aiLL 09/21/23 - The Future of AI: From Janky to Sophisticated

Description

Chapters

Transcript

0:01 foreign
0:14 [Music]
0:22 [Music]
0:26 getting the glasses cleaned off so I can
0:29 see
0:30 can I see now can I see can you see me
0:35 where are you people where are you
0:37 people why are you late
0:39 why are you late AI learning lab
0:44 we come here every night you're not
0:46 supposed to be late why are you late
0:48 all right
0:53 it's a little warm
0:56 it's okay I've got my iced coffee
1:03 oh yeah my my voice was having a rest
1:06 yeah that was that was pretty pretty
1:07 nasty last night
1:09 I think it's okay now
1:11 we'll see we'll see when I get talking
1:14 we've been hanging around here for for
1:16 an hour
1:18 hey Ted looks at life what's happening
1:20 Omar
1:22 [Laughter]
1:24 welcome everybody the beginning is near
1:29 I don't know if you knew that
1:31 but it is
1:38 I was about to give up don't give up
1:41 Stevo
1:42 I had a uh it was Denver startup week
1:45 and tonight was the uh
1:48 the last night bash I tend not to go to
1:51 Denver startup weeks because
1:53 I'm busy running a startup
1:58 so I went to the party it was fine I saw
2:00 a couple of people but
2:02 I am not a fan of uh
2:04 those Mega networking events they just
2:07 don't I went to two two uh sessions
2:10 today one was okay the other one was a
2:12 copy waste of time
2:17 oh
2:18 oh so anyway Robert Rossi what's
2:21 happening
2:23 Ted looks at life waiting for the Great
2:24 Pumpkin hey that's a fat joke isn't it I
2:28 resemble that remark I resemble that
2:30 remark see yeah yeah
2:33 his head was the size of a giant pumpkin
2:37 he looked like Rosie O'Donnell
2:40 Tom Arnold maybe yeah see yeah you
2:43 palooka
2:47 pouring water down my face
2:49 a terrible drinking problem
2:57 so welcome everyone my name is Kyle
3:00 Shannon this is the AI learning lab
3:02 you know a lot of people out there in
3:04 this world they're saying
3:06 the robots are gonna kill us the end is
3:09 near
3:10 no I disagree the beginning is here
3:15 we're at the beginning of this thing if
3:17 you're feeling behind if you're feeling
3:18 left out or you're feeling like everyone
3:20 knows more about AI than I do they don't
3:23 nobody knows more than anyone right now
3:25 everyone's trying to figure this [ __ ]
3:27 out and everyone that thinks they
3:29 haven't figured out they're going to get
3:30 their ass handed to them like say
3:32 mid-journey
3:34 because because Dolly three was
3:37 announced and it can do text and it can
3:40 do coherence meaning you can identify a
3:43 character
3:44 at the beginning of a a series of
3:46 prompts and it'll maintain that
3:48 character throughout that series of
3:49 prompts
3:50 and uh it can do really complicated uh
3:55 object integration into the scene
3:58 and and then today
4:01 Elon Musk because he's Elon Musk and he
4:04 doesn't want old Sam Altman to to get
4:06 too big in his britches tweet it out
4:10 um
4:11 mid-journey's coming up with something
4:12 pretty cool soon too
4:15 so anyway there's there's this is this
4:18 is gonna be a crazy fall if you think
4:22 all of these tools we we are we are
4:25 moving out of so one of the things I say
4:27 on a channel here because I'm just a
4:29 smart ass I'm a smart aleck I'm a smart
4:31 alec and one of the things I say here I
4:34 say it pretty much every night I say
4:38 all these AI tools are just Yankee
4:40 pieces of [ __ ] I do I say it I say it
4:42 just like that I say it so much somebody
4:46 made me a graphic for it with a little
4:48 poop it's poop on top of the word janky
4:53 we are entering
4:56 like in the next three months
4:59 I may have to retire the janky pieces of
5:02 [ __ ] graphic
5:04 I think these tools are going to get
5:06 pretty sophisticated pretty quick
5:09 so anyway listen if you're new here
5:10 welcome everyone got a lot of folks in
5:12 here
5:13 um if you see all the hearts going
5:15 that's people tapping the screen like
5:16 mad now you're thinking like what he
5:19 hasn't said anything yet why are they
5:21 what what is there's nothing to like
5:22 here exactly
5:24 those are the Irregulars there's people
5:27 that show up here night after night
5:28 after night including me
5:31 and they're just not right they're just
5:33 not right they're irregular
5:35 and so so they have little games they
5:38 play
5:39 don't we have a we have a t-shirt for
5:41 them we made like a little t-shirt
5:43 graphic see they're not weird they're
5:46 Irregulars
5:47 so they're trying they're trying to beat
5:49 their record from last night what are
5:51 you up to you're up to one point six
5:53 thousand so you're doing good
5:58 um here's what we do here if you're new
6:00 to this generative AI stuff if you've
6:02 not played with chat GPT go to
6:04 chat.openai.com
6:06 and just start playing while you listen
6:08 to me ramble you'll you'll see what goes
6:10 on in the channel
