9:11 model all right fantastic Bob is it a
9:15 Chinese data collection app for the
9:18 spying I don't know it might
9:21 be as opposed to what the you you want
9:25 our spying apps
9:27 homegrown you only want our spying apps
9:30 here on uh here in
9:34 [Music]
9:45 America I mean not for nothing Facebook
9:48 and Google
9:50 and all of them have been spying on us
9:53 for 20 years right like you know that
9:55 right like everyone knows that yes is it
9:59 worse because China's going to do
10:00 something with it maybe I don't know
10:03 maybe if we become part of the social
10:05 score of China at some point that would
10:07 impact us so so if you wna you want to
10:09 head your bets for the future now um is
10:13 it highly likely that if you go to chat.
10:16 deep seek.com you should assume that
10:19 nothing in there is
10:21 safe that that everything in there will
10:24 likely be um you know popped into a uh a
10:28 database sure you should assume that
10:29 that but you should also assume that
10:32 anything that's
10:33 [Music]
10:44 free can't sign up for deep seek been
10:47 trying for a long time today and it
10:48 keeps the gear keeps spinning that's
10:51 weird hang on I gotta move my Black
10:54 Box um I just logged into deep
10:57 seek that's Google studio that's
11:01 that I'm
11:03 in chat. deep
11:06 seek.com can't sign up yeah that's
11:11 [Music]
11:15 weird perplexity said deep seek
11:18 downloads were
11:21 paused deep seek had an attack today ah
11:25 so they are limiting new enrollments
11:27 okay thank you all right so luckily
11:31 daddy AI learning lab got in there just
11:33 in the nick of time let me see if it's
11:35 actually working right now um tell me a
11:40 bed time
11:43 story
11:48 story yeah it's working
11:52 okay by the way if you're worried about
11:56 Ye Old Chinese servers spying on your
11:58 [ __ ]
12:00 spin up a web-based GPU and install your
12:04 own version of uh deep seek R1 there's
12:07 all there's a bunch of different
12:09 sizes figure out how much you want to
12:11 spend and just throw it in a
12:13 hermetically sealed GPU instance in the
12:15 cloud and then none of your shit's going
12:17 anywhere except to the to the GPU
12:19 provider but generally they've got good
12:22 good things did this give us a good
12:25 nighttime
12:26 story the Allen the missing fireflies
12:29 you know cozy fire forests nestled
12:31 between Rolling Hills where the moon
12:33 painted silver Trails through the trees
12:36 there Liv lived a wife Wise Old Al named
12:38 Oliver seems like a story all right
12:40 there you go all right I all good people
12:43 I downloaded deep seek off the App Store
12:45 L than an hour ago works
12:47 fine okay yeah they've got an
12:55 app and it's
12:57 free and it's better than 01 or at least
13:01 01 preview probably maybe 01 I don't
13:04 know if it's better than 01 screen share
13:06 still off yeah that's okay I'm not I'm I
13:09 wasn't really sharing anything thank
13:13 you let me share so I
13:22 have carrying
13:25 on rather Cheerio pip pip
13:33 you know we don't need that right
13:35 now bless
13:37 you oh bless me I thought you literally
13:41 literally meant like bless me like you
13:43 were taking pity on me it was because I
13:47 sneezed I'm I'm a little slow in the
13:49 uptake sometimes I don't know if you
13:50 knew
13:51 that H bless your heart a a't that cute
13:55 he's on there trying to talk about the
13:57 AI with zero qualif ifications it's so
14:01 sweet oh I'll tell you what I'll tell
14:05 you what people I I'll tell you what I
14:08 do I go to that AI learning lab he he's
14:10 just so sweet he's got a dog the dog
14:12 bless his heart that little dog sing
14:15 sing sings now here's a funny thing and
14:17 it's kind of sad I mean I feel for the
14:19 gentleman um the dog sings better than
14:23 he does and it's just
14:26 it's bless his heart bless his heart and
14:30 then anyway but it's called the AI
14:32 learning lab so you go there and what
14:34 he's talking about on the AI learning
14:35 lab is it's not always AI right
14:38 sometimes
14:41 he
14:42 I and he it's learning he talks
14:46 about he's he's not a teacher he's not a
14:48 teacher he's not qualified he's he's
14:50 he's got zero qualifications um uh he
14:54 has a degree in acting and apparently
14:56 got it mail order a mail order acting
14:59 degree apparently um but he he tries he
15:02 he's so sweet he's super sweet he shows
15:04 up all the time and and it's like if if
15:06 you're going to give him an a I'll tell
15:08 you what AI learning lab they deserve a
15:11 participation trophy they really a
15:13 certificate a participation certificate
15:15 yeah trophy probably probably not
15:17 probably not if maybe if they made like
15:20 a little like you know like a little
15:22 Yoda trophy that would be fine but just
15:25 more more of
15:26 a you can make money with
15:30 so that's what we do
15:32 [Laughter]
15:35 here a fashion award for the bow oh good
15:41 lord um ocean
15:43 Kyle you know we have different
15:46 capabilities and interests when we use
15:49 Ai and GPT needs different ways however
15:53 how it's going to be able to cooperate
15:55 and stay with the human brain oh it's
15:58 not
16:00 ocean here I'll leave that up there
16:02 because in case you want to figure that
16:04 out one of the one of the one of the
16:08 questions one of the questions
16:12 uh the question here how are humans
16:16 going to keep up with these things we're
16:18 not we're
16:21 not
16:28 and what while that
16:30 is potentially caused for
16:35 [Music]
16:37 concern shy of all of this AI activity
16:40 shutting down which it's not going to
16:43 these AI capabilities I you know I
16:45 listen I would argue that for me
16:48 personally my cognitive
16:50 wall I hit when 01 preview came out back
16:55 in what was it
16:58 September um
17:00 and I realized that I didn't really
17:03 know how to um how to prompt it that I
17:08 didn't in fact the exercise that we're
17:11 going to do tonight we're going to do a
17:13 head-to-head between three reasoning
17:14 engines and the prompt that we're going
17:16 to
17:17 use we we'll write it live here so you
17:20 can see my prompting process which is
17:22 very methodical and it's it I just pull
17:24 it out of my butt basically um
17:31 [Music]
17:32 um but we're going to do we're going to
17:33 do a heads up
17:36 competition competition is not the right
17:38 word um because when you say competition
17:41 that implies there's winners and losers
17:43 and in this world we can have no winners
17:45 and
17:47 losers although tell that to the Buffalo
17:50 [Music]
17:58 Bills we're going to have a competition
18:00 between deep seek Gemini their reasoning
18:03 engine flash thinking I think they call
18:06 it and then uh
18:08 01 and maybe we'll throw 01 Pro in there
18:11 since I'm paying for it the whole point
18:13 of me paying all that money is to have
18:15 access to these better tools all right
18:18 so we're gonna do that and the uh The
18:20 Prompt that I'm going to put in there is
18:24 how would someone with
18:25 my
18:27 capabilities for lack of a better term
18:30 how would someone with my skill
18:32 set best leverage a reasoning engine and
18:35 let's let the reasoning engines tell us
18:37 how to use the reasoning engines and
18:38 then we'll compare the answers and just
18:39 see see see what the [ __ ] these things
18:41 are and how they work and why we should
18:43 give a
18:45 [ __ ] they're they're fascinating they're
18:47 fascinating I've I've been using them
18:48 more in the past week and again part of
18:50 the one of the things that I've been
18:53 talking about on the channel for the
18:55 past I don't know three weeks a
18:57 month is
18:59 and this gets back to the human the
19:01 human thing what's our
19:04 job if we have tools that are
19:07 significantly beyond our cognitive
19:11 ability what's our
19:14 job and I think our job is to a person
19:18 figure out what we want to do in the
19:21 world what difference we want to make in
19:22 the
19:23 world because we wait hang on did you
19:27 just call me yeah
19:34 [Music]
19:39 thought I was being called thought I was
19:41 being
19:43 summoned have you been able to prompt
19:45 operator to do anything
19:47 useful yeah I I mean it depends on your
19:50 definition of useful I haven't I I saw a
19:52 couple of videos of people um using
19:56 operator to go to um repet and generate
20:01 apps which I thought was kind of
20:02 interesting I saw a video today of
20:05 someone having it go to make and create
20:08 an automation which I thought was kind
20:10 of interesting um last night I did kind
20:14 of a deep research project that a uh a
20:18 thing came out for for um the University
20:22 of Colorado and shoots which is just
20:24 like you know 45 minutes that way they
20:28 just did some research Arch that they
20:30 have um they've got a new drug that
20:33 works in animal models to repair myin
20:37 sheath neuropathy which is what my boys
20:39 have um in the eyes of I don't know rats
20:43 that had whatever the disease was that
20:45 caused the myin sheath
20:47 blindness um and and it's made it's it's
20:50 uh promising to repair my and sheath
20:54 it's like the covering of your nerves um
20:57 and so I just did a deep Di on that so I
21:00 had I had um I had operator go out and
21:03 do that research and I it was
21:05 fascinating like just sitting there
21:07 watching it do the
21:08 research and what you realize and you'll
21:11 see this with deep seek tonight deep
21:12 seek is connected to the internet
21:16 like it's hard to put it's hard for my
21:20 tiny ape brain to comprehend but like it
21:25 will just go find a website like that
21:29 and then it goes to it and it's Reddit
21:31 it reads the whole website or it reads
21:33 the whole paper and in the case of deep
21:35 seek it's like okay I found 49 websites
21:38 and now I'm analyzing them and now out
21:40 comes your answer so so these things are
21:43 incredibly capable so so that was pretty
21:46 [Music]
21:49 interesting but good question I I think
21:53 I think operator is is going to force us
21:56 to think about well again what do we
21:58 want Toom
21:59 what do we want to do in the
22:01 world um you know solving some shitty
22:04 disease that your kids have or or
22:06 finding at least treatments for it
22:09 that's not
22:11 [Music]
22:16 bad so we're going to do that we're also
22:18 going to start with jelly pod if you
22:21 haven't been here if you weren't here
22:22 last
22:23 week we did this uh we did this wacky
22:27 zany
22:29 this wacky zany um podcast maker called
22:33 jelly pod I'm with Pate I don't like 01
22:36 or for 40 for Co coding for myself what
22:40 what do you use Joker do you use um uh
22:43 what you call it
22:45 Claude yeah I don't I kind of feel like
22:49 coding
22:51 um I've been coding for two weeks my
22:54 agent will do all my coding
23:00 all of the okay if you're new here if
23:02 you're new to AI apps and and these
23:05 things these AI these AI tools
23:09 today to a tool are janky pieces of [ __ ]
23:13 now they are getting better and better
23:15 and better and better but there's still
23:17 a
23:19 mess I'm in three different Creator
23:21 programs that I keep getting access to
23:24 their latest models and they still
23:27 cannot make a car that that's Wheels
23:29 roll like the car moves but the wheels
23:32 are stuck they're just
23:37 sliding so they're just everything's
23:40 janky including the coding tools
23:42 remember when IBM Watson was going to
23:44 revolutionize the world yeah but we're
23:47 in a we're in a different we're in a
23:48 very very different place here that was
23:51 you know very narrow Ai and it it was a
23:54 more traditional approach to machine
23:56 learning we now have generally capable
23:59 you know um general intelligence not not
24:02 AGI not artificial general intelligence
24:05 the threshold we haven't met the
24:06 threshold yet um but but these are
24:09 generally intelligent tools right
24:12 they're they're intelligent on lots of
24:15 different things and the reasoning
24:17 engines can reason um beyond their
24:21 trained capability right they have
24:23 emergent properties so that's very
24:27 different and that was also IBM and it
24:30 was also you know a single product a
24:33 single
24:34 division um this AI stuff is now you
24:38 know coming from every major tech
24:40 company in the world Plus open
24:43 [Music]
24:53 source um so what jelly pod is is this
24:57 is this allows us to we designed these
25:00 two voices Lily Vogue she's the one on
25:04 the right with the the bitching gray
25:06 streak in her hair and then Buster
25:10 mcgillicuty he's a little hipster dude
25:12 on the left he's stoned a lot um but
25:15 there are hosts and then we went over to
25:19 to Yan chat
25:23 GPT and we had chat GPT write us some
25:26 episodes so so this was kind of a cool
25:29 thing we we had it design the uh the
25:32 hosts we had it design the the podcast
25:37 ideas and then Corey Sandler went off
25:39 and made us episode covers so we're g to
25:41 make one of these so let
25:44 me go back to jelly
25:51 pod oh Brandon don't let me hit the
25:54 submit button before I add the music
25:57 because it's it's one of those counter
25:58 intive things where the submit button is
26:00 big and orange and then the ad music is
26:03 really tiny and gray and you can barely
26:05 see it um oh you want to hear one of
26:08 these by the way here's one on the um
26:11 the the uh episode one the premiere the
26:14 astonishing life of Mike the Headless
26:18 Chicken oh wait you can't hear that oh
26:21 God
26:22 [Laughter]
26:27 [Music]
26:38 the only podcast where absurd trivia and
26:41 fascinating history collide with well a
26:43 solid dose of charm wouldn't you say a
26:47 dose Lily I'd call it more of a heaping
26:50 pile two charm heavyweights here I'm
26:54 Buster your guide to all the weirdly
26:55 impossible stories you didn't know you
26:57 needed so this is our little host all
26:59 right so that's what that's what our our
27:00 podcast is going to be made out of so
27:02 we're going to go new episode are are
27:04 you viewing this yes you're viewing this
27:06 okay I've been using chat gbt over
27:08 Google and any okay all right so we
27:11 we're gonna go over I'm gonna go grab uh
27:14 in chat GPT the episode
27:25 description and we'll go add source and
27:28 then I'm go back over here and I'm going
27:29 to grab the
27:32 Wikipedia
27:34 article about this
27:38 topic all right we're g to add
27:43 that um or describe your episode uh I'm
27:47 going to say this is episode 4 and it
27:54 is sponsored
27:59 by the AI
28:02 salon at the
28:07 salon.
