Feb 22, 2024 - AI Revolution: Exploring the Future of Generative Technology and Creativity

Description

Chapters

Transcript

0:01 [Music]
0:17 looking at the
0:19 world
0:21 telescope wish I wish
0:29 I
0:31 [Music]
0:35 looking at this blue
0:39 telescope down a moon struck a road
0:44 [Music]
0:58 tonight thank you Danielle thank you
1:00 Robert
1:01 [Music]
1:29 Rossy
1:35 [Music]
1:40 I didn't mean to cause you on a
1:43 [Music]
1:45 sorrow didn't mean to cause you in a
1:49 [Music]
1:54 pain only one want time to see you
1:58 laughing thank you robt
2:01 Ry only want to see a laugh in that bule
2:06 rain purle rain purle
2:12 rain Purple Rain Purple
2:17 [Music]
2:19 Rain Purple Rain Purple
2:23 [Music]
2:29 Rain want to
2:31 see
2:36 [Music]
2:47 in that's actually in that purple rain
2:51 thing there's remember the opening of
2:52 Twin
2:53 [Music]
2:56 Peaks huh that's the Twin Peaks opening
3:01 [Music]
3:17 [Music]
3:23 [Laughter]
3:23 [Music]
3:26 welcome everybody my name is Kyle
3:27 Shannon this is the AI learning
3:29 lab the lead singer there is
3:32 [Music]
3:34 Champ he likes to open up the shows
3:38 [Music]
3:58 here well you done don't me you better I
4:01 felt it tried to beat you after the M
4:05 fell out through the
4:07 cracks now I'm trying to get back before
4:12 the cool R out I'll be giving in my
4:14 business nothing going to stop on the v
4:18 w again my turn to win some or some but
4:24 I
4:26 won't take no more no more
4:33 it I'm
4:35 [Music]
4:49 your all
4:51 right enough mediocre
4:53 [Music]
4:55 singing I know I probably shouldn't call
4:57 champ mediocre
4:59 right
5:01 I think if I put champ down people are
5:04 going to come after me we love champ why
5:06 would you put him down he's a lovely
5:08 singer he's clear he's clearly got
5:11 classical
5:15 training I'll tell you one thing I do
5:17 know champ he sings from the diaphragm
5:20 like a good like a good singer
5:23 should
5:27 deep all right so
5:31 welcome to the AI learning lab if you
5:33 have questions about AI pop them in the
5:35 comments below they're
5:39 okay I said it in January I said it a
5:43 lot in January shit's going to get weird
5:45 in 20124 and
5:48 then January was not weird it was just
5:51 like yeah it's like chat GPT still kind
5:53 of sucks and Google hasn't launched
5:56 their
5:57 [ __ ] and then February came and
6:00 it was still kind of
6:03 nothing and like Google launched their
6:06 Gemini Ultra and it was like a piece of
6:08 [ __ ] was like yeah whatever it's all
6:10 right it's okay not great not what they
6:14 said and then in the past week holy
6:20 [ __ ] you want to see something [ __ ] up
6:23 well it's not [ __ ] up to you cuz
6:24 you're normal people but when you when
6:27 you got the neuro Divergence when you're
6:29 neuros spicy my own
6:33 [Music]
6:36 self before I went live tonight I wrote
6:39 down that many
6:42 things this many things happened this
6:45 week that I think are kind of batshit
6:46 crazy and
6:50 big some are bigger than
6:53 others and they're not really in
6:56 order but I figure I figure we'll like
6:59 we'll look at some of this [ __ ] tonight
7:01 so we had Sora right the open AI video
7:06 model like oh
7:07 [ __ ] I guess these things are doing
7:10 video now like making like really
7:14 good nearly can't tell it from reality
7:17 video I guess we're there now now it's
7:20 not launched yet so we're not there
7:24 there but we're
7:26 there 11 Labs then said oh we've got a
7:31 um oh there's another one here wait let
7:34 me put another one
7:39 here then 11 Labs did the text to sound
7:43 effects thing where they put the audio
7:47 behind the open AI Sora video which was
7:52 really
7:53 interesting then today uh 11 Labs has
7:57 been accepted into Disney's accelerator
8:00 which is really interesting because if
8:02 they're doing voice
8:04 synthesis what does Disney have lots of
8:06 lots of intellectual property would say
8:08 Disney
8:10 characters and 11 Labs is also
8:13 behind um oh someone said no for
8:19 what 11 Labs cloning your voice yeah
8:22 exactly 11 Labs is also the voice Gemini
8:25 is a train wreck I know it really is oh
8:28 they also you know how I was bitching
8:29 about
8:31 you can't get Gemini if you're if you
8:33 have a work group account now you can so
8:38 either for $20 or $30 you can add Gemini
8:42 to the thing you're already paying
8:43 for and then at some point they say
8:47 we're going to give you access to Gemini
8:49 [Laughter]
8:56 Ultra so 11 Labs is now part of the
8:59 Disney Disney accelerator like and 11
9:02 Labs is the one that sits behind Heen
9:04 right the H Heen video translation stuff
9:08 that's that's 11 Labs sitting behind
9:12 that
9:14 then just because we [ __ ] all over
9:16 Google let's unshit on them a week after
9:20 they
9:21 launch Gemini
9:25 Ultra they announce and give limited
9:28 access to a thing called Gemini Pro
9:31 1.5 which I guess is their middle model
9:34 but it's the 1.5 version of it and it's
9:36 got a million token context window which
9:39 you're like I don't know what that
9:41 means it means you can upload like six
9:45 complete white papers and it just will
9:48 understand them all or you can up it's
9:51 multimodal you can upload an hour of
9:53 video Ethan mullock today said he did a
9:56 screen recording of himself reading a
9:58 bunch of articles and uploaded that
10:01 video to to Gemini and it just
10:04 understood all of what he did There's he
10:07 he also uh
10:09 uploaded him
10:10 working and Gemini understood what he
10:13 was doing and suggested
10:18 improvements like Ethan mullik so if you
10:20 don't follow Ethan mullik he's a Wharton
10:22 business school Professor you should
10:23 follow him he writes like I don't know
10:26 once or twice a day for the past
10:30 three four days he's been writing like
10:33 seven eight 10 times a day all about
10:36 Gemini
10:37 1.5 and how a million token context
10:40 window changes everything everything's
10:41 different this is a guy that's been
10:43 paying attention to this [ __ ] every
10:45 single day for
10:49 years he's like shit's
10:52 different Deco here which is a tool that
10:55 we've played with in here before that
10:57 makes quick quick images it's using SD
10:59 XL turbo and they added a thing today
11:03 called a reference photo which I'll show
11:05 you later where you just upload a
11:06 picture of yourself and it's like
11:07 instant AI of
11:09 you it's like holy
11:14 [ __ ]
11:15 um Mid journey is apparently in talks
11:18 with
11:20 x.com
11:22 about getting getting
11:25 hitched million million tokens is how
11:28 many pot how many Harry Potter books I
11:31 think it's probably all of
11:35 them I think it's probably all of
11:40 them
11:42 Nvidia announced their quarterly
11:44 earnings and they dramatically exceeded
11:48 the
11:49 market when did we get oh wait when did
11:52 this start uh 8:30 I just started I'm
11:54 just I'm just reading through a list of
11:57 [ __ ] that happened I haven't really
11:58 started yet
12:00 um the vision The Vision Pro goggles
12:03 like people are just continue to lose
12:05 their mind over them like that that
12:07 wasn't just a blip of Apple Fanboys
12:09 going this is good there's been like a
12:11 sustained two we holy [ __ ] Vision Pro
12:15 changes things and then you had Mark
12:17 Zuckerberg in there going you know I
12:19 think that uh I think that the uh the
12:22 the Oculus Rift is uh is every bit as
12:25 good as the Vision Pro and it's 17th the
12:27 priz and you know I knew it would be
12:29 better value I just didn't think it
12:31 would be a better product but but it is
12:34 our product's so much better cuz it's
12:36 made with plastic and there's has some
12:38 sort of see-through glass and you know
12:40 it's it's just plastic so much better so
12:43 that
12:44 happened then grock not that grock not
12:48 Elon Musk grock the grock with a Q not
12:51 with a
12:53 K grock is a chip manufacturer and they
12:57 manufacture some AI specific chips
13:00 and you can go to gro.com
13:02 gq.com
13:04 and go type a prompt in there go do it
13:06 right now 500 tokens a second it's like
13:10 350 words a second it just goes like it
13:13 just like vomits out answers at you like
13:16 bang you're like oh okay so not only are
13:21 these tools going to be able to answer
13:23 anything they're going to be able to
13:24 answer them like
13:26 that and then right before the last
13:29 thing I just remembered is uh
13:31 reddit reddit is announcing their
13:35 IPO so holy
13:38 [ __ ] and it was it was um Matt wolf
13:41 wrote today he goes wasn't it last March
13:43 when like all of the companies launched
13:46 all of the [ __ ] that blew all of our
13:48 minds I have a feeling we're about to
13:50 enter funny
13:54 season end of the first
13:58 quarter
14:00 shit's getting good but shit's getting
14:03 weird and it's it's getting chaotic it's
14:05 getting really chaotic it going get more
14:09 chaotic have you embraced the
14:14 Jank have you embraced the Jank
14:17 everything in AI right now is janky
14:19 learn to love it because it's not going
14:21 to be here that long you can make money
14:23 with we're going to look back at 2023
14:26 and go HH it was so sweet when the AI
14:28 was slow and made mistakes
14:32 and the the images had had personality
14:36 cuz they were
14:38 janky it's going to get weird
14:42 people it's going to get
14:45 weird no I know I know it's 2024 20 I'm
14:49 just saying 2023 was like normal 2023
14:52 the last normal
14:56 year
14:58 2024 it started starting to
15:00 accelerate so we'll go we'll go mid mid
15:04 February is when the uh when the Rockets
15:07 kicked in my feet are cold my feet are
15:10 cold Listen People my feet are cold I
15:13 got to go deal with that okay okay I'll
15:17 be back I'm just going to go put on shoy
15:19 shoes and then we'll be back and we'll
15:20 do something big news today Google took
15:22 down its own image
15:24 generator yeah I saw that did they
15:26 actually do that also AT&T went down
15:28 today and they're trying to figure out
15:29 if it was a
15:31 hack cuz apparently well I can't say it
15:34 on here some people have infiltrated our
15:38 infrastructure more than the government
15:40 suspected let me get shoes hang on I'll
15:42 be back talk about
15:58 yourselves
16:28 e
16:32 I I'm afraid my rabbit will be [ __ ] by
16:34 the time I get it I know me too my
16:36 rabbit the rabbit phase one is supposed
16:39 to come
16:40 out mid-march which probably means mid
16:46 April their discord's been really slow
16:49 too I actually bitched about it I went
16:50 in their Discord and I'm like does
16:52 anyone that works at rabbit is anyone
16:54 still posting in here
16:56 like someone from rabbit was like oh
16:58 we're still here we're still
17:00 [Laughter]
17:07 here man all
17:10 right what are we gonna do what do you
17:13 guys want to do what questions do you
17:14 have let me let me let me see what's
17:15 going on let me see what's going on in
17:18 here are we full of trolls tonight are
17:20 we full of full of lovers we got empathy
17:23 tonight oh we got hate and rage and
17:27 bitterness which one is it What's it
17:29 gonna be
17:31 tonight embrace the
17:36 Jank um is the Google is the Google
17:39 thing true did Google really pull down
17:42 their image
17:45 generator let me go to let me go to
17:47 Gemini Ultra because I have access to
17:50 ultra
17:58 Gemini yeah see
18:01 yeah hello Kyle how can I help you today
18:05 make a
18:07 picture of janky
18:10 AI
18:16 [Music]
18:22 technology doing
18:24 something I got twirling Stars it did
18:28 it all right let me go to non
18:33 Ultra let me go to Just Gemini little
18:36 old Gemini and this isn't 1.5 by the way
18:39 whoever told me that last night was
18:41 lying lying liing a cheat uh make an
18:46 image of janky AI
18:58 technology I'm still learning to create
19:00 images so I can't help you with that yet
19:03 I'm sorry Dave I'm afraid I can't do
19:06 that all right if we're good at chat GPT
19:10 do we really need things like jasper. a
19:13 that is an excellent question Teresa um
19:17 technically
19:19 no Jasper AI um copy AI tools like
19:27 that are basically just pre-written
19:30 prompts sitting on top of chat GPT or on
19:33 top of GPT 4 or three whatever they're
19:35 using probably
19:37 four if they're doing copy um you could
19:42 technically go do all that [ __ ] yourself
19:44 right just learn learn to
19:46 prompt the the problem with sites like
19:49 Jasper and copy AI is that they start
19:52 out with like oh you know we should do
19:54 10 things like okay we'll do a short
19:56 paragraph blog post and then we'll do a
19:59 tweet and then we'll do this and then
20:00 we'll do that and then before they know
20:02 it they've got 8,000 buttons and options
20:05 and things that are all prompts and sub
20:08 prompts and things like that and it's
20:10 like it takes you as long to learn the
20:12 interface as it would just to go to chat
20:14 GPT and ask it to write you that [ __ ] so
20:16 you're absolutely correct ah LinkedIn
20:19 how do I get to your office hours okay
20:21 so
20:22 tomorrow if you want to hang out in a in
20:26 a more like just hangy out session I do
20:29 a thing called AI office hours at 11:00
20:32 a.m. mountain is that right
20:36 yeah 10 a.m. Pacific 1 p.m. Eastern on
20:40 LinkedIn and so how you get to it is you
20:44 go to
20:49 my you go to my profile so I'm at Kyle
20:53 Shannon on
20:56 LinkedIn and then
21:03 are there
21:12 events
21:19 huh
21:22 resources show all five resources I
21:25 don't know where events are on LinkedIn
21:27 like when I go to my profile
21:31 hang
21:32 on so if I just go to
21:38 LinkedIn just look up AI office hours
21:42 and Kyle Shannon you should find it and
21:44 I don't have tomorrow's up I'll put it
21:46 up in the
21:47 morning but but you should be able to
21:54 see so here uh oh here's the event link
22:02 where can I put this I could put this oh
22:06 you know where I'll put
22:07 this I'll put this
22:11 in the Irregulars channel in the AI
22:15 salon and
22:17 then let's
22:20 see I'll put it in the
22:24 chat all right so
22:28 tomorrow
22:29 AI uh office
22:33 hours on
22:37 LinkedIn 11
22:40 a.m.