28:10 um do I need anything else other than
28:16 that oh I'm adding that as a source okay
28:20 fine copy that out of there all right so
28:24 create boom
28:30 so now what it does is it's going to
28:31 write us an outline which we're going to
28:33 be able to augment which is super
28:37 sweet got into deep seek on Firefox on
28:40 laptop okay great so yeah you might have
28:43 had some sort of cookie some sort of
28:46 Bonkers cookie issue on on
28:48 [Music]
28:49 Chrome Valerie Cox what do she have to
28:52 say while we're looking at that thing
28:54 I've been using chat GPT over Google
28:59 on any questions that I have or on any
29:02 explanation on a deep
29:04 subject I have set up several tasks to
29:07 get me top news tell me a joke produce
29:09 random pictures for fun yeah that's
29:11 pretty cool are you you're using tasks
29:12 for that all right that's
29:18 cool as an educator
29:21 in Mar Meridius Marius I cannot believe
29:26 how resistant they are to embrace in AI
29:30 which is being something I assume you
29:32 know they're they're using AI detectors
29:35 I it is staggering to me
29:39 how far up their asses most educational
29:42 institutions are right now with AI
29:44 you've got honest to God I mean just
29:46 think about what it is it's it's one of
29:49 the most
29:50 profound knowledge augmentations in the
29:53 history no it is the most profound
29:56 knowledge augmentation in the history of
29:58 humanity and our educational
30:00 institutions are be like well we don't
30:02 want people to
30:03 cheat it's going to [ __ ] change
30:05 everything it's going to change how our
30:07 brains
30:09 work it it it's absolutely mind-blowing
30:12 to me absolutely mind-blowing all right
30:16 what the fact image archives number four
30:18 sneaker culture oh wait you can't see
30:20 any of that because
30:23 tabs morius mortius is a country in East
30:27 Africa very cool
30:30 um I got to change this again hang on
30:41 people the college kids can't even get a
30:44 smart laptop per instruction per
30:47 instructions from their professors yeah
30:50 well some of them are going to figure it
30:52 out anyway here's the Corey Sandler
30:54 cover for what the what the fact from
30:57 Humble Footwear to the hype
30:59 machine so that's our
31:02 uh file too large all right we had this
31:05 problem before I've got to I gotta go
31:07 fix all these things probably before we
31:09 go live shouldn't
31:10 [Laughter]
31:15 I
31:20 yeah just say yes 768 by 768 resample
31:27 the image let's do do
31:29 it duplicate is Tiff sure fine whatever
31:43 [Music]
31:48 export all right let's see if we got it
31:52 here
31:55 desktop we're going to work looks like
31:57 we're going to work
32:00 all right we're going to add a chapter
32:02 here it puts
32:05 chapter the new chapter behind chapter
32:08 one which is weird chapter
32:11 title episode 4
32:16 intro and then I'm going to put the [ __ ]
32:18 I wrote before this is episode four and
32:20 sponsored by the AI
32:23 Salon I think that's all I need to put
32:25 there am I missing anything Brandon I
32:27 think that's it
32:31 sure fine
32:32 whatever my response to my computer all
32:35 the time exactly you have to have a
32:38 sense of humor with these [ __ ]
32:39 computers okay so now what we
32:43 do what we do I definitely know what to
32:46 do here yeah okay there it is so now
32:49 we're going to generate the script
32:52 bang and so it's now writing a script
32:56 for us
32:59 down here all this [ __ ] on the right
33:01 where there's nothing there's going to
33:02 be something so it's going to take our
33:04 little intro kicks the change the game
33:08 Lily dramatically introduces sneaker
33:10 cultures
33:12 Origins fun fact when I was doing
33:14 agency.com
33:17 we went out and we pitched Nike you know
33:20 the sneaker company they make they make
33:23 kicks as the kids call them
33:27 today oh yeah this is a free account
33:29 that I'm on although he did give me some
33:30 extra credits so uh I couldn't be doing
33:33 these on just a normal free account
33:34 actually I don't know how many free
33:35 credit credits he gave me so we might
33:38 not be able to to generate this we'll
33:40 see um
33:44 uh yeah we pitched
33:46 Nike and they are a madly impressive
33:51 company to pitch because they've
33:53 basically tried everything and the thing
33:55 that you think is radical and Innovative
33:57 and marketing they're like yeah we tried
33:59 that a decade ago that doesn't work um
34:01 they're a little full of themselves but
34:03 they're [ __ ] Nike right but we so we
34:05 didn't get the a account but no agency
34:08 did what we did get was we got the
34:10 Jumpman 23 site so we actually built
34:13 Michael Jordan's first website that was
34:15 pretty slick okay um our thing has been
34:19 generated now we're gonna add intro
34:21 music good catch there
34:24 Brandon
34:27 what generate audio first I
34:32 did add intro music oh oh I got to
34:35 generate the audio here got
34:37 it welcome to the Intriguing episode of
34:40 what the fact where we uncover the
34:41 stories behind things sneakers a
34:45 quick not quite Buster but thanks to the
34:48 AI Salon this
34:52 episode uh styish feet so let's Jump
34:56 Right In um
34:58 we'll say
35:00 check them out
35:03 at the
35:06 salon.
35:12 [Music]
35:15 a by the way while we're talking about
35:17 the salon.
35:19 a go there if you've not gone there and
35:24 check that [ __ ]
35:25 out that's the salon the salom bye Tony
35:30 see you later Tony all right let's
35:32 generate our
35:34 audio
35:38 Bang Bang audio generation started this
35:42 may take a few
35:43 minutes all right then add intro music
35:47 then publish beautiful
35:50 beautiful I'll tell you another thing
35:52 Tony hey
35:54 Tony hey it's
35:56 Frank Frank how you doing I'll tell you
36:01 what these these uh this this [ __ ]
36:03 Chinese deep seek R1 comes in holy [ __ ]
36:06 comes in and knocks everyone off their
36:08 [ __ ] Keers you know you know what I'm
36:10 saying all the big AI companies are like
36:13 I just raised half a trillion dollars
36:15 they're like hey I put together a couple
36:17 of buckets and a string and I got a
36:19 model better than
36:21 yours
36:25 anyway also go check out are you ready
36:28 for ai.com that's a course that uh I'm
36:31 putting together with an Murphy and
36:32 Vicky Baptist uh to train people how to
36:36 get ready for what's coming the fact of
36:38 the matter is you can't get ready for
36:40 for what's coming we were going to
36:42 create an AI um literacy course and then
36:45 we realized you can't there's nothing to
36:48 master right now this stuff's changing
36:49 too fast but what you can do is you can
36:52 start to shift your mindset and so
36:53 that's the course we're putting together
36:55 based on the um
36:59 the the the really cool event that we
37:01 had called AI festiv us so all right
37:05 let's go.
37:11 [Music]
37:16 MP3 uh where is
37:20 my
37:22 huh let's go
37:26 what the
37:34 name contains what the
37:36 no what Dash the hang
37:39 on what Dash
37:45 Thee I just looked this up hang on a
37:53 sec champion no that's Mama
37:59 what the fact there it is
38:01 okay hang on
38:10 people boom all right there's our
38:19 music welcome to another intriguing
38:21 episode of what the
38:24 fact and then we're going to publish the
38:27 episode
38:29 should we hear it
38:30 first how do you preview it can I
38:32 preview this thing oh yeah the giant
38:34 play button oops sorry hang on go check
38:38 out demystifying AI space in the salon
38:41 for all the uploads including the
38:43 notebook LM podcast very cool thank you
38:45 Pate we'll do
38:53 that oh are you kidding me
39:00 hang
39:04 on for people that don't know what my
39:07 exasperation is about you have to change
39:10 the way you're sharing your screen in
39:13 Chrome only because they're in some sort
39:15 of [ __ ] Feud with Apple so if you're
39:18 running a Chrome browser and you're
39:19 trying to share your audio you can only
39:22 share it in a tab which is [ __ ] up if
39:25 you have a a channel like chat add where
39:27 I'm flipping around between tabs like I
39:30 I have
39:31 memory
39:44 anyway welcome to another intriguing
39:46 episode of what the fact sounds
39:49 like doesn't it all right publish boom
39:53 and we're done I think we'll see
39:58 once you publish episodes you won't be
40:00 able to make
40:03 changes you can subscribe on
40:08 pocketcast yeah Pocket
40:11 Cast but wait so if I want to share this
40:14 I click this yeah there's the episode
40:17 share the
40:18 episode and then we're going to go to oh
40:21 I gotta I got to unshare hang
40:24 on dang it
40:28 now I got to go back and
40:29 reshare Tik Tock pin need more Cowell
40:33 yeah exactly it's definitely got some
40:36 SNL Vibes to it that little that little
40:38 songy song we'll do this with audio so I
40:42 can actually play this after we upload
40:47 it black bar I know I'm I'm a
40:50 professional I don't I don't need a
40:55 producer can you imagine to show with
40:57 that one well yeah we can imagine it we
41:00 experienced it for [ __ ]
41:04 years it was
41:07 awful I would I would seriously
41:10 sometimes talk for 45 minutes without
41:12 sharing the screen I thought I was
41:15 sharing um deep seek R1 this is from Sam
41:18 Alman an hour ago deep seek R1 can
41:21 everyone see this yes is an impressive
41:24 model particularly around what they're
41:26 able to deliver for the price why
41:28 everyone's stock price tanked we will
41:30 obviously deliver much better models and
41:33 it's also legit invigorating to have a
41:36 new competitor we will pull up some
41:39 releases one of the things that I think
41:41 all the naysayers I you
41:44 know it is amazing to me how
41:48 quickly um the tech bro culture turned
41:51 on all the big uh Frontier Model
41:54 companies they're like oh deep seeks
41:56 around you're all you're all worth L and
41:58 and you know out of out of business it's
42:01 like does no one understand that [ __ ]
42:04 changes so it's like the the
42:07 reason deep seek R1 is so good for so
42:12 cheap is that we put an export
42:16 restriction on the best AI chips so the
42:19 h100 chips the ones that all of our
42:22 models were trained on we basically
42:24 hobbled that thing and and gave China a
42:28 shitty version of it called the h800 the
42:30 r800 whatever I don't know it's some
42:32 different number it doesn't matter we
42:34 gave them a shittier chip now the rumor
42:38 is that uh that deep seek actually has
42:41 50,000 h100s that they're not supposed
42:43 to have but they don't talk about that
42:45 but we'll just we'll just let that rumor
42:47 slide that's conspiracy
42:49 theory because they had
42:52 shitty chips they completely rethought
42:55 how they trained them and whatever they
42:57 did to completely rethink how they
42:59 trained them made it really good really
43:02 fast
43:03 so you don't think that our Frontier
43:07 companies are going to take that
43:09 methodology and adopt it and improve it
43:11 and put it on the big ass things and
43:13 then we'll have some crazy ass [ __ ]
43:15 GPT 17.4 of course they will anyway
43:19 whatever just it's it's nuts okay so we
43:23 are going to say um on this
43:29 episode of what the
43:36 fact
43:39 sneakers let's
43:43 see
43:44 sneakes sneakers on on this episode I
43:49 guess we'll do four in parenthesis no
43:53 on how about here's episode four
44:01 here's episode four of what the fact we
44:04 drop that in there we'll go grab the
44:09 graphic what the
44:12 fact can
44:16 I add a link to this I don't think I can
44:20 right
44:22 no that's
44:24 stupid that's stupid Twitter hey Elon
44:28 why don't you make your stupid sight so
44:30 it's not so
44:31 stupid all
44:33 right thanks at jelly
44:45 POD at uh what was his name um Pier Pier
44:51 Pier uh not Piers Morgan not Brock
44:54 Pearson Pearson
45:01 Pearson Pearson
45:05 something
45:07 Pearson oh Pearson why do you have to
45:09 have such a fancy name anybody remember
45:14 Pearson there's no guarantee deep seeks
45:16 algorithm and training will lead to more
45:18 powerful models it already has what do
45:20 you mean what do you mean there's no
45:22 guarantee that that their algorithm and
45:25 training will lead to more powerful
45:26 models they trained on on 2,000
45:30 gpus
45:32 now according to them Pearson marks
45:34 that's what it is
45:36 marks
45:39 um let's do atso because that was what
45:43 where the song was
45:45 from uh let's tag Corey Sandler because
45:48 she made us our swell graphics
45:59 Cory Sandler
46:00 Pottery at Nike yeah why
46:03 not at
46:06 Nike we'll do at
46:09 Jumpman I wonder if Jumpman 23 is still
46:12 a
46:13 thing and no
46:17 no and we'll do scobleizer
46:22 not when you spell it that
46:25 way SC Eliz
46:36 wait oh scoer okay all
46:43 right there we go okay so here's my
46:45 request for all you people out there we
46:48 got 40 people over on the YouTube and
46:50 the X we have 68 people here on Tik Tok
46:54 go here go to
46:58 Twitter my account at Kyle Shannon and
47:01 Pump Up episode 4 let's turn this into a
47:03 hit
47:04 people well we don't even know if it's
47:06 any good yeah know
47:09 whatever oh tell me tell me also if
47:12 that's still broken if you try to open
47:15 this thing on
47:18 um on desktop at works and on mobile it
47:21 didn't liked by Kelly Camp Kelly Camp
47:24 gets the gold star she went first on x
47:27 she she liked this on X before you did I
47:29 just got it on my phone got a
47:32 notification if you want a gold star
47:34 next
47:44 time okay beautiful beautiful fantastic
47:49 Bob all right what are we gonna do I
47:51 guess let's switch over
47:53 to um so this Saturday I'm doing
47:58 um I'm doing an actual reading with
48:01 actors of the musical Sydney that I'm
48:03 working on so I thought maybe it might
48:07 be fun to play some songs from there but
48:09 I don't have to we can jump right into
48:11 maybe we do a little little poll on on
48:13 Tik Tok or something like that um or
48:16 maybe maybe I'll just make a call as if
48:19 I'm running the
48:21 show but we can either we can listen to
48:24 some Sydney songs folks waiting for 01
48:27 fine we'll do reasoning F we'll do it
48:29 live we'll do it live all right we'll do
48:35 reasoning we will
48:36 do um maybe we'll maybe I'll do a Sydney
48:39 thing Thursday night or maybe I'll I
48:41 don't know when I'm GNA get
48:48 in yeah maybe I'll go live Friday night
48:51 and I'll just do a special Sydney
48:53 episode Friday night I think that's what
48:55 I'll do all right
48:58 it's as if I don't know what I'm
49:02 doing you're not sharing sound on
49:04 YouTube oh God
49:10 Jesus well that's okay we're not gonna
49:12 need to share sound for what we're about
49:13 to do so let me go back to sharing
49:20 everything yeah Friday night date night
49:22 with a soundtrack I think that's right
49:24 it opened for me okay great fantastic
49:28 yeah maybe it just didn't work that
49:29 first night because the the you know URL
49:32 hadn't propagated or whatever the [ __ ]
49:34 has to happen with
49:36 domains there okay there's that that was
49:39 that
49:40 comment okay so what we're GNA do
49:43 now if you want to play
49:46 along jump on
49:49 over to chat. deep
49:55 seek.com and this is very much like chat
50:03 GPT God I hate how this thing displays
50:06 on Tik
50:08 Tock there's just no way to
50:13 control the look and feel oh well what
50:17 are you gonna do what are you gonna
50:19 [ __ ]
50:23 do more powerful means better reasoning
50:26 and more accurate results it has it is
50:29 more powerful and it has more accurate
50:31 results that's the that's the that's why
50:34 everyone's freaking out is that this
50:36 thing is is really quite good okay
50:39 so so what we'll do so so new app
50:43 profile am I logged in I must be where's
50:47 my history though oh failed to load my
50:52 history failed to load try reloading
50:57 shut up champy okay there's my history
50:59 great all
51:01 right all right fantastic they need a
51:04 better UI yeah they're I mean their UI
51:06 is just like chat gpts they just stole
51:08 it from chat
51:09 GPT so so yes you're correct they need a
51:13 new
51:16 UI okay so here's what we're gonna do so
51:18 so so what we're looking at
51:20 here we're playing with reasoning
51:22 engines the difference between your chat
51:25 GPT and a reasoning
51:28 engine uh is what's called Chain of
51:30 Thought reasoning so when you type a
51:33 prompt into chat
51:34 GPT it does this probability calculator
51:38 thing where it's predicting the next
51:40 token or the next batch of tokens and
51:42 you can effectively translate that into
51:44 it's predicting the next word or the
51:47 next series of
51:49 words and so you put in a prompt and it
51:52 basically answers you and it's got no
51:55 real idea of what it answered it's
51:58 literally just spitting out fragments of
52:01 words that it stitches together and then
52:03 we as humans read and go yes that's
52:05 awesome or no that's completely wrong
52:08 that's why we get things like
52:09 hallucinations right where it just does
52:11 [ __ ]
52:13 wrong reasoning engines rather than just
52:17 vomiting out the stuff they have this
52:19 whole background conversation going on
52:22 where you give it a goal and it
52:24 understands that goal and then it
52:25 creates a plan
52:28 right and it it and as it's creating the
52:30 plan it's talking to itself and figuring
52:33 [ __ ] out and and going step by step
52:35 through the process there there was a
52:37 prompt for chat GPT think step by step
52:40 will will improve a chat GPT prompt what
52:42 what that was effectively doing was
52:44 telling chat GPT to act like a Chain of
52:46 Thought Reasoner um they're just doing
52:48 it in a more sophisticated way
52:51 okay sound is amazing over here versus
52:54 Tik Tock great fantastic
52:57 thank you Lord digital gods for
53:00 thees that ni I like it I like it a lot
53:05 it's very
53:06 nice okay so what we're gonna
53:09 do the
53:12 the one of the things that
53:17 um that emerged over the weekend for me
53:20 is that the the reason people like deep
53:24 seek so much other than the fact that it
53:27 was you know way cheaper than than Chad
53:30 GPT um is that its
53:35 personality is pretty good so when it
53:39 like people are more excited about the
53:41 reasoning than they are necessarily
53:42 about the answers and so so we're going
53:45 to do that so what I'm going to do is
53:47 I'm going to come up with a single
53:48 prompt and then we're going to apply
53:50 that prompt in three different reasoning
53:52 models and then we'll look at the The
53:55 Chain of Thought reasoning and then
53:56 we'll look at get the answers and then
53:57 we'll just kind of compare them that's
53:59 kind of the plan for tonight um with
54:01 deep seek you can turn on reasoning with
54:04 with the left button you can turn on web
54:07 search with the search button and so you
54:10 can actually do a reasoning response
54:13 with web search which chat GPT doesn't
54:15 have and I we'll find out I can't
54:18 remember if if Gemini has it it may or
54:20 may not I can't remember but I'm I'm not
54:22 gonna I'm not going to use web search
54:24 I'm just going to use reasoning
54:29 oh yeah and there's there's a new one
54:31 that came out um today called keimy
54:35 1.5 U let's not worry about that right
54:38 now oh what I had
54:41 a I had a
54:44 metaphor that popped up today like
54:46 trying to keep track of all these things
54:48 oh it was so when I got really into
54:51 blockchain and nfts I I did like a
54:53 two-year Deep dive into that world and
54:56 when I was trying to keep up with the
54:57 nft projects I started creating
55:00 spreadsheets because I couldn't keep
55:01 track of everything and that's where I
55:03 am right now with all these AI models
55:06 with all the like the video tools and
55:07 the image tools and now the reasoning
55:09 tools and who's got what features and
55:11 wait was that feature on that tool or
55:13 was it on that tool I feel like we're
55:15 just in this place where if you're
55:17 trying to keep up with this stuff you
55:19 you like you literally need AI to help
55:22 you make sense of it um YouTube question
55:25 is deep seek safe
55:28 well if you're on chat. deep
55:32 seek.com you should assume that any data
55:35 you put in here is going straight into a
55:37 Chinese
55:40 database period end of
55:42 story
55:44 now what they did with this model
55:47 however is they they gave it to us in
55:49 open source so if you go to I assume
55:52 it's on GitHub or or on uh hugging face
55:56 it's on one of those two I assume it's
55:57 on GitHub hang on I
56:01 gotta do that
56:05 um you can you can run these models
56:08 either locally on a machine they G they
56:11 gave different Siz models from 1.5
56:13 billion parameters up to 70 billion
56:16 parameters and there's people right now
56:18 that are putting that 1.5 billion
56:20 parameter version of Deep seek on phones
56:22 they're running them locally on phones
56:24 so if you're running it locally your
56:26 data is not going anywhere so in that CA
56:28 you know in that in that sense it's safe
56:32 within your hermetically sealed
56:34 container where you're running it right
56:36 the most safe would be run it on a local
56:38 computer or a local device the next safe
56:41 would be spin up a GPU instance in the
56:44 cloud that's
56:45 secure you know and then the least
56:48 secure is go to chat. deep seek.com but
56:50 if you're lazy like
56:53 me or if you go download the iOS app
56:56 that's that's using their servers just
56:58 assume that anything you put in here is
57:00 not safe um but but you know is it is it
57:04 is it unsafe no it's not unsafe it's
57:06 just assume that your data is gone being
57:10 used elsewhere okay I'm gonna go
57:14 to this thing and go
57:17 [Music]
57:19 here all right and then I'm going to go
57:21 g g g and I'm going to go uh this is the
57:25 Deep seek
57:31 prompt and let's change this to title as
57:35 well yeah so you can see
57:40 it we'll call this we'll call this the
57:42 reasoning
57:43 prompt so I'm going to write one prompt
57:46 for all of these
57:50 engines okay let's see
57:57 I
58:04 am I
58:07 am
58:09 a a creative
58:18 professional
58:20 that does not
58:23 have PhD level
58:27 problems to
58:30 [Laughter]
58:35 solve
58:36 however comma I
58:39 want to take
58:45 advantage
58:50 of the new AI
58:53 reasoning engines
58:57 that use Chain of Thought
59:06 reasoning and
59:08 I'm hoping you can
59:13 help here are some of
59:17 my
59:19 interests
59:22 and
59:23 uh areas of
59:32 aptitude if you're thought right now is
59:36 couldn't you have done this
59:39 before um good
59:47 point yes yes this must be insufferable
59:51 to watch but that's
59:53 okay because I'm
59:57 I I think this is well I I'm hoping that
1:00:00 this is valuable because we're all going
1:00:03 to have to kind of muddle through this
1:00:05 like we've never had I don't I have
1:00:07 never personally had like a 30 PhD
1:00:12 buddy right so I don't know what to ask
1:00:15 a 30 PhD buddy the closest thing I have
1:00:17 is pate and when he answers me I'm like
1:00:20 I don't know what he's saying and then
1:00:22 he's like all right I'll dumb it down
1:00:23 for you
1:00:26 he does it pretty good
1:00:29 actually um okay let's make this a
1:00:32 little bigger so you can see it
1:00:41 okay let me Let Me Entertain
1:00:47 [Music]
1:00:50 You all right there we go beautiful
1:00:53 fantastic
1:00:55 blob okay
1:00:56 Tik Tock question Kyle Irregulars
1:01:00 Irregulars post of my random words as a
1:01:03 prompt in deep seek okay
1:01:06 cool hey I'm a college student trying to
1:01:09 be a software engineer remember to
1:01:11 always check the sources yes I
1:01:15 agree um always check the sources I saw
1:01:18 a Tik Tok video on that today that
1:01:20 someone put a legal brief in and had it
1:01:22 write a summary of the legal brief and
1:01:25 then um
1:01:27 chat GPT got them all wrong and then
1:01:29 basically just she she called it it
1:01:31 quiet quit on her where it said it was
1:01:33 trying but it just stopped trying uh and
1:01:36 then deep seek um actually got all the
1:01:39 the summaries correct so um so it this
1:01:41 might be better for summarization but
1:01:43 anyway okay so here's my some of my
1:01:45 areas of aptitude um
1:01:48 um creative writing for
1:01:53 theater um fiction
1:02:01 marketing um uh let's
1:02:05 see
1:02:08 um
1:02:11 performance and
1:02:14 acting um
1:02:18 [Music]
1:02:20 entrepreneurship and being a
1:02:24 Founder um
1:02:27 first
1:02:29 online
1:02:32 magazine
1:02:35 first digital
1:02:38 agency
1:02:39 um
1:02:43 patented video uh
1:02:47 Auto
1:02:50 mated video
1:02:55 platform uh uh uh Community
1:02:58 Builder um
1:03:02 World
1:03:03 [Music]
1:03:05 Builder worldwide web artists
1:03:11 consort
1:03:13 1994 and AI Salon the
1:03:19 salon.