22:43 MST and then I'll put the uh the that
22:48 thing there you go so it's
22:54 there there listen there are much more
22:57 professional ways to do that like I
22:59 could have that thing up for more than
23:01 an hour in the morning on a Friday
23:06 but and welcome to Chad
23:10 ad I got organizational skills can I
23:13 sell you some organizational skills you
23:15 can make money with Chi I'll teach you
23:17 how to organize a chat chippy you just
23:19 give me uh uh $49.95 402 79.95 for the
23:25 first two hour one hour two hours
23:32 yeah yeah you understand what I'm saying
23:35 all right a friend called and said he
23:37 finally started using chat after months
23:40 of me prompting no pun intended he's
23:44 hooked every [ __ ] skeptic I know
23:47 every [ __ ] skeptic I know with the
23:51 exception of Eric marier he's sticking
23:52 to his guns cuz he's using chat GPT and
23:55 he still hates it I still am not
23:57 convinced he's using it right though
24:00 he's bitter about it and I like that I
24:02 appreciate that in him at least he's
24:04 [ __ ] using it right um every skeptic
24:08 I know when they actually use this [ __ ]
24:11 they're like oh that's better than I
24:13 thought it would
24:14 [Laughter]
24:16 be hey Kyle you know um I know you've
24:19 been talking about this AI stuff for for
24:21 a little bit lately I are you on Tik Tok
24:25 anyway so listen I was on have you used
24:29 the chat GPT yeah yeah so I went on the
24:33 chat GPT and I thought it was just like
24:34 Google right and then I um I did like a
24:38 search and it like it said I don't have
24:40 that latest information it was like old
24:43 and I was like well that's kind of
24:44 useless and and then I asked it to make
24:47 me a business plan and so anyway so I
24:49 just launched a new business I didn't
24:51 know I could do
24:55 that every [ __ ] one
25:01 why do they always talk like that
25:03 too I'm a skeptic and so this is how I
25:06 talk this is just how I talk I don't
25:09 know what's going on with my mouth this
25:11 is just how God gave it to me you know
25:13 bless
25:19 him what is with those
25:24 people figure I'd give chat jbt a try
25:27 and uh hey did you know that it makes
25:30 pictures no but you know what's funny is
25:33 I I put in uh I put in a it's a the John
25:37 Deere lawn mower the push mower the the
25:39 green one and I I said wouldn't it be
25:42 hilarious if it if it had a picture of a
25:45 a a John Deere uh push mower on a on a
25:48 lawn and sure enough it had a picture
25:51 right in there like I don't how many
25:52 pictures are in there it's a lot right
25:55 that was that was that was really
25:57 something
25:59 what was the AI that came out today that
26:01 said to that you upload a photo to turn
26:04 it into AI it's on your list Deco here
26:06 so let's go play with
26:08 that we'll get we'll get back to the
26:10 Skeptics because
26:13 they by the way if you're in here right
26:15 now right now I I see you if you're in
26:20 here and you're like if you're one of
26:22 these sitting on the sidelines like well
26:24 I don't know if you know this but it
26:27 hallucinates you know it's the world's
26:29 greatest plagiarism engine if you're if
26:32 you're sitting on the sidelines and you
26:33 haven't [ __ ] tried
26:35 it go right
26:38 now is he always this aggressive he
26:40 seems Mighty
26:44 aggressive go right now to chat.
26:47 open.com just do it oh yeah I'll get
26:50 through it tomorrow no now now
26:53 now now no do it
26:56 now chat
27:02 open.com and just start playing with it
27:04 while you're listening to me ramble and
27:06 show off this Deco here [ __ ] go ask it
27:08 to write you a poem ask it to write you
27:10 a song ask it to write you a business
27:12 plan ask it to come up with a new pizza
27:14 recipe ask it to write some code for you
27:18 just go watch what it
27:19 does and then and then when you're like
27:22 oh that's cute push past the cute and
27:26 take something you know like I don't
27:27 know maybe you're really into I don't
27:30 know uh uh 15th century
27:33 literature ask it about some obscure
27:36 [ __ ] from your literature degree and
27:38 watch what it does push it push it push
27:42 it go now go all
27:47 right all
27:51 right I'm not putting up with
27:54 it I'm not putting up with You Nero well
27:58 SK ICS who are sitting on the
28:00 sideline if you want to be against AI
28:03 after having tried it I'm happy to have
28:05 that
28:06 conversation but if you're just over
28:08 there listening to what you heard on the
28:09 New York [ __ ]
28:12 times all
28:15 right sorry you asked Jim right all
28:19 right that had nothing to do with Jim by
28:22 the way all right where are we going
28:23 we're going to go to app. deoh
28:27 here.com uh uh uh uh
28:31 r. dcoh
28:37 here.com and we're going to sign in with
28:39 the Google single sign on Google made it
28:43 so easy for us to all just sign on
28:45 they're definitely not keeping our data
28:48 of all the apps we sign on with
28:50 definitely not they take privacy really
28:56 seriously okay so how we going to do
29:02 this
29:05 uh okay so this is it so if you haven't
29:08 seen it before you can just go in here
29:10 and you can type in a
29:14 large come
29:18 on
29:22 semi
29:25 truck in
29:27 pink polka
29:32 dots
29:36 zooming down
29:39 the
29:42 highway
29:43 in New
29:46 Jersey
29:48 at
29:51 dusk all right so if you haven't seen
29:55 anything built on top of sdxl Turbo
29:58 that's what this is sdxl Turbo just like
30:02 makes [ __ ] in real time as you're typing
30:04 it it's Bonkers it's crazy and then
30:06 there's like a whole bunch of pictures
30:07 here and then you can take any one of
30:10 those pictures I don't know why I'm
30:12 talking like this all of a sudden GE
30:14 williker
30:17 Mr G Willer Mr McGill
30:22 Cy I don't think I've had enough water
30:24 today the Ye Old vocal cords are season
30:28 up on
30:28 [Laughter]
30:33 me all right I can take this thing and I
30:36 can click this little button right here
30:37 and just say movie and it's wait oh yeah
30:42 it's queued up okay so it's going to
30:44 make a movie out of that
30:47 image so that's the lower thing down
30:49 here okay so that's all cool right but
30:53 they've added this little button right
30:56 here called use reference
31:00 person and so oh I don't know let's take
31:04 someone like o Taylor
31:08 Swift and so it's now got the reference
31:12 in there and so I can
31:16 say
31:18 um woman sitting outside uh oops woman
31:25 sitting
31:27 outside
31:31 a chrome
31:34 Diner in the
31:42 1950s I mean holy
31:45 [ __ ] I
31:47 mean holy
31:53 [ __ ] now okay couple of things that I've
31:56 discovered that are weird about it
31:59 I can't
32:00 say I can't say like closeup or like far
32:05 away shot like we can't see like her
32:08 standing in front of the diner we can
32:10 only see from like the shoulders up cuz
32:12 that's what the reference photo
32:14 is
32:16 but holy good Lord right and now let me
32:20 change out that reference photo
32:23 for who's that Brie
32:26 Larson yeah so now it's Brie Larsson in
32:30 front of Jan 50s
32:36 diner that actually kind of looks like
32:42 her and then you're like oh well they
32:44 got all those pictures in there already
32:47 that's how it can do that Kyle yeah I
32:49 don't know if you know this but they
32:50 they put they put like you know they
32:53 create a whole database of pictures and
32:55 then they're just pulling it up there
32:56 and they're copy and pasting her face in
32:58 front front of it no you Dum dumb it's
32:59 not like that so we'll go get I I
33:03 actually paid for a
33:06 $9
33:10 um dreams I paid for a $9 a month
33:14 subscription for that stupid thing so I
33:15 could show y all
33:21 this I invest in you people I do it I do
33:25 it I do it all right that's not what I
33:29 was looking for dang
33:31 it
33:34 um contains dreams I should have Kyle
33:37 Shannon dreams in here somewhere
33:41 [Music]
33:45 no Kyle Shannon
33:54 [Music]
33:57 dreams
34:05 oh there
34:07 look we'll do serious me all right
34:10 there's serious me all
34:15 right let's say um let's say
34:26 um okay that's just rude rude that is
34:29 just rude
34:35 that okay we'll do this oh I got it I
34:38 got it I got it okay um
34:41 Sears carpet
34:45 salesmen
34:47 from the
34:57 1970s
35:03 [Laughter]
35:08 oh hey Tom lungren here can I interest
35:11 you some in some in some high
35:16 pile listen I'll tell you what the new
35:18 thing's going to be the new thing's
35:21 going to be tiled carpets yeah you don't
35:23 buy it in rolls anymore you buy it in
35:25 squares and I'll tell you what the new
35:28 stuff it's got adhesive on the back
35:31 anyway here's my card Tom lungren give
35:33 me a
35:38 ring uh let's see um so let's see um JC
35:46 Penny
35:49 um JC Penney yeah salesman yeah
35:52 salesman's good uh let's see the
35:54 salesman we'll say uh vice president
36:00 from the
36:01 1970s what what was why was there a gun
36:04 there
36:06 in the
36:09 corporate
36:23 headquarters all right let's see
36:27 um
36:28 tattoo
36:33 artist
36:36 in grenwich Village New York City in the
36:57 80s
37:11 listen I'd like something Little Lacy La
37:14 see across the chest so I just shaved I
37:16 you know I shaved down to my nipples I
37:17 don't you know I don't like to use that
37:19 term but I don't want to be forward
37:21 anyway uh so I'm clean you know just
37:24 just down to sort of you know water
37:26 level let's call it water level okay so
37:28 anyway so what I'm thinking is I would
37:30 love just a little lace if I could have
37:32 like a not so much a necklace just kind
37:35 of like the like a like a you know
37:37 whatever you would call that right up
37:39 through
37:41 [Laughter]
37:44 here let's see from the 80s uh with a
37:51 tribal
37:53 tattoo on his
37:56 face
38:00 come on let's get some face tattoos
38:04 going listen I was
38:13 want look at the
38:18 hair you got to show your wife I think
38:22 my wife has seen enough she's seen me
38:25 with this
38:26 hair
38:32 oh good
38:37 lord all right let's see
38:41 um let's say
38:44 uh
38:46 motor
38:49 cycle gang member
38:53 in southern
38:56 la
38:58 in
39:02 1999 on a
39:06 big
39:13 Harley yeah yeah we call him baby face
39:16 you don't [ __ ] with baby face yeah he
39:18 looks sweet
39:20 enough he all right he's all right I
39:23 just wouldn't I wouldn't [ __ ] with him I
39:25 mean look at his hair
39:28 anyone that goes around with that you
39:30 know they're willing to take you on and
39:33 they're going to [ __ ] take you down
39:35 he's not going to risk messing that hair
39:37 up you know what I'm saying you don't
39:39 [ __ ] with baby
39:41 face should have invested in Invidia
39:44 honest to God I like I was doing this I
39:47 started the AI Salon last [ __ ]
39:49 January Nvidia was like what was it 100
39:53 bucks or 70 bucks now it's
39:56 760
39:59 $760 it's like I think I think nvidia's
40:03 market cap just exceeded Canada's GDP or
40:06 some [ __ ] like that it's like good
40:13 Lord
40:15 oh man man man man man okay so
40:19 anyway that's Deco here's oh let's let's
40:22 look at the the video that it made for
40:24 us so here's the pink
40:26 truck
40:28 yeah see yeah the vi the these videos
40:31 like the the uh the stability AI videos
40:35 right now are kind of
40:36 shitty they're not great but I did a
40:39 cool thing I took this picture this was
40:41 one I did
40:43 earlier and I threw I took the picture
40:46 so this was the picture that Deco here
40:49 made out of my photo
40:52 earlier and then I took it into chat
40:56 Je
41:07 I took it in the chat
41:12 [ __ ] and it made that
41:17 image oh
41:20 yeah
41:21 [Applause]
41:22 yeah it's
41:25 electric it's middleaged B how you
41:32 doing I told it to give me salt and
41:34 pepper hair and it made it like some rad
41:36 Punk
41:38 thing Kyle you never told chat what the
41:42 app was yeah I did Deco yeah app. deoh
41:48 here. this this other app that we were
41:56 doing this one this is Deco
42:01 here app a. deoh here d e c o h e r e.