1:03:23 2022 um
1:03:29 educator Tik tock.com
1:03:35 slash AI learning
1:03:38 lab so what I'm doing here is just
1:03:41 trying to give it who I am what I'm you
1:03:44 know the kinds of things that I do and
1:03:46 know how to
1:03:48 do um what else is on this list um
1:03:53 leadership and uh people
1:03:57 management
1:04:01 consulting
1:04:05 um comedy
1:04:11 jokes oh
1:04:14 musician
1:04:17 um you should ask chat gp01 mini to
1:04:20 write this bio yeah I could do that well
1:04:23 that's okay all right so that that's
1:04:26 enough that's enough kind of stuff there
1:04:28 okay so
1:04:30 now let me go back here and make this
1:04:33 big again for you so I'm going to
1:04:37 say I want you to think long and
1:04:43 hard about the kinds of things I could
1:04:49 use a
1:04:54 reasoning engine
1:04:58 four
1:05:01 and break the use
1:05:05 cases
1:05:14 into pink bow
1:05:17 owner um break the use cases
1:05:23 into um categories
1:05:31 you think will be
1:05:34 Val Val
1:05:38 valuable and then in each
1:05:43 category
1:05:45 provide
1:05:48 a provide
1:05:53 three use cases
1:05:58 and then we'll say
1:06:01 basic
1:06:04 intermediate and
1:06:10 advanced and describe the use
1:06:16 case why a
1:06:19 reasoning let's see
1:06:22 um why it makes sense
1:06:29 and then
1:06:36 include the prompt
1:06:41 for the prompt to
1:06:45 use
1:06:47 in a reasoning
1:06:51 engine okay good Lord Okay so
1:06:56 I basically
1:06:58 say I
1:07:00 don't I don't have the training or or
1:07:03 the intellectual capacity to really know
1:07:06 how to use this [ __ ] tool that's in
1:07:08 front of me however here's some [ __ ]
1:07:11 I've done right so you could do this you
1:07:13 make your own list of
1:07:15 [ __ ] and then basically I want you to
1:07:18 think of ways that I could use a
1:07:20 reasoning engine right does that make
1:07:21 sense this is amazing on my huge TV
1:07:25 that's terrifying but but awesome okay
1:07:27 what a great idea Valerie what did
1:07:29 Valerie say oh I watch him on TV on
1:07:31 YouTube and chat on my phone oh that's
1:07:33 yeah the dual screen thing people do
1:07:35 that a lot we're all here for the fun
1:07:37 that's great that's really cool Tams is
1:07:40 loving it all right rock on Dax is
1:07:42 loving it cool all right Pate says he
1:07:45 has some PhD problems you can borrow
1:07:47 yeah well actually you know maybe that's
1:07:49 something we do here one night Pate is
1:07:52 because I got 30 days of deep Dives
1:07:54 maybe actually Pate you know what could
1:07:56 be kind of fun is you come up with a uh
1:08:00 you know PhD problem or two and then
1:08:02 maybe I'll bring you up on stage and
1:08:04 we'll we'll just dick around in these
1:08:06 reasoning engines and see if we can get
1:08:08 them to do anything
1:08:10 interesting um Source Camp my prompts
1:08:12 include some form of I don't know what
1:08:14 I'm doing yeah ex great oh that's
1:08:17 perfect I'll do that um you know uh wait
1:08:21 hang on let me make it a title so you
1:08:23 can see it uh as
1:08:29 reminder uh I'm a bit of a dumb
1:08:34 [Laughter]
1:08:37 dumb but
1:08:40 um I I'm really
1:08:43 sweet and well-meaning
1:08:46 that's gonna become part of our prompt
1:08:49 all right so we're gonna copy this whole
1:08:53 thing pat down for whatever great
1:08:56 cuno impostor prompting yeah exactly
1:09:01 exactly all right so here's my impostor
1:09:04 prompt and and listen I mean I'm doing
1:09:07 this a bit tongue and cheek right you
1:09:09 know that that that I'm a bit of a dumb
1:09:11 Dum and and you know I'm not really
1:09:13 smart enough to use this thing but but
1:09:16 to I I I mean the point still stands
1:09:19 just in my life you know I don't have a
1:09:23 lot of people presenting me with with
1:09:27 big hairy problems to solve um or or
1:09:31 like the ones that like like when I
1:09:32 wanted to come up with what storyvine
1:09:34 has become and what that technology has
1:09:37 become you know that was two or three
1:09:39 years of solving a problem I could now
1:09:41 probably take the problem that I solved
1:09:43 with storyvine and I probably could have
1:09:45 solved it you know in a [ __ ] week
1:09:48 right that's that's the difference for
1:09:50 where we are right now but knowing
1:09:52 having the problems is part of the
1:09:54 challenge so let's go use these tools
1:09:58 all right I'm an
1:10:25 problem hang on hold
1:10:29 please I'm like that doesn't make any
1:10:32 sense that you can't paste it in a long
1:10:35 prompt boom okay so there we go I'm I'm
1:10:39 a creative professional that does not
1:10:41 have PhD level problems to solve however
1:10:45 there's what I do I want you to think
1:10:47 long and
1:10:48 hard all right and then we're going to
1:10:53 go make that bigger put this in here so
1:10:57 you can see it error between seat and
1:11:00 keyboard
1:11:02 chat
1:11:08 huh in the
1:11:14 prompt never mind chat okay
1:11:17 whatever I'm using that I'm Kelly and
1:11:20 I'm
1:11:22 anomor um all right so here we go bang
1:11:31 thinking oh it's slow everyone's using
1:11:34 the [ __ ] out of this thing this was much
1:11:35 faster earlier okay so the first thing
1:11:37 you so what it's giving you right now so
1:11:39 this little line down the side says that
1:11:41 it's giving you its thinking so it's now
1:11:44 it's thinking so this is the what we're
1:11:47 getting to witness here is the Chain of
1:11:49 Thought reasoning so this is its
1:11:52 conversation with
1:11:53 itself what's cool about it is it's very
1:11:56 conversational okay let's start by
1:11:58 understanding the users's background and
1:11:59 needs they're creative professional with
1:12:01 experience in writing performance
1:12:02 entrepreneurship Community Building
1:12:04 education and more they want to use AI
1:12:06 reasoning agents but don't have PhD
1:12:08 level problems great first I'll break
1:12:10 down their listed interests into
1:12:12 categories the user mentioned creative
1:12:14 writing performance entrepreneurship
1:12:16 Community Building education right gives
1:12:18 me the long list these use cases each
1:12:21 each of these should be a category for
1:12:23 each category I I need to so so look at
1:12:26 this for each category I need to think
1:12:28 of basic intermediate and advanced level
1:12:31 use cases so it took that basic
1:12:33 instruction I did and it kind of
1:12:35 internalized it right so it's responding
1:12:38 as if it's a just like a a person right
1:12:41 let's play with Janice Janice Pro images
1:12:44 uh yeah I would love to do that actually
1:12:46 um ramik do you know what uh
1:12:49 youl like where where are they hosted I
1:12:53 assume they're not on on chat. deep
1:12:55 seek.com yet but if they are let me
1:12:58 know um okay starting with creative
1:13:01 writing um it's coming up with ideas
1:13:04 next is performance or acting it's
1:13:06 coming up with ideas here's
1:13:07 entrepreneurship ideas here's community
1:13:10 building ideas here's education ideas
1:13:12 right so it's giving itself all the
1:13:13 ideas then it says each use case needs
1:13:16 to prompt needs needs a prompt that
1:13:18 guides the a AI through the reasoning
1:13:20 Pro process the prompt should be
1:13:23 specific and trigger a Chain of Thought
1:13:26 right so it's so then now here's the
1:13:30 answer so now we get it in bright white
1:13:33 right so the thinking is in this gray
1:13:34 with the little line and then here's the
1:13:36 answer let's break this into categories
1:13:39 aligned with your skills and use cases
1:13:41 basic to Advanced and tailored prompts
1:13:44 for reasoning engines like chat gbt
1:13:46 gemini or Claude these prompts will then
1:13:48 guide the AI to think step by step okay
1:13:51 great well we don't need it to add the
1:13:54 think step by step
1:13:57 all right but let's see what it did
1:13:58 creative writing basic use case
1:14:01 brainstorming unique story premises why
1:14:04 overcome writer block The Prompt act as
1:14:06 a creative writing partner I need five
1:14:08 story premises that blend genre
1:14:12 theme for each premise suggest a
1:14:15 protagonist flaw and a symbolic okay so
1:14:18 that's that intermediate character
1:14:20 motivation Mapp mapping deep charact
1:14:24 deep deepen character arcs by forcing
1:14:28 contradictions all right these prompts
1:14:30 seem a little thin I'm going to I'm
1:14:32 going to tell it that I'm going to
1:14:36 say um your prompts
1:14:40 seem a little
1:14:44 thin
1:14:47 and lack
1:14:52 specificity um expand on each use
1:14:58 case with more real world
1:15:03 examples to
1:15:06 inspire me to think outside the
1:15:11 box or whatever that [ __ ]
1:15:17 [Music]
1:15:20 means all right let's go back up here
1:15:22 while it's while it's thinking we see
1:15:24 see some of these other ideas as it gave
1:15:25 us so there was creative
1:15:27 writing um Advanced meta narrative
1:15:30 structure design experiment with
1:15:32 nonlinear storytelling that's pretty
1:15:34 cool that's actually a really good one
1:15:36 for a reasoning engine right
1:15:39 entrepreneurship and Consulting audience
1:15:41 paino generator pretty
1:15:43 cool business model Paradox resolver
1:15:46 tackle contradictory goals scale fast
1:15:49 but stay Niche cool anti-brand strategy
1:15:53 all right now oh it's thinking again
1:15:56 all right let's see let's see what it
1:15:58 does okay let me try to figure out how
1:16:01 to approach this the user found my
1:16:03 previous prompts a bit thin and lacking
1:16:06 specificity they want more real world
1:16:08 examples to inspire them to think
1:16:10 outside the box so I need to expand each
1:16:12 use case it's fascinating like how this
1:16:15 thing interprets my prompt and then
1:16:17 expands it into what it's going to do
1:16:19 right so this is now a plan but it's
1:16:22 kind of internalized it like it's my job
1:16:24 to help this guy all right sidebar
1:16:27 question on Tik Tock um I want to train
1:16:29 my chat gbt but my son uses the same
1:16:32 account must I pay for a new a new thing
1:16:37 um
1:16:38 yeah you do you can get a team account
1:16:42 so that you you he's got access
1:16:46 and but that I don't I just yes you do
1:16:50 talk to Cindy [ __ ] on here or on the on
1:16:53 the AI Salon she's got
1:16:56 this happened with her and her daughters
1:16:57 her daughters were were like Mom would
1:17:00 you shut up about the AI we're we hate
1:17:02 the AI the AI is
1:17:04 stupid and then she had them do
1:17:07 something where it solved a really cool
1:17:08 problem for them and they're like oh my
1:17:10 God thei so cool we want our own and so
1:17:15 so I think she's a three chat GPT house
1:17:17 at this
1:17:18 point so yes that's a pain in the ass
1:17:23 you could just have him work on on deep
1:17:26 seek deep seek is free and it's better
1:17:28 than chat GPT it's some of the basic
1:17:31 [ __ ] it's also Wicked slow right
1:17:34 now Wicked slow this this was not this
1:17:37 slow earlier first let's look at the
1:17:40 categories they
1:17:41 mentioned all right so it's still thing
1:17:44 okay so let's go on so now let me go let
1:17:46 me go blow your mind with something
1:17:47 we're gonna go back to get the prompt so
1:17:50 we'll copy the prompt from here while
1:17:52 that works we're going to go to grock
1:17:57 um and grock is not grock is not Elon
1:18:02 Musk grock but this is
1:18:04 grq so how is this prompt different than
1:18:07 a chat gbt the The Prompt is not
1:18:10 different how a reasoning engine
1:18:12 responds to The Prompt is different what
1:18:15 I'm trying to get is not so much the
1:18:17 prompt I don't really give too much of a
1:18:19 [ __ ] about the the the phrasing of the
1:18:22 prompting however I I wanted it to be a
1:18:24 little better than it it gave me but I I
1:18:27 want to come up with use cases just
1:18:29 thinking of like what is a thing I could
1:18:31 have a reasoning engine do versus normal
1:18:34 chat GPT part of my challenge
1:18:38 is normal chat GPT is already well
1:18:42 beyond like it can solve almost anything
1:18:45 I throw at it right so my question is
1:18:48 why am I using a reasoning engine on top
1:18:50 of
1:18:51 that all right Janice Pros on hugging f
1:18:55 okay great we can go play with that
1:18:56 there yeah so
1:19:00 so um deep seek R1 is the reasoning
1:19:03 engine and today they came out with a
1:19:06 multimodal um also open- Source engine
1:19:11 um that can make images so so just like
1:19:13 DOI makes images it's just like that
1:19:16 only it makes images so all right Tik
1:19:18 Tock pin might want to engage all
1:19:21 right no it's not it's dangerous it made
1:19:23 fake FDA letters on its own you can get
1:19:26 AI to make anything you can you can get
1:19:29 AI to make fake FDA letters you can get
1:19:32 it to make you know real well researched
1:19:35 ones we're still at a point with
1:19:39 AI that we need humans in the
1:19:42 loop right and AI is like any other tool
1:19:48 you could make a fake a you could make a
1:19:49 fake FDA letter with Microsoft Word and
1:19:54 a fake FDA logo with Photoshop right the
1:19:57 the tools are the tools now I'll give
1:20:00 you that these are much more powerful
1:20:01 tools and people that don't understand
1:20:04 them and don't understand they might
1:20:05 hallucinate they might inadvertently do
1:20:08 stupid [ __ ] but that's on us as people
1:20:12 it's on us as people to to understand
1:20:15 one of the reasons I go live five nights
1:20:17 a [ __ ] week to teach this stuff is to
1:20:22 encourage everyone to play with these
1:20:26 tools not to believe in them like
1:20:28 they're you know [ __ ] you know the
1:20:30 savior of of of humanity but to play
1:20:33 with them because they're not going away
1:20:36 and they're getting more and more
1:20:37 powerful and it's our job in my opinion
1:20:41 to to understand what they're capable of
1:20:44 and where the boundaries are and right
1:20:47 now the boundaries are are relatively
1:20:50 inclose like they're within they're
1:20:52 within grasp of us right
1:20:56 so when it does research of a legal
1:20:58 brief and it summarizes it and it puts
1:21:00 citations in there that are not correct
1:21:03 that's on the person that made that
1:21:05 summary to check that
1:21:07 [ __ ] there will be a point at which
1:21:10 these things are so good that they won't
1:21:11 make those mistakes but we're not there
1:21:13 right now but yeah any tool can be used
1:21:15 for any any bad because it can be used
1:21:18 for evil doesn't mean that it's
1:21:20 inherently evil just doesn't
1:21:23 okay I ask for practical non-digital use
1:21:26 cases that's great yes but it said it
1:21:29 hacked
1:21:30 itself and was 18% its previous
1:21:33 version it also made
1:21:36 up I cracked Claud yeah if listen if you
1:21:40 want to go find examples where it did
1:21:43 shitty things and it did stuff that's
1:21:44 scary you can do
1:21:47 that I I don't I don't understand the
1:21:50 point of your thing like should no one
1:21:52 use these tools
1:21:53 because they did some things that were
1:21:56 kind of scary that kind of felt like a
1:21:57 sci-fi
1:21:59 movie or do we want to learn that they
1:22:01 can do that and work with companies to
1:22:04 say let's make them more safe so they
1:22:06 don't do that right the whole purpose of
1:22:09 putting these tools in the hands of
1:22:11 people is to understand what these
1:22:13 things are and what our responsibility
1:22:15 is and putting them out there all right
1:22:17 archetypal architect on YouTube oh
1:22:21 that's my fault I had it thinking about
1:22:23 a prompt that would really check
1:22:25 challenge it and
1:22:26 then no reason on that prompt all right
1:22:29 Sky Timbo um I've even reached a factual
1:22:33 error with notebook LM even though it
1:22:35 only responds to its uploaded docks yeah
1:22:37 these tools are jankier than [ __ ] just
1:22:40 like again these are not all good
1:22:44 they're not all bad they are what they
1:22:46 are they are
1:22:48 algorithmic knowledge
1:22:53 engines what I would argue is that AI in
1:22:56 its current
1:22:58 state what these things really do is
1:23:01 they take any idea that you have and
1:23:02 they amplify it right they allow you to
1:23:04 get it out of your head and get it into
1:23:06 the world and you can get any D any idea
1:23:09 in your head out in the world to about
1:23:12 an 80% complete level not much more than
1:23:16 that that 20% is really [ __ ]
1:23:19 important right right now that's on us
1:23:23 as these models get better and better
1:23:25 that 20% goes to 18% then 15 then 10
1:23:29 then five three years from now you're
1:23:32 going to be able to put things out of
1:23:34 these things that you know are very very
1:23:37 close to final if not final but we're
1:23:39 not there right now anyway all
1:23:41 right g g g so what I want to show you
1:23:45 here is grock gro.com gq.com
1:23:51 is a hardware company and they create
1:23:54 these special specialized inference
1:23:56 chips and inference is the moment when
1:23:58 you give it a prompt and it gives you
1:24:00 the answer that's inference that's the
1:24:02 the the moment of inference and so these
1:24:05 chips are hyp specialized for just text
1:24:08 inferencing text large language model
1:24:11 inferencing so you're going to see you
1:24:13 saw how slow um chat. deep seek.com was
1:24:17 watch how fast this is so it's still the
1:24:20 same model deep seek R1 but it's running
1:24:22 on this faster Hardware so there's myig
1:24:25 ass
1:24:26 um apologetic prompt and watch this
1:24:34 bang like it's just wow it's thinking
1:24:38 deep holy
1:24:43 [ __ ] like it's all
1:24:50 done okay so I'm trying to figure out
1:24:52 how to use I use AI reasoning engin for
1:24:56 my creative projects first let's look
1:24:58 into categories starting with creative
1:25:00 writing okay so it did all the different
1:25:02 categories so yeah it approached it the
1:25:03 same that makes
1:25:05 sense creative
1:25:07 writing prompt analyze so it's it's
1:25:10 still given us short little
1:25:12 prompts theme suggestion timeline
1:25:15 consistency create a timeline of events
1:25:17 based on my plot ensuring oh that's kind
1:25:21 of cool so you could give you could give
1:25:23 an outline of a story and it'll create a
1:25:25 timeline that's kind of cool performance
1:25:28 and acting character dialogue analysis
1:25:30 that's that's although that's regular
1:25:33 chat GPT could do that physical
1:25:35 mannerisms this is pretty this is pretty
1:25:38 shitty it looks like they've got um like
1:25:41 the the amount of token outputs for this
1:25:43 really low um so while it's fast it's
1:25:47 very thin so I don't like that all right
1:25:48 let's go to
1:25:50 Gemini okay so where I am right now is
1:25:53 I'm at AI stud i.g
1:25:57 google.com and there's a lot going on on
1:26:00 screen here on the left hand side is so
1:26:02 so this is if if you've ever used um
1:26:05 open AI playground where you can design
1:26:09 your prompts for applications You're
1:26:10 Building you've got things like
1:26:12 temperature and you've got you know some
1:26:15 more uh input options like one of the
1:26:19 things that we can do here is I can
1:26:20 actually put in a system prompt so I
1:26:23 could say that I want I want this uh
1:26:26 this
1:26:28 uh response to only respond in you know
1:26:31 in comedy or or whatever right so it's
1:26:34 faster I thought open AI intentionally
1:26:37 slow rolled responses so that it felt
1:26:39 more
1:26:40 human
1:26:44 um I don't I don't know what that's in
1:26:47 reference to John um I think that um
1:26:52 open AI slow rolled oper
1:26:55 operator um just so that people could
1:26:57 keep up with
1:26:59 it chat GPT is free but China based if
1:27:03 that matters it's cheaper faster
1:27:06 stronger oh that was to the tech text
1:27:08 response well what I was just demoing
1:27:11 over on gro.com this is a this is a
1:27:15 hardware company
1:27:17 demo you just happen to be able to you
1:27:20 you can choose a bunch of different
1:27:22 models um and I just chose they they put
1:27:25 deep seek R1 on here so you can just
1:27:27 compare the speed of this output with
1:27:29 the speed of of uh of the regular chat
1:27:35 why not use beam compare all the all the
1:27:38 llms in seconds using the same prompt
1:27:40 because I'm a dumb dumb we established
1:27:42 that
1:27:43 earlier
1:27:49 um um okay so in in Gemini
1:27:55 over here on the right is where you
1:27:57 choose your um your
1:27:59 models there's a bunch of different
1:28:01 models here you
1:28:04 know oh this is good one thing that
1:28:07 they've done well here is for each of
1:28:09 the different models they now give you
1:28:11 what it's good for and what its cost is
1:28:14 and so we're going to go to flash uh
1:28:17 thinking experimental so this is their
1:28:19 reasoning engine this is Gemini's
1:28:20 reasoning
1:28:22 engine so we're going to take our big
1:28:23 ass prompt
1:28:26 is this might
1:28:29 no we go back over here we'll copy this
1:28:32 we'll go back over here we'll paste
1:28:39 [Music]
1:28:41 this all right there's our prompt
1:28:45 run all right
1:28:47 thinking expand of view thoughts all
1:28:50 right so here's our
1:28:51 thoughts so remember R1 was pretty um
1:28:55 good question on Tik Tok oh my God how
1:28:57 many Tabs are open right now uh I'm
1:28:59 going to guess
1:29:06 84 oh that's not the
1:29:11 question the question
1:29:15 pinned oh Tik
1:29:17 Tock what is the point of open AI now
1:29:21 so your question GR grat crack grat
1:29:27 crack you um your question is why the
1:29:31 NASDAQ tank
1:29:34 today about a year ago maybe a year and
1:29:37 three months ago there was a Google
1:29:39 engineer that I think left Google no no
1:29:42 I think he was still at Google and he
1:29:44 wrote it I don't know if he's still
1:29:45 there P might know um he wrote a piece
1:29:49 called there is no
1:29:50 moat and what he basically talked about
1:29:53 is that the cost of intelligence is
1:29:55 going to Trend to zero and everyone will
1:29:58 be able to replicate what everyone else
1:30:00 is doing and the fact that this Chinese
1:30:02 company created a state-of-the-art
1:30:04 reasoning model that beats open Ai and
1:30:09 you know the the the other things that
1:30:11 are out there
1:30:14 um for a fraction of the price is what
1:30:18 sent Nvidia stock you know tanking and
1:30:21 Oracle stock tanking and the NASDAQ
1:30:23 tanking
1:30:26 again here here's where I
1:30:30 go while it's possible
1:30:35 that creating these powerful models is
1:30:38 trivially easy it is probably the case
1:30:41 that having state-of-the-art models is
1:30:43 going to require massive amounts of
1:30:45 infrastructure if it's not then the the
1:30:49 drop in price makes sense um
1:30:54 just because a model came out that
1:30:57 caught up and and surpassed other people
1:30:59 for way way cheaper all that's going to
1:31:01 do Sam Alman said it an hour ago that's
1:31:04 a kick in the ass for them to say we
1:31:05 better get our [ __ ] together and re
1:31:08 reestablish our Supremacy right so like
1:31:12 this ain't over like I said these tools
1:31:15 right now as impressive as they are are
1:31:18 not ready for prime
1:31:20 time like the fact that I'm having to go
1:31:22 to three different sites to compare
1:31:25 three different reasoning engines to see
1:31:27 what is what is it that they actually do
1:31:29 and are they any good like at some point
1:31:32 in the future we're not going to have to
1:31:33 choose models you're just going to ask
1:31:36 the the the chat hole as we as we
1:31:39 defined it on Friday you just ask the
1:31:41 chat hole your question or give it your
1:31:44 goal and this shit'll go out and it'll
1:31:46 figure out what models it needs to use
1:31:48 and it'll go do all your [ __ ] for you
1:31:50 we're not there
1:31:51 yet the downside of being this early is
1:31:55 you have to be a little geeky you have
1:31:56 to be hyper curious you have to be hyper
1:31:59 adaptable like I could teach you
1:32:01 something tonight here's exactly how to
1:32:03 use this reasoning engine and then deep
1:32:05 seek comes along three days later and
1:32:07 completely blows that out of the water
1:32:09 and now you got to go learn that that's
1:32:11 just where we are right now the upside
1:32:14 of being this early is that the amount
1:32:15 of opportunity for people that are this
1:32:18 early is
1:32:19 massive because most people are ignoring
1:32:22 this [ __ ]
1:32:25 like even if you're in here like the
1:32:26 person that was in here that was was
1:32:28 saying these things are bad these things
1:32:29 are evil at least you're [ __ ]
1:32:30 interested in them at least you're
1:32:32 willing to be in the dialogue most
1:32:34 people are like ah whatever hey I yeah I
1:32:36 heard it I heard about it yeah whatever
1:32:38 I used chat GPT once I didn't like
1:32:41 it that's most of the world right
1:32:46 now it's it's it's amazing to me all
1:32:49 right is training a new model right
1:32:55 or should I work on existing models
1:32:57 that's an interesting
1:32:58 question it depends on what you want to
1:33:02 do like if
1:33:04 you're 100 folks in the live hey
1:33:07 everybody if you're new here welcome my
1:33:09 name is Kyle Shannon this day I learning
1:33:10 lab the qu the question is should I
1:33:13 train a new model or fine-tune a
1:33:16 model or should I work with existing
1:33:18 models it depends what what you
1:33:21 do like if I'm working on my my
1:33:26 musical I don't need to train a
1:33:29 model current models are fine if I'm
1:33:33 trying to cure some genetic disease I
1:33:35 might want to do a specialized
1:33:37 [Laughter]
1:33:42 model um aios in its AOL disc era yeah
1:33:48 yeah a or AI is in its AOL disc yeah
1:33:52 right now
1:33:54 metaphorically speaking if if you were
1:33:57 to compare where we are with AI to where
1:33:59 we were with the worldwide web because I
1:34:02 was there for that like I know what
1:34:04 phases we went through a lot of people
1:34:07 are saying oh this is like the the the
1:34:09 dotc bubble we're not even [ __ ] close
1:34:12 to the.com bubble that was the.com
1:34:15 bubble happened in
1:34:16 2000 um and it was clear that it was
1:34:18 going to happen in
1:34:20 1999 we're way back in like 95 and a
1:34:24 half maybe 996 we're very
1:34:29 early right early in the web you had to
1:34:33 teach yourself HTML you had to we had
1:34:35 dial up modems you didn't even have the
1:34:37 internet all the time all all that sort
1:34:39 of stuff
1:34:40 so you know we we're not even close
1:34:42 we're so early we're so early no it's
1:34:46 it's because these tools are so capable
1:34:48 it doesn't feel like that but we ain't
1:34:50 seen nothing yet question human
1:34:52 creativity still priceless but not all
1:34:55 AI is ideal like being a detective you
1:34:58 have to see the clues that's correct and
1:35:00 I think you to your point human
1:35:02 creativity intuition critical thinking
1:35:06 um um discernment and judgment having
1:35:10 ideas in the first place again I think
1:35:13 that our job right now our job moving
1:35:16 into the
1:35:17 future is very much going to
1:35:21 be understanding within ourselves what
1:35:23 do we want to do in the
1:35:25 world what do we want to do how do we
1:35:28 want to impact people because these
1:35:31 tools are going to help us do
1:35:34 that but you have to have the question
1:35:36 you have to have the the imagination you
1:35:39 have to have the
1:35:41 creativity to go
1:35:44 oh I I mean here's by the end of
1:35:48 2025 I would I would conservatively say
1:35:52 by the end of 2025 if you're paying
1:35:54 attention to this stuff and you know
1:35:56 where to look and you know which tools
1:35:57 to
1:35:58 use every person that is in that hyper
1:36:01 curious hyper adaptable state for for
1:36:03 2025 by the end of 2025 we're all going
1:36:07 to have at a minimum
1:36:10 10 PhD
1:36:13 level