42:09 a Mr K what's happening look Mr K you
42:14 too can be a
42:23 biker let's see pop art post
42:31 with
42:33 blonde
42:35 spiky
42:36 hair no that
42:40 [Laughter]
42:52 what
42:54 yeah that's good that's I think that's
42:57 appropriate
42:57 [Laughter]
43:10 oh so this is one of the things on the
43:13 list let me go back to the
43:17 list it's crazy
43:20 right and like like this is like what
43:23 what's what's freaking me out cuz one of
43:25 the things when I first started doing
43:28 Kyle Shannon dreams so like fall of
43:34 22 fall of
43:36 22 I started making these images and
43:38 people were like you know how I just
43:40 want to upload a picture of myself and
43:41 like make a thing like well it's not
43:45 that easy and you got to make a model
43:46 and you got to do this and you got to go
43:48 to dream booth and do that and and then
43:51 there were these apps that came out and
43:52 it's like well you got to pay the $9.95
43:55 a month or the1 1995 a month but then
43:57 make sure you cancel cuz
44:00 just and so now it's just like yeah you
44:03 just put your picture there and you just
44:04 make yourself into some androgynous 80s
44:07 pop
44:10 star boom
44:13 done now does it look exactly like me no
44:16 but it's enough to make me laugh like
44:18 it's it's close
44:19 enough it's in the
44:23 hood crazy
44:26 right
44:29 let's turn Joe Rogan into an 80s pop
44:31 star shall we Hey Joe
44:35 Rogan look you think you're so Macho
44:38 look at you now with your pop
44:41 background come
44:42 on come
44:51 on he looks like dairo and taxi
44:56 driver
45:00 I'm a
45:01 sunflower hey it's Joe Rogan sunflower
45:05 welcome to 80s
45:08 pop oh yeah you go Joe
45:13 Rogan you're so Butch love you Deco here
45:17 has a few guard rails compared to the
45:19 other tools which is ref oh has few
45:20 guard rails yeah and in fact you can
45:24 turn them off the safety filter you can
45:27 turn off
45:28 if you if you subscribe to it which is
45:30 pretty smart this is actually pretty
45:31 smart so so this tool by the way Deco
45:34 here is just an application it's like a
45:36 website someone built on top of stable
45:38 diffusion tools okay so just like Jasper
45:42 was sitting on top of chat GPT and I'm
45:44 like it's not really worth playing with
45:46 some of these tools built on top of
45:47 stable diffusion are pretty slick
45:49 because they take all the complexity
45:50 away right so it might be worth paying a
45:53 month or two on this just to play with
45:54 it but you could go do all this [ __ ] in
45:57 stable diffusion if you want to geek out
45:58 you can go do all this stuff for
46:02 free
46:04 um but anyway they charge you you can
46:06 see that you can turn water marks off
46:08 and safety filters off only if you
46:10 subscribe which I thought was pretty
46:12 slick Hey Joe Rogan you're looking fresh
46:17 hey
46:22 thanks that's
46:26 awesome
46:28 okay I have an idea let's go find a
46:30 funny one now that's too real well
46:33 that's not bad all right that's good so
46:37 I'm going to turn motion up to
46:39 10 and we're going to turn this into a
46:43 video let's let's see how wacky it makes
46:46 Joe
46:46 Rogan what did the New York Times say I
46:49 don't know about
46:52 what what's the name of the AI please
46:55 tell us what the name of the AI is Kyle
46:57 I want to know what the name of the AI
46:59 is is this the one is this the one is
47:01 this it how do you get there it's pinned
47:05 it's pinned it's pinned it's not pinned
47:08 now deoh here.
47:11 a the hair is over
47:14 9,000 uh looks like Billy
47:18 Idol Deco here my camping neighbors must
47:21 think I'm mental all they can hear is me
47:26 laughing
47:28 oh look we got a we got a Joe Rogan
47:30 movie come on please not be boring
47:32 that's pretty
47:34 [Laughter]
47:35 cool oh you know what I'm gonna do let's
47:38 download
47:40 this uh download did I download
47:49 it yeah it did
47:52 okay we're going to go to
47:55 x and we're g to I love doing stupid
47:58 [ __ ] like tagging Sam Alman which by the
48:00 way he never got back to me and he never
48:02 rendered my movie Butthead so I'm GNA go
48:06 at Joe
48:12 Rogan at Joe
48:14 Rogan
48:16 um hey
48:21 man let's
48:26 see
48:27 you were so cool in the
48:32 80s
48:34 [Laughter]
48:39 man and then let's upload our little
48:52 movie all right so do me a favor go to
48:55 my Twitter go to my ex
48:58 which is it's at Kyle
49:00 Shannon and all right I just tagged Joe
49:02 Rogan so go boost the [ __ ] out of that
49:05 thing let's see if we can get Joe Rogan
49:07 to to
49:09 [Laughter]
49:17 respond oh Lordy all right that's some
49:20 good fun right there I'll tell you what
49:23 mm that's some good times you go out
49:25 there on the internet and you play play
49:27 with some of them their AI tools and you
49:29 make some funny picture like Joe Rogan
49:31 with a tie-dye background you put it out
49:33 there on the X they call it X now they
49:35 used to call it Twitter anyway you you
49:37 put it on x.com and then you just say to
49:40 Joe Rogan Hey Joe Rogan look at you in
49:42 the 1980s it's so funny
49:46 that's you got any more
49:56 Twinkies did he take his medication
50:00 [Music]
50:03 today all right what do you want to do
50:06 what do you want to
50:07 do Chrome gets built-in AI rating
50:11 writing tool powered by
50:13 Gemini
50:16 really oh I think I saw
50:20 that I was in
50:23 something yeah it was like three little
50:26 options came up
50:28 I think I saw that that little writing
50:30 tool inside Chrome yeah this stuff's
50:32 going to be everywhere Adobe brings oh
50:33 conversational AI to trillions of PDFs
50:36 with a new AI assistant and reader and
50:39 acrobat I haven't played with that yet
50:42 I'm I really want to play with the
50:45 new um Gemini
50:48 1.5 but I assume that they're not
50:50 releasing it because the million token
50:52 context window is probably expensive is
50:54 is my
50:56 guess
51:00 Kyle didn't take his
51:04 medications yeah but look I did do my
51:12 hair put my head in a blender did my
51:18 hair can we write some classical piano
51:21 music we could do
51:26 that you could do it at
51:29 refusion but I think refusion now I
51:31 don't know if refusion I don't you know
51:34 what I'm not good enough at the music
51:35 tools to do it in a way can you explain
51:38 stable diffusion and how it works I can
51:41 I can explain how diffusion models work
51:44 Yes stable diffusion is just one of many
51:48 tools that are using the diffusion
51:50 method or model or whatever the [ __ ]
51:54 they call
51:55 it um
51:57 I have a slide that explains how that
51:59 works if you want to
52:06 know I think I'm GNA have to pull
52:08 something down from
52:10 yi yi
52:14 iCloud but let me do that and I'll yeah
52:17 I'll I'll edim M you on how how these
52:20 image generation tools work when
52:23 you okay so one of the things that is
52:28 um is a misconception of AI right now
52:33 is that it copy and pastes from The
52:36 Source material it does not the source
52:39 material ceases to
52:41 exist once it's trained and
52:44 embedded um in the case
52:47 of text
52:50 documents they all turn into these
52:52 things called vectors which is just a
52:55 long string of numbers numbers that
52:58 just live in Thousand dimensional
53:01 mathematical space I'm not that sounds
53:04 like I'm just making [ __ ]
53:06 up that's what it is you have a point in
53:10 Thousand dimensional mathematical space
53:13 and when you type a
53:15 prompt the mathematics of the large
53:18 language model says based on that prompt
53:21 there's a probability somewhere in this
53:23 massive Vector space they call it this
53:26 thous dimensional space where one of
53:29 those points is has the highest
53:31 probability of being the right thing and
53:35 then that point in space is associated
53:37 with a word and then it puts that word
53:41 in and then it does it again and then it
53:44 does it again and again and again so
53:46 when it spits out all that
53:48 text it's generating it token what what
53:51 they call token by token but effectively
53:53 word by word it's writing original
53:56 content
53:57 content the image tools are similarly
54:02 funky and
54:04 weird and they're not copying and
54:06 pasting so all right this is how
54:09 diffusion images are
54:11 made so see that pretty picture let's
54:14 say that pretty picture is in the
54:15 training set so so when you use an image
54:20 generation
54:22 tool you've got two models that are
54:24 interacting you've got the large
54:26 language model which understands the
54:28 words and then you've got the diffusion
54:30 model which which has got some data
54:33 about images but what happens is you
54:35 would take this image for example and
54:37 you would tag it right so human beings
54:40 or maybe even computer systems using AI
54:43 would write a bunch of tags for this so
54:45 they would say I don't know
54:48 anime you know cartoon
54:51 illustration whatever those trees are
54:53 cherry blossoms or whatever the heck
54:55 they are um Cliff Cottage House
55:01 Rocks garden lawn you know bushes
55:05 whatever uh Sky clouds Mist Haze right
55:09 so there'd be a whole bunch of tags and
55:13 then this thing gets
55:14 embedded or this thing gets noised so
55:18 they they noise it so okay so so that's
55:21 so you're uploading a whole bunch of
55:24 images into the data set that have been
55:27 tagged and then the way these things
55:30 work is
55:33 this you you have the large language
55:36 model that understands the words and
55:39 then you have the image model that
55:40 understands
55:41 images the so see how that thing just
55:44 went noisy where it was an image before
55:46 and now it's been noised into just sort
55:48 of this random noisy image so the
55:51 labeled images have noise added and are
55:54 vectorized so that they're basically
55:56 turned into number into
55:59 numbers and
56:01 then sorry
56:04 no um and then based on your prompt the
56:08 large language model interprets and
56:10 filters the relevant image vectors now
56:13 that doesn't mean [ __ ] but here's what
56:14 happens so this is a very simplified
56:18 Vector space this is a two-dimensional
56:20 Vector Space X and Y right and then
56:23 these little clusters you have you know
56:26 related little clusters of points in
56:29 Vector space that are tied to the tags
56:32 of those images
56:33 right and so when you type in your
56:36 prompt which in this case is Cat on a
56:40 desk the large language model sort of
56:42 goes through its Vector space and it
56:44 says there's a bunch of these things
56:45 that are not relevant but these pink
56:49 ones you know are are all sort of cats
56:52 and these uh blue ones are all sort of
56:55 desks right they've been tagged and then
56:58 associated with all of those points in
57:00 Vector space are a bunch of these noised
57:03 images that you can't see anything on