1:36:16 employees that can be PhD level at
1:36:19 anything we want them to be so I would
1:36:22 start asking yourself right now kind of
1:36:24 why I wrote this question the way I
1:36:28 did if you could have 10 PhD level
1:36:30 employees that will work for you
1:36:33 247 and do anything you want them to do
1:36:37 what would you have them
1:36:39 do that's not an AI problem that's a
1:36:42 human
1:36:43 problem and then you as the one that
1:36:46 have that Vision have that idea you sort
1:36:48 of put your employees out to go do the
1:36:50 [ __ ] for you and they come back and then
1:36:53 it's your job to go did they get it
1:36:55 right did they get it
1:36:57 wrong is it on track is it off track so
1:37:00 just like you would run a company you're
1:37:02 going to run your own little mini
1:37:03 company you're GNA have 10 employees
1:37:06 maybe you'll have five maybe you'll have
1:37:07 a hundred maybe you'll have a thousand I
1:37:09 don't know but let's assume you'll have
1:37:13 10 can you think right now if you had 10
1:37:17 PhD level employees that could execute
1:37:20 at a worldclass level what would you do
1:37:23 with them and if you don't have an
1:37:25 answer for that [ __ ] start thinking
1:37:27 about it so anyway that's that's what
1:37:30 this prompt is new here watch the
1:37:33 backlog of AI content on YouTube oh oh
1:37:36 if you're new here yeah you can go to uh
1:37:39 YouTube uh and the account is
1:37:42 learninglab D aai and you can watch all
1:37:44 the back things thank you
1:37:49 [Music]
1:37:50 Brandon gotta have the uh the AWA
1:37:56 okay so here's Gemini's response to our
1:37:58 prompt um understand the core the users
1:38:01 core need user wants to leverage AI
1:38:04 reasoning analyze so this is a much more
1:38:06 cold response this is is this all the
1:38:08 thinking stuff oh no here's there's my
1:38:11 prompt oh here's the
1:38:14 thoughts okay here's the
1:38:17 thoughts so understand the users core
1:38:19 thing
1:38:26 okay wait I got to challenge this SEL AI
1:38:31 can't
1:38:33 think can't think of new things it works
1:38:36 on what already
1:38:38 exists my question to you is how of have
1:38:41 have you used it how often have you used
1:38:43 it because that's a Trope that I
1:38:47 actually don't agree
1:38:49 with
1:38:51 technically you're correct technically
1:38:54 there is some Big Blob of
1:38:57 data so it's been trained on what's
1:38:59 known
1:39:02 right and you put in a prompt and it
1:39:05 puts out a probability response for a
1:39:08 token and then it does that a bunch of
1:39:10 times in a row right so in
1:39:13 theory since it was trained on [ __ ] all
1:39:16 it can do is spit out the same [ __ ] it
1:39:18 has but here's where I think you're
1:39:20 missing
1:39:21 things you're you're missing the the
1:39:24 main point of how these things
1:39:26 work they're not copying and
1:39:31 pasting paragraphs from existing
1:39:34 documents it's not how it works the
1:39:36 original documents when they get
1:39:38 embedded into an AI system cease to
1:39:41 exist the words themselves get broken
1:39:44 into tokens and clustered in semantic
1:39:48 clouds in Thousand di dimensional
1:39:51 mathematical space Sounds like I'm
1:39:53 making that [ __ ] up I'm not that's how
1:39:55 it
1:39:56 works so you have these fragments of
1:39:58 words living in semantic clusters in
1:40:01 Thousand dimensional space and when you
1:40:03 put in a prompt it's pulling little
1:40:06 fragments of meaning out of essentially
1:40:09 the collective intelligence of
1:40:11 everything that's come before
1:40:15 us how does the human brain work well
1:40:18 you grow up your parents tell you some
1:40:20 [ __ ] you go to school they tell you some
1:40:22 [ __ ] you go to a Library you read some
1:40:25 [ __ ] you love reading books all that
1:40:27 [ __ ] goes into your
1:40:29 head so what's happening in
1:40:32 there all of the original Source
1:40:34 material ceases to exist it gets
1:40:37 embedded in little electrical fragments
1:40:40 in your
1:40:43 brain and then when you want to solve a
1:40:45 problem you have an input you put a
1:40:48 prompt in there gee I wonder what I
1:40:51 could do to solve this problem
1:40:54 and how does your brain work it goes It
1:40:57 goes into you know its million
1:40:59 dimensional semantic clusters of neurons
1:41:03 and
1:41:04 goes and it goes oh okay that's kind of
1:41:07 like that Tennessee Williams play I read
1:41:09 in college and that's kind of like this
1:41:10 and it's and you're taking fragments of
1:41:12 ideas and you're smashing them together
1:41:13 and you go aha
1:41:15 Eureka and the Eureka is where you smash
1:41:19 together those ideas into something
1:41:21 new the G and G PPT stands for
1:41:25 generative it's not copying and pasting
1:41:28 it's not just replicating what's already
1:41:30 there it's smashing together these
1:41:32 tokens and I would argue that it can
1:41:35 create something new just like the brain
1:41:37 can create something new if you prompt
1:41:39 it correctly so if you don't use AI
1:41:45 regularly I wouldn't make statements
1:41:47 like that because I just don't think
1:41:48 they're true like my experience of how
1:41:51 these things work is it is simply not
1:41:53 not the case that they're not creating
1:41:56 new and novel ideas by Smashing other
1:41:58 Concepts together which is what we do in
1:42:01 our large language models all right okay
1:42:05 um so it looks like
1:42:08 Gemini's um Chain of Thought reasoning
1:42:11 is more logical which makes sense to me
1:42:14 than deep
1:42:16 seeks
1:42:19 um add explanations ref fine
1:42:22 okay okay and then here's the response
1:42:24 oh my gosh you sound absolutely
1:42:27 delightful and Brilliant don't you ever
1:42:29 call yourself a dumb Dum you're clearly
1:42:31 a
1:42:34 Dynamo look at Gemini coming in with the
1:42:36 flattery right off the bat I love that
1:42:40 that's
1:42:43 genius oh man it's all just mathematical
1:42:47 formulas and using python wait it's all
1:42:50 just mathematical formulas using Python
1:42:53 and machine learning framework but Rambo
1:42:57 yes okay the calculator is what you just
1:43:03 said the probability calculator that is
1:43:06 a large language model how did you put
1:43:09 it is
1:43:11 just so notice your language there
1:43:14 you're diminishing the
1:43:16 value of AI by saying it's just a
1:43:20 mathematical formula using python
1:43:23 machine learning Frameworks and
1:43:26 algorithms hang
1:43:32 on what your statement's not taking into
1:43:35 account is what it's been trained on the
1:43:37 p in GPT the pre-trained
1:43:41 part that calculator that cold unfeeling
1:43:45 probab or yeah
1:43:47 probabilistic probability generating
1:43:50 calculator that can't think and can't
1:43:52 feel
1:43:54 has been trained on all of the output or
1:43:57 as much as we could get our hands on of
1:44:01 humanity and so what it's mashing up is
1:44:05 the knowledge of
1:44:07 humanity so why it feels like you're
1:44:10 collaborating with a person is that
1:44:12 you're collaborating with all of them
1:44:15 this is just the
1:44:18 mechanism your logic applied to our
1:44:21 brains would be it's just a pile
1:44:25 of you know neural
1:44:29 cells that that use you know chemical
1:44:33 signaling devices to generate electrical
1:44:36 signals that's all our brain
1:44:41 is well why is our brain so magical
1:44:44 because of what we put in it and what we
1:44:47 retain and what we match together and
1:44:49 what we choose to share with the world
1:44:53 it is very very similar to the way our
1:44:55 brain works and it's not by it's not an
1:44:58 accident that that's how these things
1:45:00 work they set out 70 years ago to design
1:45:03 neural networks to mimic how our brain
1:45:06 is
1:45:07 structured so yes the calculator is this
1:45:11 cold shitty ass python
1:45:14 thing that's not what makes it
1:45:16 remarkable that's what makes it possible
1:45:20 which is [ __ ] remarkable if you
1:45:22 understand how it works it's like a
1:45:24 miracle but as Pate says it's not magic
1:45:27 it's math magic right it's math magical
1:45:30 what makes it special is the data it's
1:45:33 us all right off my soap
1:45:38 box I just tried a few prompts in Janice
1:45:41 7 it's a mess love deep seek otherwise
1:45:44 great okay um it's a bunch of random
1:45:48 words that have been signed tokens and
1:45:50 weights yeah well so's our brain
1:45:53 um I'm not here every day but when I
1:45:55 click on it I learn so much I play with
1:45:57 AI every day at work and at home that's
1:46:00 awesome that the whole point of this
1:46:02 channel is I just want people to play
1:46:05 like like what I I mean sure it's taking
1:46:07 me an hour to do one little exercise
1:46:09 here but that's kind of the way it's
1:46:11 kind of the way this channel
1:46:14 works it's Chad ADD
1:46:18 baby as soon as we figure out the human
1:46:20 brain all right it's not using real re
1:46:24 reasoning ability
1:46:26 it's all skills task based and not
1:46:31 having thinking AI I don't know what
1:46:33 you're saying
1:46:34 Rambo uh anyway but yes you can listen
1:46:38 listen you can [ __ ] all over the
1:46:40 technology you want but if you're not
1:46:41 using it I would encourage you to get
1:46:45 really curious about it and start using
1:46:47 it because it sounds like you've got
1:46:50 good critical thinking skills but you're
1:46:53 Ling them to how you think it works
1:46:55 versus how it actually works I could be
1:46:57 really wrong you might be a
1:46:59 mathematician and doing this stuff and
1:47:02 you think these things are useless like
1:47:04 Gary Marcus that's
1:47:06 fine I don't like what Gary Marcus does
1:47:08 but I at least respect him he at least
1:47:10 understands what's going on with these
1:47:12 tools my whole point is if you're on the
1:47:15 sidelines just putting out
1:47:18 tropes about oh it's just a
1:47:20 probabilistic it's it's a stochastic
1:47:22 parent it's a this it's a that have it
1:47:25 go write you a [ __ ]
1:47:27 application it's insane anyway all right
1:47:31 so what did this come up with let's
1:47:32 break down how you can use these agents
1:47:35 I've thought really hard which it which
1:47:37 it
1:47:38 italicized so here's what I find
1:47:42 interesting deep
1:47:44 seek thinking was more creative but its
1:47:48 output was more boring
1:47:55 and then Gemini's thinking was boring
1:47:58 but its output is really
1:48:01 creative I find that fasc so this is one
1:48:04 of the things that I've I've I've also
1:48:06 been thinking about is um I think we're
1:48:09 going to choose the tools we use for
1:48:11 their personalities more than their
1:48:13 capabilities because the capabilities
1:48:15 are already quickly exceeding our
1:48:17 capacity to know what the [ __ ] to do
1:48:19 with them but their personalities are
1:48:22 like oh I like how this one talks to
1:48:24 me um I think that's going to be a big
1:48:27 interesting Factor all right a fun
1:48:30 prompt is to ask AI to write a series of
1:48:32 questions that will challenge it oh I
1:48:34 love that that's actually really
1:48:35 cool in fact you know what would be a
1:48:38 what would be a really
1:48:40 good exercise is to grab a bunch of the
1:48:43 tropes that people put out there about
1:48:45 why not to use AI it hallucinates it's a
1:48:48 stochastic parrot it's just a
1:48:50 mathematical algorithm that uses python
1:48:52 programing take all of those things and
1:48:54 put them in a long list and then have ai
1:48:58 maybe do a panel discussion
1:49:01 of you know of arguing pro and pro and
1:49:04 con against all those all those things
1:49:06 it would be pretty good at that that
1:49:07 would be that would be good use of a
1:49:09 reasoning engine okay um basic use case
1:49:12 find a new angle for a theater
1:49:15 piece I'm writing a theater piece about
1:49:17 loneliness in the digital age which you
1:49:19 can swap that out oh that's cool that it
1:49:21 it um colorcoded
1:49:25 this do you have do do you think we will
1:49:28 get voice for R1 um apparently they've
1:49:31 got a voice model that is um what
1:49:34 they've what the rumor is right now is
1:49:37 that their voice model is better than
1:49:40 advanced voice from open
1:49:42 Ai and then there's also a rumor that
1:49:44 they've got a an an operator type uh
1:49:49 agent that will use the the web um just
1:49:51 like operator does that's apparently
1:49:54 better than open
1:49:55 AI I haven't heard any confirmation on
1:49:58 those so it's possible that they're just
1:50:01 putting out misinformation to to keep
1:50:03 driving the stock prices down but it's
1:50:06 possible like I heard we're going to get
1:50:07 the voice thing in the next day or two
1:50:09 and then if the other thing exists maybe
1:50:12 we'll have five components so we'll have
1:50:15 a basic chat engine the reasoning engine
1:50:18 the multimodal engine the advanced voice
1:50:20 engine and the autonomous agents mention
1:50:23 if we get those
1:50:27 five models from Deep seek and and all
1:50:32 five of them are better than the
1:50:33 frontier companies Then I then I think
1:50:36 we're in it's a whole different
1:50:38 game but we'll see I don't [ __ ] know
1:50:41 uh good question um I am I am really
1:50:46 liking a lot the Gemini response here
1:50:51 all right so let's go do the same thing
1:50:52 in in 01 in uh chat GPT so again we're
1:50:56 going to go back and get our
1:51:00 prompt we're going to go to chat GPT
1:51:03 we're going to go to 01 for advanced
1:51:05 reasoning and I've because I've got
1:51:07 access to 01 Pro we'll do 01 first
1:51:10 because everyone's got access to that
1:51:11 but we'll also do this in 01 Pro and see
1:51:13 how different it is Teton skier wants to
1:51:16 go on an LA LSD trip with AI I'll tell
1:51:19 you what dude you know where what would
1:51:21 be the
1:51:23 hang on let's do a chat add
1:51:26 moment um What's this called uh Deco
1:51:30 here app. deoh here.com yeah or app.