57:05 and then you know somehow mathematically
57:08 they take all the [ __ ] that is high
57:10 probability of these noised images that
57:13 it it knows nothing about them other
57:15 than there was some association with
57:17 some word and there's some probability
57:20 and it turns it
57:22 into a like a a a like an average of all
57:26 of these IM of all of these noisy images
57:29 here's this noisy image and then what
57:32 the system does is it takes the same
57:34 algorithm that turned them into noise in
57:36 the first place and it denoises this
57:39 averaged
57:41 image oops what did I do
57:46 oh so it denises it and it gets a little
57:49 less noisy and a little less noisy and
57:52 then in the end you have a cat on a
57:55 desk so it has no idea what's in the
58:00 image so this is why early on because it
58:03 wasn't trained right you get hands that
58:05 were like this cuz when it averaged the
58:07 images together maybe it brought hands
58:09 together in a weird way I don't know how
58:11 they fixed that
58:13 [ __ ] but
58:14 like it's all just math it's all points
58:17 in Vector space and it's these these
58:19 sort
58:20 of you know messes of images that get
58:24 averaged into this one image and that's
58:26 why you sometimes get things that look
58:29 normal but not quite like if you ask for
58:31 a 70s muscle car it'll look sort of like
58:33 a Mopar car and then it's kind of got a
58:36 Chevy chevel kind of feel to it or or an
58:39 AMC Javelin kind of feel to
58:42 it so that's how they work isn't that
58:45 crazy and what's even crazier when
58:47 you're doing something like Deco
58:51 here where's Deco
58:55 here there Joe Joe's so
58:59 cute let's get rid of
59:02 Joe and we'll start over
59:06 here and we'll go um
59:10 ancient
59:12 teapot
59:14 on a
59:17 museum
59:19 pedestal all that [ __ ] I was just
59:21 talking about is happening in real time
59:25 with this model
59:26 [Laughter]
59:29 so so like there's just like [ __ ]
59:32 millions of images that have been noised
59:35 and in real time it's averaging them and
59:36 denoising them in like half a second or
59:39 I what however long it takes not
59:48 long so like these are not pictures that
59:50 exist but these are amalgamations of
59:53 some elements of pictures that exist
59:57 but like the the the actual images cease
59:59 to exist in the in the space in the
1:00:03 latent space or in the diffusion space I
1:00:06 don't know what they call it with
1:00:08 images isn't that
1:00:11 crazy it's just nutty it just nutty it
1:00:15 just nutty now I don't know you know
1:00:17 that the AI they put them in the they
1:00:20 put it in there they got the they used
1:00:22 the
1:00:23 GPU so so they used the GPU the GPU it
1:00:27 stands for graphics PR setting unit in
1:00:30 the in the in the gamers The Gamers The
1:00:32 Gamers used to buy the GPU to make the
1:00:34 game make the game go faster they they
1:00:37 wanted to 60 Herz and then they wanted
1:00:39 to to 90 Herz and then they wanted 120 H
1:00:41 Hertz so they can put hurt on the enemy
1:00:44 right and that's so they did GPU and
1:00:47 then all the other chip manufacturers
1:00:49 are like yeah you can go Nvidia you just
1:00:51 go make them gpus for all them snotting
1:00:54 those little teenagers that's good
1:00:55 that's fine everything's fine you can go
1:00:57 do that we're not going to make them why
1:00:59 don't you go make the gpus we'll make
1:01:01 the CPUs in the RAM and then you make
1:01:03 the gpus and then everybody good right
1:01:05 you good we're good and and then and
1:01:07 then the crypto came crypto came and
1:01:10 they were they were making blockchain
1:01:11 they were they were doing the the the
1:01:13 Bitcoin and the
1:01:15 nfts and they and they use the gpus and
1:01:17 all the all the chip manufacturers like
1:01:19 yeah you just Nvidia go go make them
1:01:22 them them chips for the blockchain that
1:01:25 you that's good the gamer don't want it
1:01:27 anymore but the blockchain people do
1:01:28 yeah just go make that now and then and
1:01:31 then all of a sudden the AI came out and
1:01:33 then never was like yeah Nvidia just
1:01:36 just go make them AI
1:01:37 chips and and we're good we're not going
1:01:39 to make them and then and then chat gbt
1:01:43 came out and 100 million people used it
1:01:46 and all the other people were like holy
1:01:48 [ __ ] we got to make
1:01:49 gpus what do we
1:01:53 do it was too
1:01:55 late Nvidia has the market cornered cu
1:01:59 no one gave a [ __ ] about it
1:02:01 apparently [ __ ]
1:02:09 crazy I don't know why my Billy Bob
1:02:11 character there knew so much about chip
1:02:14 manufacturing but you know stranger
1:02:17 things have happened the world's going
1:02:18 to get weird people are going to have
1:02:19 all sorts of knowledge you don't think
1:02:21 they should have access to and seen and
1:02:25 seen
1:02:27 thank you Lord digital Gods was that was
1:02:30 that payment for my
1:02:31 [Laughter]
1:02:37 performance all right let's go do some
1:02:39 more show and tell how do you get over
1:02:41 the criticism that everyone hates you
1:02:44 well that's not what she said how do you
1:02:46 get over the criticism that training
1:02:48 sets are mass theft I can never
1:02:51 reconcile
1:02:54 that um
1:03:02 it's not an easy
1:03:09 answer if it were if it were actually
1:03:13 sampling I I would
1:03:16 say um you know if if it were actually
1:03:19 like like sampling in the 80s and 90s
1:03:22 with hip-hop they were actually chopping
1:03:25 chunks out of song and putting them in
1:03:27 other songs right so they're they're
1:03:28 actually doing
1:03:30 that that's not first of all that's not
1:03:32 how these tools work so part of it is
1:03:36 just there's an assumption on the part
1:03:39 of the artist whose Works have
1:03:42 been used for
1:03:45 training that these tools work a way
1:03:47 they don't
1:03:48 now why it's gray and weird is you can
1:03:52 prompt large language models and you can
1:03:55 prompt image generation tools to get
1:03:57 really [ __ ] close to someone's work
1:03:59 and you know turns out that Ye Old mid
1:04:03 Journey they don't have any [ __ ]
1:04:05 Scruples at all and you know there was a
1:04:08 document that was leaked from mid
1:04:09 Journey that said here's all the artists
1:04:11 that we explicitly trained trained on
1:04:13 their data and they put those names in
1:04:15 the model so that you say when you say
1:04:17 in the style of some artist it looks
1:04:19 just like that artist right because mid
1:04:22 journey is just basically run by a
1:04:24 cowboy that's like [ __ ] it let's go woo
1:04:27 W Adobe Firefly on the other hand was
1:04:32 not trained on copywritten images so one
1:04:36 of your counters can be if you're really
1:04:38 concerned about that go use Adobe
1:04:41 Firefly because it was only trained on
1:04:43 open source or public domain images and
1:04:46 images that adobe already bought the
1:04:48 rights to or owned the rights to and
1:04:51 then Adobe is also allowing artists to
1:04:53 opt out of being trained opt into being
1:04:55 trained trained and they're developing
1:04:57 um attribution and compensation models I
1:05:00 don't know where they are on that but
1:05:02 Adobe is at least being conscious of
1:05:05 this mid journey is like [ __ ] it which
1:05:08 is why mid journey is better than
1:05:09 everything
1:05:10 else because they just said screw it now
1:05:14 here's here's how I think about
1:05:19 it from a conceptual standpoint the way
1:05:23 these models work it's actually a lot
1:05:26 closer
1:05:28 to if you go to college and all you do
1:05:31 is study pop art you love pop art you
1:05:33 love Andy Warhol you love lickstein you
1:05:37 love you you like all Keith Herring you
1:05:39 like all the pop art you know artists of
1:05:41 the 80s and you study everything about
1:05:44 them and just in your head are going
1:05:46 images and and words and how they think
1:05:49 about things and their poetry and who
1:05:51 were the Poets of the time and they were
1:05:53 listening to Velvet Underground and you
1:05:55 got Lou re and Lori Anderson and all
1:05:57 that shit's going in your
1:05:59 head and then you you leave
1:06:02 college and you're like I'm going to
1:06:03 make a masterpiece and the first piece
1:06:05 of art you do looks just like Keith
1:06:08 Herring or it looks just like Andy
1:06:11 Warhol
1:06:14 well is that their work are you going to
1:06:16 get sued for that well the Keith Herring
1:06:19 estate sues the [ __ ] out of people that
1:06:23 make fat marker
1:06:26 you know people that look like Keith
1:06:29 Herring drew them so the Keith Herring
1:06:32 estate you know they've put a flag in
1:06:35 their ground and like we'll sue you if
1:06:36 it looks anything like
1:06:38 it but you're you've been inspired by
1:06:42 those guys and you're doing a kind of
1:06:44 work that that
1:06:47 is you're the one coming up with the
1:06:51 idea you're the one executing it and you
1:06:54 might have been inspired by them so it's
1:06:55 more like like that and most of the
1:06:58 times what's happening when I see people
1:07:00 doing things in AI they're not going I
1:07:03 want something that looks like Andy
1:07:05 Warhol and I want it like that now you
1:07:06 can do that it's really [ __ ] lazy the
1:07:09 artists that I know that are doing this
1:07:11 are like okay I want something that kind
1:07:12 of has this vibe to it no I want it more
1:07:14 like this I want it more like that if
1:07:16 you watch Corey Sandler her art she'll
1:07:18 go through 30 40 70 100 different
1:07:24 variations of an image before she gets
1:07:26 to one where she's like o that's what
1:07:28 that's the feeling I'm going for and
1:07:31 then she pulls that one
1:07:33 out and then the other place I go is so
1:07:36 so that's kind of it conceptually and
1:07:38 then the other place I go is
1:07:44 this machine learn the the p in GPT
1:07:47 stands for
1:07:50 pre-trained and what open AI did for
1:07:53 chat gbt is they said let me go get as
1:07:55 much of the public internet as I
1:07:57 can and let me train this thing on all
1:08:00 that data out there so they're they're
1:08:03 they're
1:08:05 effectively performing the function of a
1:08:08 library or a
1:08:11 museum they're collecting the output of
1:08:14 humanity that was put in this particular
1:08:16 form called the internet and they're
1:08:19 putting it
1:08:20 in the digital
1:08:24 equivalent
1:08:27 of a
1:08:28 museum right of a library so they said
1:08:31 okay as much of the internet as we could
1:08:33 get we're going to get it we're going to
1:08:34 jam it in
1:08:36 here if you carry that to its logical
1:08:43 extension you take all of the output of
1:08:47 humanity and you make it accessible to
1:08:51 the rest of humanity there is a much
1:08:54 bigger thing at play here than
1:08:55 individual contributors going you owe me
1:08:58 a millionth of a penny every time an
1:09:01 image might slightly look like my myew
1:09:04 glasses from my famous photo
1:09:09 right I love your lives you make AI fun
1:09:12 thank you Suzy AI ninja
1:09:15 Mastermind all it's Susie AI ninja
1:09:18 Mastermind how can I help you welcome to
1:09:20 the party
1:09:23 kids so for me
1:09:31 the you should never train on someone
1:09:34 else's
1:09:35 art is kind of like saying you should
1:09:38 never put [ __ ] in the library and you
1:09:40 should never put [ __ ] in
1:09:42 museums because only the people that
1:09:45 created them or have the rights to them
1:09:47 should ever be able to do anything with
1:09:49 them there's also there's a really
1:09:51 interesting response when when Sam Alman
1:09:54 was asked about the New York times
1:09:56 lawsuit in
1:10:00 Davos they're suing
1:10:03 him because what they're claiming is the
1:10:06 the the data set was trained on New York
1:10:10 Times data because they can do a prompt
1:10:13 in a certain way that will pull up a
1:10:15 direct quote from a New York Times
1:10:18 piece so Sam Alman had a couple couple
1:10:22 of interesting
1:10:23 points the first one was
1:10:26 we don't need the New York Times for
1:10:28 training data like we've we've got
1:10:30 enough and the analogy he used
1:10:33 was if you want to get like a high
1:10:35 school biology
1:10:39 education and you've read 2,000 books on
1:10:42 biology does the 2001st book really make
1:10:45 a difference no it
1:10:48 doesn't what he said was the reason they
1:10:51 they were in negotiations with the New
1:10:54 York Times to cut a publishing deal
1:10:56 was on the output side so that when
1:10:58 somebody did a search on chat
1:11:01 GPT and it came up with an answer like
1:11:04 they could actually reference New York
1:11:06 Times data and say hey here's an actual
1:11:09 reference for the New York Times click
1:11:10 here to go there they were trying to cut
1:11:13 an attribution deal not on the training
1:11:15 data but on the output
1:11:19 side the other thing he said
1:11:22 is the New York Times publishes digital
1:11:27 data and so when someone's got a
1:11:30 subscription to the New York Times and
1:11:32 there's a really good article there
1:11:34 sometimes they'll copy and paste that
1:11:36 and stick it on LinkedIn or stick it in
1:11:38 a PDF or stick parts of it in their in
1:11:41 their
1:11:43 article so while open AI could easily
1:11:47 say don't train on New York
1:11:50 times.com it's effectively impossible
1:11:53 for them to say find every in of every
1:11:56 mention of the New York Times at every
1:11:59 location on the internet for all of
1:12:02 time could they do it well they
1:12:06 could but you know they go goes back to
1:12:09 the library thing should no New York
1:12:12 Times ever be in a library should you
1:12:14 never be able to go back and look at
1:12:16 those why do libraries exist for the
1:12:19 good of
1:12:21 humanity so if you play out the open AI
1:12:24 data training set to logical extension
1:12:27 you put all of the data from all of
1:12:29 humanity in
1:12:32 It Mary Mary thank you Kyle this is so
1:12:35 common sense to me as
1:12:36 well now
1:12:40 listen I don't think it's Crystal
1:12:43 Clear but I think it's
1:12:51 impossible it's an impossible solution
1:12:54 right here here's here's the solution as
1:12:57 I see
1:12:59 it right now everyone's chasing the
1:13:03 outputs everyone's saying well if I put
1:13:05 out a song and you copi my song I'm
1:13:07 going to sue you because that's
1:13:09 copyright we're moving into a world of
1:13:14 infinite
1:13:18 infinite content
1:13:22 generation so how many millions of fans
1:13:24 does TW Taylor Swift have 100 million a
1:13:27 billion I let's say let's say a
1:13:30 billion so how many millions of fans
1:13:32 does TW Taylor Swift have 100 million a
1:13:35 billion I let's say let's say a
1:13:39 billion that talk about my dog named
1:13:43 Sam and my sister named
1:13:46 Lily
1:13:47 and the Little Creek that we go swimming
1:13:51 in and a song will start
1:13:54 playing that is Taylor Swift singing
1:13:57 that
1:13:58 song and at some point I promise
1:14:04 you one of those billions of people that
1:14:07 writes one of those trillions of Taylor
1:14:09 Swift songs that song is going to become
1:14:11 a
1:14:13 hit and then what happens does Taylor
1:14:17 Swift sue her fans or partner with
1:14:21 them
1:14:24 right so I think it's
1:14:28 ridiculous even even in a three-year
1:14:31 window to go after the outputs because
1:14:34 the outputs are gone like copyright as a
1:14:38 mechanism was based on the fact that it
1:14:40 used to be hard to replicate
1:14:43 content and then the internet came and
1:14:46 digital music came and they sort of
1:14:48 finagled some [ __ ]
1:14:51 Byzantine laws about sample sizes and if
1:14:54 you use less than 11 seconds and
1:14:57 it's
1:14:59 common like all that [ __ ] to work
1:15:02 around the fact that like it's just
1:15:04 infinitely easy to copy and distribute
1:15:07 Digital Data well now we're moving to
1:15:09 the next evolution of this which is it's
1:15:13 infinitely easy to generate original
1:15:16 content that sounds like whatever the
1:15:18 [ __ ] we want it to sound like
1:15:21 so here's where I think the solution
1:15:23 goes and YouTube has already teased this
1:15:27 so YouTube about a month ago put out a
1:15:30 thing I forget what they called it it
1:15:31 was some sort of audio
1:15:33 thing where if you want original music
1:15:36 behind your YouTube
1:15:39 short they cut a deal with Charlie poth
1:15:42 and
1:15:43 Tay and you can go I have a video about
1:15:47 me ranting about copyright issues could
1:15:50 you please have Charlie poth sing my
1:15:52 background
1:15:54 music and it makes a song written in the
1:15:58 style of Charlie poth with the voice of
1:16:01 Charlie poth about your little topic and
1:16:03 they let you put that song behind your
1:16:06 your YouTube video and play it for
1:16:09 everyone to
1:16:11 hear they're paying him on the
1:16:15 generation Tay and Charlie poth have an
1:16:18 original model that is licensed to
1:16:21 YouTube on the
1:16:24 generation [ __ ] them let them make all
1:16:27 they want just pay us every time they
1:16:30 push the button every time you go make
1:16:32 me a Charlie poth song send me a nickel
1:16:35 Charlie po song nickel Charlie poo song
1:16:38 nickel and then if that song becomes a
1:16:40 hit [ __ ] it that was cool I got my
1:16:44 nickels Sora red team memb said that it
1:16:47 takes one hour to render one minute and
1:16:50 said 90 hours for a 90-minute movie yeah
1:16:53 that'll get faster Tak one hour to
1:16:56 render I figured that would be the case
1:16:58 because when Sam wman was saying hey you
1:17:01 know send me your ideas I'll render your
1:17:03 videos all of the videos that Sam Alman
1:17:06 was doing were 10 seconds
1:17:08 long so if it takes an hour to do a
1:17:11 uh to do a a 60c clip yeah so so it's 10
1:17:16 minutes for 10 seconds so it's it's a
1:17:18 sec it's it's a minute it's a minute per
1:17:21 second right that's what it sounds like
1:17:24 so that will get faster but that doesn't
1:17:26 surprise me at all like it doesn't
1:17:28 surprise me at all and that's why that's
1:17:29 also why I think we're probably not
1:17:31 going to see Sora for two three months
1:17:33 maybe even
1:17:35 more but they're putting it out there to
1:17:37 let us know what's
1:17:39 coming and I don't think Sora is going
1:17:41 to be the last one from open AI I think
1:17:43 open AI is going to give us four or five
1:17:44 of these things I heard boxes wrestling
1:17:47 and woke
1:17:48 up I have my box
1:17:52 out this is the world's biggest library
1:17:55 and biggest
1:17:57 Museum I didn't go the whole way through
1:18:00 it but I was yelling about
1:18:02 uh about
1:18:05 copyright so anyway that's my two cents
1:18:08 on it that's my that was that was my
1:18:10 $187 on
1:18:15 it and the other thing is this you can
1:18:18 [ __ ] all you want you can whine all you
1:18:20 want you can pick it all you
1:18:23 want the horse gun be out of the
1:18:28 barn they shouldn't train on our data
1:18:32 they already
1:18:33 did they shouldn't have access to all of
1:18:36 our private data well that's Google and
1:18:38 Facebook we gave it to them 20 years ago
1:18:40 so that ship's also already
1:18:44 sailed they should pay me for my blog
1:18:48 post okay how many blog posts do you
1:18:50 think are out
1:18:52 there I if you took your your stupid
1:18:55 little blog post let's say they weren't
1:18:57 stupid let's say they were brilliant and
1:18:58 you did a blog post a day for the last
1:19:00 10
1:19:02 years how many other blog posts are
1:19:05 there out there that if if someone
1:19:08 prompted chat GPT in a way that
1:19:11 somehow said something that sounded sort
1:19:14 of like something you wrote in a blog
1:19:15 post if we could figure that
1:19:20 out what percentage of a
1:19:23 penny would be fair to give give to
1:19:26 [Laughter]
1:19:30 you I made six Sense on my Collective
1:19:34 writings over the past decade I don't
1:19:37 know it's it's
1:19:41 just I have
1:19:44 deep deep empathy for artists and
1:19:47 creators who are feeling like their
1:19:51 reality is being ripped out from under
1:19:54 them
1:19:55 because it
1:19:59 is I don't think it serves them to just
1:20:04 [ __ ] about it and cross their arms and
1:20:06 go [ __ ]
1:20:09 AI because it's not going to make ai go
1:20:13 away and what's going to happen is
1:20:15 you're going to have an artist like
1:20:17 Grimes shortly after Spotify took down
1:20:21 the the weekend and Drake song that was
1:20:25 an AI song that that was a hit on
1:20:28 Spotify and then Drake and the weekend's
1:20:31 Publishers said take that [ __ ] down
1:20:33 right
1:20:35 now the week after that or maybe even
1:20:37 the day after they took that down
1:20:40 Grimes Elon musk's
1:20:42 ex who's a
1:20:45 musician said to her
1:20:48 fans synthesize me all you want deep
1:20:52 fake me all you want if you make any
1:20:54 money give me
1:20:59 half she's going to make a
1:21:06 fortune do you do you know who's going
1:21:08 to make money for the artists that are
1:21:10 trying to sue their
1:21:12 fans the
1:21:14 lawyers the IP attorneys right now are
1:21:17 [ __ ] dancing they're yacht shopping
1:21:20 they've got the big glossy yacht
1:21:22 cataloges they're like dud do I go with
1:21:26 120 M meter
1:21:29 huh okay it's going to take it's going
1:21:32 to take a crew of seven at a
1:21:34 minimum all right so my annual burn is
1:21:37 going to be $20
1:21:41 million or do I go with the
1:21:45 150 30 million bucks a year to run my
1:21:48 yacht I think I'm going to 150 I think
1:21:50 I'm going to 150 oh yeah sorry yes Sue
1:21:53 the [ __ ] out of the fans absolutely yep
1:21:56 no we're right there with you just we