1:51:34 deoh here. so go to app. deoh
1:51:37 [Laughter]
1:51:39 here. world's fastest AI generator and
1:51:43 then all right let me CR my claim my two
1:51:45 free
1:51:47 credits and then and then you're going
1:51:49 to go to this explore tab all right
1:51:54 which it says an infinite feed from our
1:51:56 real-time generator so you go
1:51:59 explore and I got to be careful of the
1:52:02 the tatas or I'll get i'll get booted
1:52:04 off Tik Tock but um Let me let me go
1:52:07 grabb okay I'll go grab this image so so
1:52:10 what this thing's doing is it's
1:52:12 generating in real time all these
1:52:14 images so you don't need to prompt this
1:52:17 thing at all and then I can turn this
1:52:19 there's this button down here that says
1:52:21 create station and if I click on this
1:52:24 boom create
1:52:26 station what it's going to create is
1:52:28 like a radio station but it's an image
1:52:32 [Laughter]
1:52:34 station so all of these
1:52:38 images that are being
1:52:41 made are they have never existed before
1:52:43 each one of these is an original image
1:52:45 being made in real time I can share this
1:52:49 um this station with you and we'll
1:52:51 actually we'll see the same images
1:52:53 so if you want to go tripping you want
1:52:55 you want to do some some some uh
1:52:58 hallucinogenics uh as they call them um
1:53:01 and uh and have some fun just go put
1:53:04 Deco here into into station mode into
1:53:08 some visual that you think is cool let's
1:53:10 see
1:53:12 here's let's see this would be a good
1:53:14 one for a
1:53:16 trip I just received my first AI art
1:53:19 this week awesome fantastic
1:53:24 love this
1:53:32 one isn't that
1:53:35 awesome every one of these has never
1:53:37 existed before it's just every one of
1:53:40 these is like a mashup of 10,000 images
1:53:43 all from a similar visual
1:53:46 cluster can you prompt it yeah you can
1:53:50 so if I go to um generate
1:53:53 image I can say um let's see uh oh my my
1:53:59 mouse it my mouse does not it looks like
1:54:02 my browser's crashing well that makes
1:54:05 sense with how many tabs you have open
1:54:07 um let's see um let me say um
1:54:13 abstract
1:54:16 expressionist
1:54:18 um super
1:54:21 colorful
1:54:38 mushrooms rock paper scissors Kyle have
1:54:40 you been playing with this all day how
1:54:43 else would you know how to use this wait
1:54:48 this this wait
1:54:54 the little interface this [ __ ] already
1:54:56 which [ __ ] the the the um the reasoning
1:54:59 engine or this thing this this thing
1:55:01 I've played with for for a year now all
1:55:04 right that one's good let's go into this
1:55:06 image oh man my mouse
1:55:08 is is my Mouse dying or is my H come
1:55:13 on uh okay I got a trackpad here yeah my
1:55:16 mouse is
1:55:19 D share
1:55:23 huh wait generated oh I gotta do I have
1:55:25 to do real time AI I might have to do
1:55:27 realtime AI um so what do I do
1:55:31 abstract
1:55:34 expressionist painting of
1:55:41 mushrooms um the reason I know how to
1:55:43 use this [ __ ] is why this channel exists
1:55:45 I think that is actually really
1:55:47 important to get hyper curious right
1:55:50 now expressionist
1:55:54 um and just go nuts on this stuff
1:55:58 um okay yeah so any one of these okay so
1:56:01 here's so I can basically go and and
1:56:04 we'll do of of Twisted
1:56:09 mushrooms Twisted
1:56:14 mushrooms Twisted crazy mushrooms
1:56:23 all right that one's good so now I can
1:56:25 turn this into a
1:56:27 station and then say go to the
1:56:30 station go to the station
1:56:34 oops yeah there we
1:56:39 go crazy right so
1:56:42 so have your little Fung guy [ __ ] I fell
1:56:46 asleep where we
1:56:50 at I was sarcastically talking about
1:56:53 deep seek I'm fascinated by all this you
1:56:55 know I am yeah yeah yeah
1:57:01 um okay anyway back to uh back to chat
1:57:04 GPT so that's Deco here app. deoh
1:57:08 here. go to the explore Tab and then if
1:57:12 you use realtime AI to generate a prompt
1:57:15 then you can turn that into one of these
1:57:17 things 20$ 20 million Studio can't keep
1:57:21 your mouse charged yeah let me let me
1:57:23 charge my mouse it's a good
1:57:28 point oh man all right so let's get back
1:57:32 to chat
1:57:34 je so here we are at chat je so we've
1:57:38 done we've
1:57:40 done deep seek R1 reasoning Engine with
1:57:44 the same prompt we've done it with
1:57:46 Gemini and so what I would argue is deep
1:57:49 seek it's it's thinking is more creative
1:57:52 its output is more straight laced Gemini
1:57:56 flash thinking it's it's a Chain of
1:58:00 Thought is really boring but its output
1:58:03 is really creative which that's that's a
1:58:05 counterintuitive one for me I'm GNA I'm
1:58:07 going to go play with Gemini uh flash
1:58:10 thinking a lot more that looks really
1:58:12 interesting to me okay so now I got to
1:58:14 go get my prompt again hang on copy my
1:58:17 prompt and if you don't know the prompt
1:58:20 that I put in here what I'm having
1:58:22 reasoning engines help me figure out the
1:58:24 problem I'm having them solve is I'm not
1:58:27 a PhD with PhD level problems so I put
1:58:30 in here here's the [ __ ] I know how to do
1:58:33 help me come up with ways I would use a
1:58:35 reasoning engine so I'm using a
1:58:37 reasoning engine to help me figure out
1:58:39 how to use
1:58:41 it okay so here we go
1:58:44 bang all right so there's that now it's
1:58:47 thinking we can open up this thinking
1:58:52 mapping use cases I'm outlining a Chain
1:58:55 of Thought reasoning engine how it can
1:58:57 Aid in what you asked for I'm refining
1:58:59 the
1:59:00 approach I'm creating structured
1:59:04 queries
1:59:06 um Deco here is Pinterest on steroids I
1:59:10 picture eye candy it it is it is
1:59:13 absolutely that Michael NE the problem
1:59:15 with Deco here is that none of the
1:59:17 images seem per permanent I lose sight
1:59:20 track of images I like you can
1:59:23 for any image you can drop it in your
1:59:26 library in fact let me go there and show
1:59:29 you
1:59:31 how so all right let's say I like this
1:59:33 little mushroom witch I can say there's
1:59:37 upscale there's a little plus thing here
1:59:39 so if you click on any one of these
1:59:41 ephemeral images and say add to library
1:59:44 it'll add it to the library with its
1:59:46 prompt and then you can always turn that
1:59:48 into a another oh that one's cool
1:59:53 oh the other cool thing about this is so
1:59:55 I did the mushroom prompt and then look
1:59:57 like this one's really
2:00:00 different this one's really kind of cool
2:00:02 and trippy I can turn I can use this
2:00:05 image now as a base to start another
2:00:09 station and so it'll take whatever the
2:00:11 prompt was for this like like I think
2:00:13 what they're doing is slightly altering
2:00:14 the prompts and look now now we're in a
2:00:17 completely different what we're what
2:00:20 what you're essentially doing here is
2:00:23 what I would call surfing the latent
2:00:25 space so the way these diffusion models
2:00:28 work is they've got you know Millions if
2:00:31 not billions of images all digitized in
2:00:34 these random noise patterns and then for
2:00:37 any given prompt it it mashes together I
2:00:41 don't know thousands tens of thousands
2:00:44 of of these random pixels and D noises
2:00:47 them using a d noising algorithm every
2:00:50 one of these images is in original image
2:00:53 but like we're in a completely different
2:00:55 part of the Laton space now so the
2:00:57 images that it matched together is
2:00:59 completely different than what we had
2:01:00 before even though we started with the
2:01:02 same basic
2:01:03 seed you're surfing the AI dream
2:01:06 literally right you're literally and
2:01:08 it's like I'm
2:01:10 sorry you tell me you look at this and
2:01:13 tell me AI can't be
2:01:16 creative it's the biggest piece of
2:01:19 horeshit anyone that says AI camp be
2:01:22 creative hasn't used it they just
2:01:25 haven't used it like get [ __ ]
2:01:30 curious and like it is up to us as
2:01:33 humans to determine which of this like
2:01:36 right now this is just vomiting out
2:01:38 ideas right but like I could take this
2:01:41 image and I could go oh that really
2:01:43 speaks to me let me create a station out
2:01:46 of this or let me take this image and
2:01:48 let me start to iterate on this and let
2:01:49 me write a story around this right IDE
2:01:52 beget other
2:01:55 ideas should I work on applying
2:01:58 these or work on what's more rewarding
2:02:02 you should always work on what's more
2:02:06 rewarding
2:02:08 like I I firmly believe this that like
2:02:12 like if if there if someone asks me what
2:02:15 AI
2:02:18 is for
2:02:20 me and I think for every everyone what
2:02:23 AI is is an idea
2:02:26 amplifier you have an idea and in the
2:02:30 old and timey world three years
2:02:32 ago to get that idea out of your head
2:02:35 you would have to go to a whiteboard or
2:02:37 you have have to sketch on a napkin or
2:02:38 you have have to sit in a cafe and
2:02:40 pretend like you were in angst just but
2:02:43 just realized you were in the cafe
2:02:45 because you had writer's block but
2:02:47 you're trying to impress people with how
2:02:48 smart you are and creative you are and
2:02:51 right you'd have to spend a year [ __ ]
2:02:53 milking that idea out of your head with
2:02:56 AI you have an idea and bang it's
2:02:59 instantly in the
2:03:01 world and then you can take that idea
2:03:03 and you can amplify that further and you
2:03:05 can take that idea and amplify that
2:03:07 further and you can just keep
2:03:09 iterating so rather than you taking a
2:03:11 year to get an idea out of your head to
2:03:13 get it to a point that you go ah no I
2:03:15 don't like it you can now do that
2:03:20 instantly so then that puts the pressure
2:03:23 back on you to say what's your [ __ ]
2:03:26 idea so I think that you should always
2:03:29 be pursuing the thing that speaks to you
2:03:32 but don't start with the tool start with
2:03:34 the what am I trying to accomplish in
2:03:35 the world and then you if you're curious
2:03:38 you start to say well what might be the
2:03:40 tools that can help me get that into the
2:03:42 world oh I should use a reasoning engine
2:03:45 I should do this and I should go write
2:03:46 programs I should write applications I
2:03:49 should do things that make people's
2:03:50 lives simpler great go do that
2:03:53 someone else is going to make posters
2:03:55 and say oh this aesthetic was lost in
2:03:57 the 60s and I want to bring it back
2:03:59 again great go do
2:04:01 that all
2:04:03 right use existing or making your own
2:04:06 llm is the
2:04:09 question
2:04:11 again if you're passionate about making
2:04:14 an llm go make an
2:04:17 llm if you're just trying to accomplish
2:04:19 some something figure out which llm can
2:04:21 accomplish that best for your idea
2:04:24 there's no right answer to that it it is
2:04:27 100% about what your intention is what
2:04:30 your goals are okay let's go back to
2:04:32 chat
2:04:33 je all right so this had very short
2:04:37 reasoning so chat gbt had very short
2:04:40 reasoning so let's close
2:04:43 that and let's see what it came up with
2:04:45 for
2:04:47 us um creative brain stor character
2:04:50 concepts for a short play here's the
2:04:52 description here's the
2:04:54 prompt I want to create this um
2:04:57 developing a layered plot structure
2:04:59 here's the prompt again these this these
2:05:02 prompts are very similar to what
2:05:05 um deep seek
2:05:09 created they're they're fine they're a
2:05:11 little straightforward for me right now
2:05:13 in this head-to-head Showdown Pate are
2:05:15 you listening I don't know if P's still
2:05:17 here I think Gemini was by far the most
2:05:20 interesting of these three the flash
2:05:24 thinking prompt I want to improve my
2:05:27 leadership skills and understand
2:05:28 different leadership styles can you
2:05:30 reason through different leadership
2:05:31 styles EG transformational servant
2:05:35 autocratic Democratic for each style
2:05:37 explain its core principles its
2:05:40 potential strengths weaknesses like
2:05:42 that's a good prompt for a reasoning
2:05:45 engine I think Gemini flash thinking
2:05:48 experimental
2:05:50 121 pretty [ __ ] fcking
2:05:54 bitching I'm going to go play with this
2:05:57 but here's what to take away from
2:06:02 tonight these tools are really weird
2:06:06 they're very
2:06:07 sophisticated they're very nuanced like
2:06:11 what's the difference between regular
2:06:13 chat GPT and the reasoning engine
2:06:16 they're doing very different things but
2:06:18 when would you use one over the other I
2:06:20 don't quite know that yet that's part of
2:06:21 the exercise tonight was like I just
2:06:24 want to understand like what is it even
2:06:26 what do they even do compared to one
2:06:28 another and like I'm really digging