1:21:58 we're writing up the we're drafting the
1:22:01 lawsuit right now hang I just got one
1:22:03 thing yeah let's go with the 150 let's
1:22:05 go with the 150 meter yeah
1:22:08 fantastic that what oh I can only keep
1:22:11 it in certain ports it it it the draft
1:22:14 is too deep that's all you know what it
1:22:18 could we uh could we dredge the uh the
1:22:20 ported ABA we could fantastic all right
1:22:24 let's let's go with the 150 any anyway
1:22:27 yeah we'll be suing those fans just
1:22:29 shortly there
1:22:35 Lars that's who's making money in this
1:22:39 [ __ ] so my my advice my my point of view
1:22:44 is if you're an artist get AI literate
1:22:46 figure this [ __ ] out learn how it
1:22:50 works understand what are the mechanisms
1:22:52 cuz I guarantee you there are going to
1:22:54 be some companies that come out that
1:22:55 make it really easy for artists to make
1:22:58 a model of themselves and monetize
1:23:03 it in fact 11 Labs is one of
1:23:08 those
1:23:21 look
1:23:23 payouts
1:23:26 uh
1:23:27 voices create let's
1:23:30 see
1:23:32 um
1:23:33 11 Labs um forget what they're
1:23:40 called um
1:23:45 voice for
1:23:47 pay the [ __ ] was it
1:23:50 called when you share your voice in the
1:23:52 in the voice Library so on 11 labs
1:23:54 payouts 11 labs when you share your
1:23:56 voice in The Voice Library you can earn
1:23:58 cash rewards every time it's used so in
1:24:01 11 Labs right
1:24:03 now if
1:24:05 you upload a 30 minute or longer sample
1:24:09 of your
1:24:10 voice so this is targeted as at voice
1:24:14 actors you can create a model or
1:24:16 multiple
1:24:18 models let's say you do a funny voice
1:24:20 like this and then you do a fantastic
1:24:23 voice like this
1:24:25 and then you do do normal Kyle voice I
1:24:28 could make those three
1:24:30 models put them on 11
1:24:33 Labs Market them and then every time
1:24:36 someone uses that model I get
1:24:40 paid that's What actors should be doing
1:24:42 that's what artists should be doing and
1:24:43 you know what's going to happen a lot of
1:24:46 artists are going to be sitting like
1:24:47 this going AI is stealing from
1:24:52 us and the other ones are going to go
1:24:56 huh this is scarier than [ __ ] but I'm
1:24:58 going to learn it and I'm just going to
1:25:00 act like it's just another in a series
1:25:02 of Technologies that's going to change
1:25:04 the game and I'm going to be ahead of
1:25:05 the curve and I'm going to go make my
1:25:08 model in 11 labs and then I'll go make
1:25:09 one over on Apple and then I'll go make
1:25:11 one over here and I'll go make one over
1:25:13 there and then meta's going to do one
1:25:14 and and then at some point it's all
1:25:15 going to shake out and there will be two
1:25:18 leaders and I'll be one of the Premier
1:25:21 voices on one of those two sites that
1:25:23 survives
1:25:26 and I'll have a career still and the
1:25:29 people that are like this they're
1:25:30 stealing from me are just going to be
1:25:33 drunk and
1:25:36 angry well I could have done that but
1:25:38 you
1:25:42 know and again I have deep empathy for
1:25:44 them but it's just it's not going away
1:25:47 it doesn't matter what you think about
1:25:49 it it doesn't matter if you think it's
1:25:50 morally not
1:25:53 cool
1:25:55 it's coming
1:26:01 anyway it's a
1:26:05 tsunami you can run around the beach all
1:26:08 you want arguing about whether or not
1:26:10 tsunamis are
1:26:13 good it's still coming
1:26:17 right so buy a [ __ ] surfboard and
1:26:21 start
1:26:23 paddling
1:26:26 man that was a soap box and a half
1:26:28 wasn't it so I don't know if that
1:26:31 answers your question about how you deal
1:26:32 with people that say this [ __ ]
1:26:37 stealing I understand why people think
1:26:40 that I I I don't disagree with some of
1:26:43 their points and this [ __ ] ain't going
1:26:51 away so figure it out get ahead of it
1:26:54 license your [ __ ] before someone steals
1:26:57 your
1:27:02 [ __ ] recording artists earn from the
1:27:04 live performances now less from
1:27:06 recording so I've heard yeah that makes
1:27:08 sense and I listen tonius I think the
1:27:11 value of live performances goes up in a
1:27:13 world of infinite content being able to
1:27:16 see your favorite artist get up
1:27:19 there and kill
1:27:21 it pretty
1:27:23 cool
1:27:26 I forgot that Charlie thing this could
1:27:27 be a cool essay yeah I think this yeah
1:27:30 this would be a good
1:27:32 essay I should I should grab the
1:27:34 transcript of my rant and see if there's
1:27:36 any coherent thought in
1:27:40 there New York Times thinks they're
1:27:42 dealing with old school search engine
1:27:43 they're Clueless yeah
1:27:45 exactly exactly well I think I think all
1:27:47 the people that are making these lawsuit
1:27:49 if you look at the language of the
1:27:52 lawsuits they they're they're basically
1:27:55 based on the 80s and
1:27:57 90s music
1:28:00 sampling technology
1:28:04 Theory we found clips of our [ __ ] in
1:28:07 your [ __ ] therefore it's in your [ __ ]
1:28:10 but it's
1:28:13 not and so so we'll see ultimately it
1:28:18 comes down to who's got the better
1:28:19 lobbyists
1:28:21 right and also the other thing about the
1:28:23 New York Times
1:28:25 It's mighty convenient that after they
1:28:27 did aund million deal with Google they
1:28:29 sued open AI right I'm sure I'm sure
1:28:33 Google had nothing to say about about
1:28:36 yeah no you probably I wouldn't I
1:28:38 wouldn't cut a deal with open AI I mean
1:28:41 haven't they trained on all your data I
1:28:43 mean I'd sue them if I were you anyway
1:28:45 you do you here's your $100 million see
1:28:48 you next week
1:28:49 [Laughter]
1:28:53 Bob not that bitter or cynical or
1:29:05 anything I wonder how all these will
1:29:07 continue develop after the war will
1:29:10 start what what are you talking
1:29:15 about I want to create one with a lot
1:29:18 strong Russian
1:29:23 accent
1:29:25 uh but you need a paid act before you
1:29:28 can
1:29:28 upload you need a paid
1:29:32 act before you can upload to 11 Labs oh
1:29:36 paid account yeah you
1:29:39 do you
1:29:42 do that's the thing about the media the
1:29:45 media stuff the audio and the video
1:29:46 stuff right now is pretty expensive I
1:29:49 wonder how these will continue to
1:29:51 develop after the war will start I don't
1:29:53 know what that means still don't really
1:29:55 care it did in a
1:29:58 way but small artists barely get any
1:30:01 opportunities for live
1:30:04 shows yeah so I mean I have a degree in
1:30:09 acting I started and ran a theater
1:30:12 company I'd work for 3 months to put on
1:30:14 a show that 17 people would come to
1:30:17 see if you're in the Arts and you have
1:30:21 some expectation that you should be
1:30:23 given a [ __ ] career
1:30:27 tough
1:30:28 [ __ ] [ __ ]
1:30:34 grind no one owes you [ __ ] anything
1:30:37 like get on Tik Tok and get you know
1:30:40 500,000
1:30:43 followers and get them to come to your
1:30:45 show if small artists can't get people
1:30:47 to come to their show it's because they
1:30:50 don't have talent or they're not
1:30:51 hustling
1:30:53 sorry it's the entertainment
1:30:55 business for every job there are 10,000
1:30:59 people that would do away with a family
1:31:02 member to have a tenth of an ounce of a
1:31:10 career it is it is not a career for the
1:31:13 faint-hearted and you know you don't go
1:31:16 into
1:31:17 that
1:31:20 world because you expect shit's going to
1:31:23 be given to you
1:31:27 like go live in New York for three years
1:31:30 and and try to get your ass on a
1:31:32 stage try to get yourself an
1:31:37 agent I did it for two years and I'm
1:31:40 like [ __ ] it and started a theater
1:31:41 company I we'll do our own
1:31:44 shows I did that for four and a half
1:31:50 years I don't do do you think I ever had
1:31:53 a complete completely sold out run maybe
1:31:57 once in 4 and A2
1:32:03 years it's a brutal business and it's
1:32:07 just been made more brutal
1:32:09 but here's the thing people with
1:32:14 talent that get AI
1:32:16 literate are going to be like [ __ ]
1:32:23 superheroes
1:32:24 because if you can put on a live show
1:32:27 and you can wield this
1:32:30 technology and you can spin things up in
1:32:33 real time and you can do big
1:32:35 performances and you've got talented
1:32:37 people around you and you can Market it
1:32:41 and you can get venues and you can get
1:32:43 asses in the
1:32:45 seat with very little money talented
1:32:49 people are going to be able to put on
1:32:52 unbelievable spec
1:32:56 spectacles with original music every
1:32:59 night and original video every night and
1:33:01 Original Stories every night but it
1:33:03 still kind of has a
1:33:09 theme so I don't have a lot of sympathy
1:33:11 for but you know small artists have
1:33:14 tough times selling out tickets [ __ ]
1:33:16 get out there and grind you
1:33:20 know you you came you came to the wrong
1:33:23 out of work actor for for you know no
1:33:26 one owes anyone's
1:33:30 [ __ ] I'm as empathetic as as the next
1:33:34 guy I'm I'm I think I'm I index high on
1:33:38 the empathy empathy
1:33:40 scale and I I also have very little
1:33:45 tolerance for just expecting [ __ ] to be
1:33:48 handed to you especially in the
1:33:51 Arts especially in the Arts
1:33:55 go watch Corey Sandler just she's
1:33:58 [ __ ] grinding right now right she's
1:34:01 doing her Pottery she's doing cooking
1:34:04 singing now she's doing gpts she's just
1:34:07 cranking cranking cranking
1:34:12 cranking if you've ever done
1:34:16 that you know in a year goes by and then
1:34:19 3 years go by and then a decade goes
1:34:22 by and you're still
1:34:26 going you've essentially
1:34:29 won but when you're in the middle of it
1:34:31 it doesn't feel like you're winning
1:34:32 you're like I just need that [ __ ]
1:34:34 break so I'm going to keep going because
1:34:36 you're called to do
1:34:40 it all right anyway we're supposed to be
1:34:42 talking about
1:34:45 AI what did I put them is that water or
1:34:48 or Schmutz I think it's
1:34:52 Schmutz it's like so schmutzy I thought
1:34:55 it was a
1:35:00 button and ticket prices oh my don't
1:35:04 excellent points Kyle you're the best
1:35:05 thank you Vegas Moxy Moxy thank you for
1:35:07 being a subscriber also I missed the
1:35:09 rant it's still going I've been ranting
1:35:12 now for two
1:35:13 hours or an hour and a
1:35:16 half someone asked me about copyright
1:35:18 Joker and I just I haven't really
1:35:23 stopped
1:35:27 Oh Lordy bring it Kyle oh yeah who reads
1:35:31 the New York Times love your passion
1:35:33 Kyle thank you
1:35:34 Edgar what is the new Sora a what is
1:35:37 this new Sora AI technology I'll show it
1:35:39 to you oops not sure what I posted it
1:35:44 was a good rant he's still
1:35:49 going H let's go look at what this new
1:35:51 Sora [ __ ] is so I'll show you this I'll
1:35:54 show you the Sora
1:35:55 [ __ ] and then I'll show you the 11 Labs
1:35:58 response to it which I thought was super
1:36:02 cool and I'll also show you how the Sora
1:36:04 stuff works because the sorus stuff is
1:36:06 really
1:36:07 significant it's really
1:36:10 significant it's so funny too because
1:36:12 it's
1:36:13 like when when open AI puts out
1:36:16 something like
1:36:18 Sora there's two kind of naysayers right
1:36:21 there's the naysayers that are like the
1:36:23 oh we're we're all assumed everyone's
1:36:24 going to die there's that that's to be
1:36:27 expected that's kind of boring but then
1:36:29 there's the technology naysayers that
1:36:31 are like it's not that big a deal I'm
1:36:33 like well if it's not that big a [ __ ]
1:36:35 deal where's your version of it you
1:36:39 know the [ __ ] Chief dude from meta
1:36:43 came out like three days after Sora
1:36:45 that's not really that SI technology is
1:36:47 not really that significant you know
1:36:50 it's um and they're just taking some
1:36:52 mathematics and they're applying it in
1:36:54 in a in a both a parallel and a
1:36:56 serialized fashion over time and you
1:36:58 know they claim to be simulating reality
1:37:00 but you know from from a true
1:37:02 perspective they're they're not actually
1:37:03 modeling a 3D World and they're not
1:37:05 actually simulating the physics of the
1:37:07 world they they're just approximating it
1:37:09 in a diffusion kind of model and you
1:37:11 shut the [ __ ]
1:37:14 up
1:37:16 Jesus go look at the video where's
1:37:21 yours if it's that trivial
1:37:25 show us
1:37:27 yours so anyway let's go look at
1:37:30 it I'm [ __ ] Saucy tonight I'm Saucy a
1:37:34 [ __ ] had it
1:37:37 right good we
1:37:40 [Laughter]
1:37:44 good all right so if you want to see
1:37:46 this Sora stuff go to
1:37:48 open.com SLS
1:37:52 o
1:37:54 and then you get
1:37:57 [Laughter]
1:37:57 [Music]
1:38:02 this all
1:38:05 right so there's our fancy little
1:38:12 website okay before I show you any cool
1:38:14 video other than that background there
1:38:16 which is pretty cool that's sort of
1:38:22 video this first
1:38:26 sentence for
1:38:28 me is a huge [ __ ] hint because the
1:38:33 second sentence talks about video right
1:38:37 this is a creating video from text right
1:38:41 is what the product is and their first
1:38:43 sentence has nothing to do with
1:38:46 video what does it have to do with we're
1:38:49 teaching AI to understand and simulate
1:38:52 the physical world in motion
1:38:54 with the goal of training models that
1:38:56 help people solve problems that require
1:39:00 real world
1:39:06 interaction like I have to when I first
1:39:09 read that I read that sentence like 10
1:39:13 times wait you're what we're teaching
1:39:16 the AI to understand and
1:39:19 simulate the physical world in motion
1:39:23 with the goal of training models to help
1:39:26 people solve problems that require real
1:39:29 world inter inter what what does that
1:39:31 mean does that mean they're going into
1:39:34 like Unreal Engine 5 creating worlds
1:39:37 that have physics on it and training
1:39:40 like like just letting those worlds
1:39:43 interact and exist with some sort of
1:39:45 characters in there and training the AI
1:39:48 on that are they are they creating
1:39:51 simulated physics world with material
1:39:53 like what what does that even
1:39:56 mean with the goal of training models to
1:39:58 help people solve problems that require
1:40:00 real world interaction so what that says
1:40:02 to me
1:40:03 is this thing this video [ __ ] is the
1:40:08 it's like it's like the tippity tip of
1:40:10 the iceberg of of whatever they're
1:40:13 doing so then it says introducing Sora
1:40:16 our text to video model Sora can
1:40:19 generate videos up to a minute long
1:40:21 while maintaining visual quality and
1:40:23 adherence to to the user's prompt and
1:40:24 then this is the this is kind of the
1:40:26 hero clip that they they keep
1:40:29 showing this is a 59
1:40:34 second and even if this took an hour to
1:40:43 render this is pretty
1:40:46 something look at there there's some
1:40:48 weirdness in her feet but most of the
1:40:50 other stuff here is pretty coherent and
1:40:53 for me the thing that Impressions me the
1:40:54 most is all of the signs in the
1:40:56 background stay and all of the
1:41:00 people are as they started out and then
1:41:03 that edit that cut to the closeup of her
1:41:07 that's not an edit the Sora model that's
1:41:09 part of this 592
1:41:14 generation it cuts it cuts its own
1:41:17 scenes together as part of the
1:41:22 generation
1:41:25 so that's that
1:41:27 one that one I think's a piece of crap
1:41:30 this one's
1:41:33 amazing again this is a single
1:41:37 prompt with all these scenes that
1:41:40 generated all these
1:41:52 scenes I mean good Lord
1:42:00 right prompt a movie trailer featuring
1:42:04 The Adventures of a 30-year-old Spaceman
1:42:07 wearing a red wool knitted motorcycle
1:42:11 helmet look he's got a red so it was
1:42:14 funny someone said oh it's not realistic
1:42:17 his his helmet is knitted it's in the
1:42:20 prompt it's in the
1:42:22 prompt he's got a red knitted motorcycle
1:42:25 helmet why cuz I don't know someone
1:42:27 prompted it that
1:42:31 way look at
1:42:33 this this drone shot of Big Sir look at
1:42:37 the look at the lighthouse look how it's
1:42:39 not moving and it's not warping and
1:42:41 morphing notice how the detail in the
1:42:43 cliffs is still there the detail in the
1:42:45 waves is still
1:42:49 there so even if this [ __ ] is rendering
1:42:52 at 60 Seconds per
1:42:59 second you know that's a 10-minute shot
1:43:02 rather than hiring a drone pilot to go
1:43:05 out fly around there or or paying the
1:43:08 stockhouse to do
1:43:11 it and you know the shit's going to get
1:43:14 faster if you don't think it's going to
1:43:16 get
1:43:17 faster if you if you don't think it's
1:43:19 going to go faster go to here we'll go
1:43:22 here watch this
1:43:24 gro.com
1:43:26 gq.com
1:43:29 have you seen
1:43:32 this watch this
1:43:35 um let's
1:43:38 see write
1:43:40 me
1:43:42 three
1:43:44 sonets about three people who each love
1:43:50 the other to
1:43:58 look how fast that went Sonet one the
1:44:02 engineer in numbers and in Street the
1:44:04 lover of concrete and the true with
1:44:06 every beam and girder does reveal the
1:44:09 strength and beauty of designs so new
1:44:11 yet in his heart a poet's Soul resides a
1:44:15 tender loving artist he may be so this
1:44:18 is this is not this is just using mix
1:44:20 drill's 8X 7B
1:44:22 model 524 tokens a second 524 so like
1:44:27 350 words a second to generate the
1:44:31 response it's not copy and paste it it's
1:44:34 generating it like
1:44:36 that so this [ __ ] that's too slow right
1:44:39 now it's not going to be that slow for
1:44:40 long anyway so
1:44:43 Sora what else is there's there's
1:44:46 another one that I think is super
1:44:49 impressive this one's super impressive
1:44:51 so the The Prompt here
1:44:54 is
1:44:55 um closeup of a 24-year-old woman's
1:45:00 eye
1:45:02 in maresh during magic
1:45:05 hour and then okay so you're like okay
1:45:08 that's cool but then wait
1:45:10 right there the reflection in her Iris
1:45:14 or in her yeah in her iris is maresh at
1:45:22 sunset
1:45:25 ah holy
1:45:28 [ __ ] what does this thing know
1:45:31 well they're understanding and
1:45:34 simulating
1:45:35 reality or physical motion this is this
1:45:39 is the one that blew me
1:45:42 away did you see that wait I'll wait for
1:45:45 it to
1:45:46 Loop
1:45:52 right
1:45:55 when the dark building goes by outside
1:45:57 the train you see the reflection of the
1:45:59 woman filming this video The Prompt here
1:46:04 is Reflections in the window of a train
1:46:06 traveling through the Tokyo
1:46:14 suburbs
1:46:17 and there she
1:46:21 is that's [ __ ] B Bonkers so that's
1:46:25 Sora so then 11
1:46:31 Labs 11 Labs two days after that came
1:46:39 out drop this
1:46:43 thing so what 11 Labs said was gee isn't
1:46:46 there something wrong with this super
1:46:48 sexy video like say it's
1:46:52 quiet so they said we have a new product
1:46:54 coming out you might want to get on the
1:46:56 wait list it's called text to sound
1:46:59 effects or text to audio or text to
1:47:01 whatever it's called but
1:47:03 basically they use a text to audio
1:47:07 generation model to create audio for all
1:47:11 the Sora videos or you know this this
1:47:14 edit of
1:47:22 them
1:47:28 that one's
1:47:30 amazing so's
1:47:45 that in a place beyond imagination where
1:47:48 the Horizon kisses the heavens one man
1:47:51 dares to Journey where few have vent
1:47:53 ured armed with nothing but his wit and
1:47:56 an unyielding Spirit he seeks the
1:47:59 answers to Mysteries that lie beyond the
1:48:22 stars all of these sound effects were
1:48:24 generated by 11 Labs without
1:48:35 modification well it's not that
1:48:37 impressive you know from a technical
1:48:39 perspective all they're doing there is
1:48:41 they're taking this uh algorithm that
1:48:44 came out in in
1:48:45 2020 is the paper is is the is the they
1:48:49 take the string algorithm and
1:48:51 then yeah
1:48:54 yep exactly it's not that
1:48:58 impressive go watch a 4 second Runway ml
1:49:02 video and then go watch these and I
1:49:04 don't give a [ __ ] if it took an hour to
1:49:05 render these and you can render a Runway
1:49:08 ml video really quickly these are like
1:49:11 usable commercially
1:49:13 usable right Runway mLs are not unless
1:49:16 you're making some sort of you know
1:49:17 Twisted anamorphic an you know morphing
1:49:21 sort of
1:49:22 thing
1:49:24 um so let's go look at how these things
1:49:26 work or how this system works because
1:49:29 it's [ __ ] Bonkers and I don't give a
1:49:32 [ __ ] if if they're taking two relatively
1:49:35 simple mathematical Concepts and
1:49:37 smashing them together they're the ones
1:49:39 that did it
1:49:42 first and no one else has got anything
1:49:44 close to this people are [ __ ] all
1:49:47 over open a well they don't they're
1:49:48 doing this and they're doing that and
1:49:49 they're doing this well they're their
1:49:51 [ __ ] GPT is still better than your
1:49:53 [ __ ] chat
1:49:56 gbt and honest to God I swear to God as
1:50:00 soon as someone comes out that's better
1:50:01 than open AI I'll shut the [ __ ] up about
1:50:08 him but ain't no one caught him
1:50:11 yet and when they do [ __ ] like that like
1:50:13 I think they did this because I think
1:50:15 they knew that something was coming from
1:50:17 Google that was going to take the little
1:50:19 Shine off their their
1:50:22 marble
1:50:25 so cuckoo all right here's what we're
1:50:28 going to do learn more in our technical
1:50:31 report I want I'm not going to go deep
1:50:33 into this
1:50:43 but okay here's video This is how video