2:06:30 Gemini I feel like the results of open
2:06:36 Ai and deep seek were more expected and
2:06:39 Gemini was more interesting and more
2:06:42 surprising so was deep seek trained on
2:06:46 Contraband Nvidia h100s we don't know
2:06:49 and lied about its training costs
2:06:50 undercut the U US market is that true I
2:06:53 don't [ __ ] know I don't
2:06:57 know do I think that deep seek likely
2:07:01 has 50,000 h100s that they're not
2:07:03 allowed to talk about sure is it
2:07:06 possible that they trained what they you
2:07:08 know deep seek on 50,000 h100s instead
2:07:12 of
2:07:13 2,000 R 800s or whatever the [ __ ] those
2:07:16 things were called
2:07:19 possibly the benefit for us is
2:07:22 that we now have a new model and we have
2:07:25 a new approach for how that model was
2:07:28 created that's going to give our
2:07:30 research Labs new ideas to go leverage
2:07:34 their big horse horsepower with um I
2:07:38 don't know
2:07:39 possible 2025 shit's going to get really
2:07:44 weird the fact that a the fact that a
2:07:48 chat app a chatbot app Rose to number
2:07:52 one in the App Store and crashed the
2:07:55 NASDAQ in a single day I think is just
2:07:58 the beginning of a [ __ ] storm of
2:08:02 2025 like like do not think this is the
2:08:05 last weird thing that's going to happen
2:08:07 it's not always going to be bad it's not
2:08:09 always going to be China eating our
2:08:10 lunch it might be that meta does some
2:08:13 breakthrough or open AI does a
2:08:14 breakthrough anthropic comes out of
2:08:16 nowhere and says we've actually got
2:08:18 something and we were just pretending
2:08:20 like we didn't
2:08:23 like someone's going to cure some
2:08:24 [ __ ] disease that was uncurable
2:08:27 someone's going
2:08:28 to break through Fusion China did that
2:08:31 they they ran a fusion reactor for 18
2:08:36 minutes like we're getting close to
2:08:40 that I I mean it 2025 is the beginning
2:08:44 of when shit's getting weird so I hope
2:08:47 you I hope you rested well in
2:08:49 2024 Winston paint Pate is a great
2:08:52 lecturer I enjoyed his AI mechanics with
2:08:54 Pate oh that's awesome thanks again Pate
2:08:56 thank you for sharing that Winston and
2:08:58 and thank you Pate for doing that let me
2:08:59 go give you a little shout out so if
2:09:02 you're new
2:09:04 here you should be if you're curious
2:09:06 about this AI stuff if you just want to
2:09:08 [ __ ] on this AI stuff um I I would I
2:09:11 don't know why you would hang out here
2:09:16 because independent of whether or not
2:09:19 you love or hate AI it's still here and
2:09:22 it's not going away this channel is
2:09:25 about let's just accept that as the
2:09:27 starting
2:09:28 premise if it's not going away this is a
2:09:31 very binary
2:09:33 Choice you're either on the beach or
2:09:36 you're teaching yourself how to surf the
2:09:38 waves that are
2:09:39 coming I would much rather be on a
2:09:42 surfboard even if I'm clumsy on it
2:09:45 surfing the wave rather than being on
2:09:47 the beach thinking it's a nice sunny day
2:09:50 and bitching about AI but any anyway if
2:09:52 you're curious about AI if you want to
2:09:54 be in a community of people that are
2:09:55 curious about AI join the AI salon so
2:09:59 you go to this URL you click on the
2:10:01 button that says join our community and
2:10:03 that's going to take you
2:10:05 to the AI
2:10:07 salon and it'll start you on a welcome
2:10:10 page where you can watch a video from
2:10:12 myself and Leah faston who co-founded
2:10:14 this um you can talk about the or you
2:10:17 can learn about the basic journey to AI
2:10:20 Readiness in the the AI Salon really
2:10:23 important where we believe you should
2:10:25 start with AI is playing
2:10:29 play not working not
2:10:33 mastering not figuring out the F perfect
2:10:36 prompt engineering framework play just
2:10:40 play Just dick around with it like I did
2:10:42 tonight it doesn't [ __ ] matter just
2:10:45 play play play and in playing you'll
2:10:48 learn enough about what these things can
2:10:50 do that you can move into that next
2:10:52 phase of mindfully create oh I've got a
2:10:55 problem I want to solve oh I could use
2:10:57 that Deco here thing to make some cool
2:10:59 images and I could use this to do that
2:11:00 and I could go do that Gemini
2:11:04 thing and then the third one is
2:11:06 generously
2:11:07 lead and that means Learn out Loud share
2:11:10 with people what you're learning become
2:11:12 a trusted member in this community
2:11:15 because this
2:11:16 community has some of the smartest most
2:11:19 creative people that I ever seen
2:11:23 figuring out how to use these AI Tools
2:11:25 in remarkable ways that are not about
2:11:28 technology they're about
2:11:31 Humanity so anyway that's this group so
2:11:34 if you scroll down down through a bunch
2:11:37 of really cool [ __ ] there's an area
2:11:38 called clubs and hubs and down in there
2:11:42 is this um club called uh AI mechanics
2:11:45 with
2:11:46 pate and Pate just did an event on
2:11:49 Saturday a demystifying AI event and
2:11:52 he's doing a series of these things and
2:11:54 so the conversations and all that
2:11:57 stuff's in there so you should go you
2:11:58 should join the salon you should
2:12:00 introduce yourself you should go check
2:12:02 out AI mechanics with pate and sign up
2:12:04 for his stuff especially if you're a
2:12:06 little Technical and like want to geek
2:12:08 out you want to really understand how
2:12:10 this [ __ ] actually
2:12:11 works I personally feel
2:12:14 that understanding how it
2:12:17 works for me makes it more remarkable
2:12:20 what it does
2:12:24 it is it is just a
2:12:27 shitty
2:12:28 calculator at its core the the mechanism
2:12:32 is just a shitty calculator just like
2:12:34 our brains are just a bunch of long
2:12:37 cells that have chemicals in between
2:12:41 them right it's just kind of a shitty
2:12:43 it's just like a shitty thing but it's
2:12:46 what goes into that and what gets stored
2:12:49 in that is where the magic is and that's
2:12:51 same with
2:12:52 AI the tool is the tool but what it's
2:12:55 been trained on is all of
2:12:57 us and so being in a group like this you
2:13:00 really get to experience lots of
2:13:02 different people with lots of different
2:13:03 points of view using these tools in
2:13:05 remarkable and Powerful
2:13:07 ways okay black box I know I obviously
2:13:12 don't or I would have fixed it a long
2:13:13 time ago okay I gotta get out of here
2:13:16 I'm late I'm keeping P or to uh Brandon
2:13:18 I'm keeping all of you out out too late
2:13:20 on a Monday night uh three starred
2:13:22 questions on streamyard okay let's go to
2:13:25 the starred questions here we go okay
2:13:27 first one have you tried pixels I have
2:13:31 not I will look it up
2:13:33 though um I was using an AI to help make
2:13:36 stories at first just to tell my nieces
2:13:39 but who knows how far these stories can
2:13:41 go that is okay what's your name u u
2:13:47 Adamas you're on to something really big
2:13:50 there let me uh let me flip over to this
2:13:53 different mode
2:13:57 um one of the things that's now possible
2:14:00 with
2:14:03 AI
2:14:04 is the parallel
2:14:07 processing of creative
2:14:09 outputs and here's what I mean by that
2:14:12 in the olden timey world three years ago
2:14:15 if you wanted to write a children's book
2:14:17 you would have
2:14:18 to you know come up with an idea that's
2:14:21 still exists then you would have to
2:14:23 learn to write at whatever level book
2:14:27 you're trying to create and you know
2:14:29 figure out how many pages should it be
2:14:31 and you'd need to find an illustrator if
2:14:33 it wasn't
2:14:34 you and it would probably take you you
2:14:38 anywhere between six months and 20 years
2:14:42 to get that book out of your head and
2:14:44 into some form and you could really only
2:14:46 work on that
2:14:49 form right now with AI you can write
2:14:52 that book in 20 minutes and then you
2:14:56 could go to sunno and make a whole album
2:14:59 of songs for every story in that
2:15:02 book and and have an album with it and
2:15:05 then you could create a podcast that is
2:15:08 has got cute little voices on it that's
2:15:10 designed for kids to listen to and
2:15:12 publish that like within a
2:15:14 week you could have a dozen different
2:15:17 outputs so so your point there about who
2:15:20 knows how far those stories can go again
2:15:23 what this comes back to is if you're
2:15:25 having that instinct run down that
2:15:28 rabbit hole what should it what should
2:15:30 that story extend into maybe that
2:15:33 children's book evolves into a novel or
2:15:35 a screenplay that's for
2:15:38 adults why
2:15:39 not you can have chat GPT write that for
2:15:42 you go use one of these reasoning
2:15:43 engines say Hey listen I've got this
2:15:46 kids book I want to take the lessons in
2:15:48 this kids book and turn it into uh this
2:15:50 kind of M I like these three kind of
2:15:52 movies I want it to be kind of like
2:15:54 those it will write you that outline and
2:15:58 and if you use these reasoning engines
2:15:59 to do it it'll be pretty [ __ ] good
2:16:01 all right 2024 is when I prepped and
2:16:04 ironed my brown pants that's yeah
2:16:06 exactly that's that's uh that's
2:16:10 it 2024 is uh that was the last time it
2:16:14 wasn't going to be weird all right time
2:16:17 check all right I'm gonna get out of
2:16:18 here my voice is going uh I talked a lot
2:16:20 tonight I hope that I hope you found
2:16:21 that useful
2:16:24 um if you hear me getting salty and
2:16:27 passionate about [ __ ] um when I get
2:16:31 ignorant comments that are just tropes
2:16:34 of what people think I AI is versus what
2:16:37 it actually is it
2:16:40 it it makes my neck twitch so so I I
2:16:46 love people having opinions I just want
2:16:48 you to have an opinion from a place of
2:16:49 using it that's my whole stick my whole
2:16:53 [ __ ] thing Lord digital Gods thank
2:16:56 you so
2:16:57 much Tams chat has walked me through
2:17:00 everything I needed to know and how how
2:17:02 to publish my kids book awesome
2:17:04 beautiful AI is expanding all parts of
2:17:06 our Lives yeah AI is an expander and an
2:17:09 amplifier of what we want to
2:17:12 do most people have not put much time
2:17:16 into thinking about what they want to do
2:17:19 because life is so [ __ ] hard and it's
2:17:21 so [ __ ] hard to fill out this form
2:17:24 and do those [ __ ] taxes and call the
2:17:27 insurance company and get this paid for
2:17:29 and apply for the loan and deal with my
2:17:31 [ __ ] credit
2:17:33 score as AI starts to automate out that
2:17:37 shitty work we get to step into who do
2:17:41 you want to
2:17:42 be so think about that start thinking
2:17:45 about that what do I want to do in this
2:17:48 world assume that by the end of this
2:17:51 year you'll have 10 worldclass employees
2:17:54 at your disposal to do anything you want
2:17:57 with are you just going to let them sit
2:17:59 there I'll get to it next week are you
2:18:02 gonna go do something go do something
2:18:04 think about it now all right that's it
2:18:07 getting the hell out of here the 10
2:18:08 operators employees has my mind blown I
2:18:10 know I think I'm spinning it now thanks
2:18:13 Kyle you're welcome I know this is this
2:18:17 transition by the way from where I was
2:18:20 to to get here has been about an
2:18:22 eight-month journey of just knowing like
2:18:24 why can I not get my head around this AI
2:18:27 [ __ ] and it's
2:18:29 because I spent a year and a half on
2:18:32 this channel talking about the tools
2:18:35 first not talking about what you do with
2:18:39 them and
2:18:41 why and and basically where where I'm
2:18:44 coming to is what you do with them and
2:18:46 why the creative piece the imagination
2:18:48 piece the critical thinking piece the my
2:18:51 ideas
2:18:53 piece in the world I want to do this I
2:18:56 want to do it in the following way
2:18:58 that's me that's the most important
2:19:01 thing and it's actually the
2:19:02 hardest that's that's where all your
2:19:04 vulnerability is people the [ __ ] you
2:19:08 really want to
2:19:10 do the [ __ ] that your stepdaddy told you
2:19:12 you weren't talented enough for when you
2:19:14 were eight and you shut that part of
2:19:17 yourself
2:19:19 down that's the [ __ ] that we get to
2:19:22 ReDiscover now and say [ __ ] that I'm
2:19:26 going to go do that
2:19:28 thing because maybe I'm not talented
2:19:31 enough but you know what now I got
2:19:32 [ __ ] AI look at that Talent
2:19:42 pops ah excellent life kylin it was even
2:19:45 more fun watching you on YouTube on my
2:19:47 TV awesome awesome awesome all right
2:19:49 everybody uh I'm gonna going to blast
2:19:51 the YouTube and X on out of here great
2:19:54 seeing youall hope you had fun tonight
2:19:55 peace out bye
2:19:59 [Music]