1:50:47 works if you don't know how video works
1:50:49 it's a stack of images
1:50:52 right
1:50:55 thing called Persistence of
1:50:57 vision the sort of minimum Persistence
1:50:59 of vision is 24 frames a second which is
1:51:02 what movies are generally rendered at
1:51:05 and that's
1:51:07 why movies tend to have that kind of
1:51:09 warm analog kind of feel cuz it's like
1:51:11 right at the edge of where you know they
1:51:14 don't flicker it just looks like motion
1:51:15 to us and then videos 30 frames a second
1:51:19 video games are 60 or 120 frames a
1:51:22 second
1:51:26 but our our eyes above 20 I think it's
1:51:30 22 can't can't really see the individual
1:51:33 frames we just see the
1:51:35 motion so
1:51:38 what and and so if you've been on here
1:51:41 for a while earlier I showed how
1:51:43 diffusion models work right so it
1:51:46 amalgamates a bunch of images that have
1:51:48 been noised into a single noisy image
1:51:52 and then it Den noises that image it has
1:51:55 no idea what's in the image none which
1:52:00 is why you get weird hands and weird
1:52:02 feets and a nose on the side of the head
1:52:04 and an extra leg is because it just
1:52:07 smashed some [ __ ] together denoised it
1:52:09 and what it gets it gets it's completely
1:52:11 unaware of of what it
1:52:15 is what open AI is doing here
1:52:19 is they're in the coding process when
1:52:24 they when they encode it on video when
1:52:27 they're teaching it about video they're
1:52:29 taking each frame they're doing visual
1:52:31 encoding of it and they're breaking it
1:52:33 into what they call Patches so the front
1:52:36 of this grid here is a single frame that
1:52:39 they've broken into patches and they do
1:52:41 that for every frame so they've got this
1:52:44 kind of
1:52:45 volumetric understanding they've got
1:52:47 this understanding of all of the
1:52:49 different
1:52:51 items sub elements of the image over
1:52:57 time and then if you take any one of
1:53:00 those little
1:53:01 patches the way they're render the way
1:53:04 they're maintaining continuity is that
1:53:07 they bring a single patch in and and
1:53:11 they have like a 100
1:53:15 frames so like three seconds of video
1:53:18 and I don't I assume they're going front
1:53:20 and back or maybe they're just going
1:53:22 back and I don't know it doesn't [ __ ]
1:53:24 matter but let let's say down here is
1:53:27 like the wheel of a sports car in that
1:53:29 little patch they can look at the wheel
1:53:32 of that sports car over time and
1:53:34 maintain continuity across it so they're
1:53:36 doing micro diffusion
1:53:39 models but they're checking them over
1:53:41 time so they're they're improving
1:53:43 continuity over time for a little patch
1:53:45 so that's how they're maintaining so
1:53:48 what it means is this
1:53:50 system understands this world it
1:53:53 understands that it's a reef with a fish
1:53:55 in it and a scuba diver in the
1:53:58 distance and it understands what they
1:54:00 are over time and because they're
1:54:02 training the model to understand the
1:54:06 physical
1:54:08 world it actually understands the
1:54:11 relationship between a scuba diver and a
1:54:14 fish in a
1:54:17 reef holy [ __ ] [ __ ]
1:54:21 Batman
1:54:24 this has nothing to do with video
1:54:27 although it does it makes really cool
1:54:34 video every time I learn about how they
1:54:36 do this [ __ ] it it it's like who's
1:54:39 figuring this [ __ ] out well here's the
1:54:41 deal people have been figuring this [ __ ]
1:54:44 out for the past three
1:54:46 decades it's just now coming to
1:54:51 fruition
1:54:53 and and as overhyped as it is this stuff
1:54:57 is coming to fruition right now because
1:55:00 open AI released this thing called chat
1:55:03 GPT on November 30th 2022 and 100
1:55:06 million people started using it within 6
1:55:09 weeks Word of Mouth
1:55:12 only and that changed
1:55:15 everything because the focus of every
1:55:18 single tech company starting at the
1:55:21 beginning of 2023 until today today is
1:55:24 100% on generative
1:55:26 AI not just the AI that's been sitting
1:55:29 in our machines for 15 20 years
1:55:32 generative AI the stuff we get to play
1:55:35 with they're all focused on it so what
1:55:38 we're starting to witness right now all
1:55:40 the all that list that I read at the at
1:55:42 the beginning of the
1:55:43 live is a year of development of people
1:55:47 dumping tons of money and tons of focus
1:55:49 on these tools that's why they're all of
1:55:51 a sudden starting to get better and
1:55:53 better and better all of a sudden it's
1:55:55 like how is it possible this is all just
1:55:57 happening right now because it's been
1:55:59 building for years and then starting
1:56:01 last year all this Focus like laser
1:56:04 focus went on generative
1:56:06 AI so all the development's turning this
1:56:09 way all the money's turning this way all
1:56:11 the research is turning this
1:56:14 way and you've got someone like open AI
1:56:16 that you know they've been at it for for
1:56:19 five or six years in a very focused way
1:56:22 they put out a thing that they didn't
1:56:24 think was a big
1:56:26 deal they had drinks the night before
1:56:29 they launched chat GPT and said ah maybe
1:56:31 we'll get a blog post or two and a
1:56:33 couple of thousand
1:56:36 comments 100 million users in six weeks
1:56:39 fastest adoption of technology in
1:56:41 history with no
1:56:46 marketing crazy
1:56:51 right B conkers all right I got to go
1:56:55 people what else did I have on here on
1:56:59 my little list my list 11
1:57:03 Labs yeah the Google pro 1.5 stuff I
1:57:06 don't have access to it yet but
1:57:08 like the people that are talking about
1:57:11 it like Ethan mik go if you're not
1:57:13 following Ethan mik go follow Ethan mik
1:57:16 follow him on Twitter follow him on
1:57:17 LinkedIn and go to his blog one useful
1:57:22 tricky Point how many people will make
1:57:24 special spectacles in the future will be
1:57:27 lost in the crowd
1:57:29 many that's art true Kyle grind grind
1:57:32 grind
1:57:34 yep just you know and you know Corey
1:57:38 you're grinding is you know I say grind
1:57:40 grind like a Gary vaynerchuk thing
1:57:42 that's not it it's like you have a
1:57:44 passion for what you do right and it's
1:57:48 not about the commercial success
1:57:50 necessarily like be nice to have I think
1:57:55 as artists we all dream of having it and
1:57:58 then it's art sometimes it resonates
1:58:01 sometimes it
1:58:03 doesn't Tyler Perry calls on
1:58:05 entertainment industry and government to
1:58:07 cral AI before everyone's out of
1:58:09 business
1:58:10 exactly the robots are going to kill us
1:58:13 I'm never going to use it I'm a writer
1:58:16 and I will never use
1:58:20 AI you're not going to use the the most
1:58:23 powerful writing tool in the history of
1:58:24 humanity really
1:58:27 Skippy
1:58:29 really you should try
1:58:32 it it's pretty [ __ ] miraculous I've
1:58:35 written seven screenplays I'm kind of
1:58:37 partial to the
1:58:41 word I think AI gives me [ __ ]
1:58:45 superpowers you know how I know that cuz
1:58:48 I used it I didn't just [ __ ] about it
1:58:57 so [ __ ] bitter
1:58:59 [Laughter]
1:59:06 tonight oh
1:59:14 Lordy
1:59:21 again
1:59:24 I'm extremely late but I have a good
1:59:25 excuse I was making a GPT one useful
1:59:28 thing yeah one useful thing.
1:59:30 org is Ethan mullock's
1:59:34 blog just go look at what he's writing
1:59:36 about Gemini yeah Gemini 1.5
1:59:40 Pro like Ethan mullik like he's he's
1:59:43 really even keeled and he's like here's
1:59:45 some research that shows this that and
1:59:47 the other and you know he's very much
1:59:48 he's very much a a uh an AI Optimist and
1:59:54 you should be using AI but he's he's
1:59:56 thinking about it very pragmatically go
1:59:58 look at what he's writing about the
2:00:01 million token context window in Gemini
2:00:03 1.5 he's like this is different shit's
2:00:07 different feels different something's
2:00:09 different this is
2:00:11 different if Ethan mullock's saying
2:00:13 something's changing something's [ __ ]
2:00:16 changing so anyway all right I got to go
2:00:19 people thank you to all of you thank you
2:00:22 to the Irregulars thank you to the mods
2:00:24 I know this was a non-standard evening
2:00:26 but
2:00:27 but I had I had to rant I had to rant I
2:00:31 just had to do it people
2:00:33 um if you want to support the channel
2:00:36 subscribe to the lives buy one of the
2:00:38 video series in the corner or the video
2:00:40 series in the corner follow my channel
2:00:43 um go
2:00:48 to go to there the salon. a check out
2:00:52 this community this is a community I
2:00:54 started the week after chat GPT launched
2:00:57 it doesn't cost anything but go read
2:01:00 about who we are and what our values are
2:01:03 and if you want to hang out with people
2:01:04 that are up to [ __ ] in AI this is the
2:01:08 group and by up to [ __ ] I don't mean
2:01:11 like they figured it out this is a group
2:01:13 that understands that no one's got it
2:01:15 figured out and they're just exploring
2:01:17 and they're playing and they're
2:01:18 collaborating and they're supporting one
2:01:20 another and they're not being [ __ ]
2:01:22 to one another and in fact when anyone
2:01:25 in here is slightly assholish it's like
2:01:27 this weird thing cuz no one in here's
2:01:29 like
2:01:33 that so if you
2:01:35 can put down your [ __ ] hat and
2:01:38 connect with people and you want to
2:01:40 learn some [ __ ] this is a remarkable
2:01:43 group non-standard is the standard all
2:01:46 right everybody I am out of here
2:01:48 Amelio's wife great to see you Vegas
2:01:51 Moxy thank you as all always good good
2:01:53 good Danielle thank you investor
2:01:55 friendly agent peace out laughing
2:01:58 Thunder thank you thank you thank
2:02:04 you Brian Whitney sherid D thank you
2:02:08 y'all are so swell you're so swell I
2:02:11 just love you up Lord digital Gods good
2:02:14 evening uh LinkedIn office hours
2:02:16 tomorrow 11:00 a.m. mountain time go to
2:02:19 my LinkedIn Channel @ Kyle Shannon on
2:02:22 Linkin in and just check in the morning
2:02:24 I'll have uh the the link there or you
2:02:27 can go to the Irregulars
2:02:29 Channel at the
2:02:31 salon um and I have the link the the
2:02:35 link to the Google meet for my AI office
2:02:37 hours to M tomorrow against the norm
2:02:40 been watching you every night thank you
2:02:41 very much keep coming back keep coming
2:02:43 back it it does get better no it doesn't
2:02:45 this is pretty much
2:02:47 it Cory Sandler Pottery good night TK
2:02:51 good to see you Joker
2:02:52 a human being good to see you
2:02:55 Debbie
2:02:57 uh all right I think I got most of the
2:02:59 people if I missed you I love you anyway
2:03:02 you know that Mr it good to see you
2:03:04 peace out people have a good night bye