6:12 um I'll answer any questions you have
6:14 I'll demo stuff if you want to see it
6:17 um and the the uh the non-profit uh
6:20 fundraising person in here
6:23 um dm'd me before the things that you
6:25 wanted me to talk about the future work
6:26 I don't know if you're in here and but
6:27 if you are
6:29 just sort of pop in kind of what you
6:30 want me to talk about I'm happy to talk
6:32 about that but go to chat.openai.com
6:34 start playing with it and when you're
6:36 like but Kyle I don't know what that
6:38 means I'm not very technical I don't you
6:41 say play with these tools you can't play
6:42 with these tools this is technology this
6:45 is artificial intelligence you have to
6:47 be a mathematician to do that no you
6:51 don't
6:51 you don't
6:53 chat CPT is just like the Google box you
6:55 know the Google box where you put [ __ ]
6:57 in and you get a bunch of ads
6:59 this you put [ __ ] in and you get back
7:00 remarkable stuff
7:03 so go play with it if you don't know
7:06 what to say there go to prompts.chat
7:08 that thing
7:09 will teach you a little bit about how
7:11 these things work
7:13 and then I'll talk about those in a
7:14 second but anyway
7:16 um pop any comments you have down below
7:18 a bunch of stuff going on oh cool little
7:20 cool little star thank you digital Gods
7:22 appreciate that what do you think about
7:24 Bard using it to search Gmail yeah so
7:27 so I did finally get it to work today so
7:30 here's here's the skinny on Bard so Bard
7:33 my top three right now are chat.openai
7:35 uh that's the the chat GPT
7:39 I think it is worth paying the 20 bucks
7:41 for that oh by the way
7:43 I want to go see something
7:46 there's a guy on Tick Tock named Abram
7:51 apron
7:52 a Brawn like LeBron a brawn
7:56 who's an open AI Ambassador and he did a
8:00 video today on Dolly three
8:02 and he said dolly three is going to show
8:04 up as a plug-in
8:06 not as a mode like code interpreter but
8:11 I'm gonna go look for something I'm
8:12 gonna go check and see if it's there
8:15 it's not
8:16 but I am going to go look in plugins for
8:19 a second hold on please let's
8:21 let's see if we got a little action
8:24 snuck into our
8:26 chat GPT we could we could play with
8:29 this stuff for free
8:31 d-a-l-l
8:34 Dash e
8:37 nope nothing all right
8:40 it's worth a shot okay anyway
8:43 um yeah so go play with these tools
8:47 um the question about bard
8:49 so if you don't know bard
8:52 yesterday
8:54 released a new version of bard
8:57 which is good they improve the interface
9:00 it still seems to be decently functional
9:03 and they added a thing called extensions
9:05 and extensions is pretty wild they're
9:08 kind of like chat CPT plug-ins except
9:11 what does Google have well Google has
9:14 all of your Gmail and all of your Google
9:17 Drive
9:18 and travel
9:20 and all that sort of happy horseshit so
9:25 so you can now
9:28 um
9:30 search across your Google Drive
9:33 now here's the deal
9:36 from my limited experience with it it
9:38 ain't great yet but the fact that you
9:41 can essentially GPT across all of your
9:45 Google Drive and and your emails you
9:48 know your gmails from history it's
9:50 pretty wild
9:52 let me get that let me get logged into
9:54 that because I oh oh here's the here's
9:57 the other thing about the Google thing
9:58 so all day yesterday I couldn't get it
10:00 to work I even went on the the salon
10:02 Discord and I I said can anyone help me
10:04 make this work like the old man I am hey
10:08 Marge how do you make the tick tock work
10:11 is this is this thing on how do I know
10:14 if it's on I see myself I'm right here
10:18 um so I was doing that a little bit
10:20 yesterday but
10:23 um
10:24 if you have a Google workspace account
10:28 so basically if you pay for Google
10:31 you don't get these new services
10:33 which is stupid
10:35 now I understand why it's like that
10:37 because I would imagine that the
10:39 permissions for the workspace you know
10:43 if you're you know
10:45 gpting across Google Drive for multiple
10:48 users is probably a [ __ ] nightmare so
10:50 I kind of get why it's not there but
10:52 they didn't they don't give you any
10:54 indication that
10:56 like there's no dialog box that pops up
10:58 and says this is a Google workspace
11:00 account go [ __ ] yourself it's just like
11:02 just doesn't work so let me see if I can
11:04 get logged in here
11:07 so you go to bard.google.com
11:16 and I gotta switch over to
11:20 my other account
11:27 accept my problem is LastPass has the
11:30 wrong password in it so I think
11:34 open the Gmail app
11:36 continue
11:39 okay
11:42 so uh let's go look at this
11:45 so here we are at Bard so this is Bard
11:47 experiment
11:48 just so you just so you know this is not
11:50 a real thing this is just an experiment
11:53 they reserve the right to steal all your
11:55 data and turn this thing off just when
11:57 you get addicted to it
12:00 um
12:01 so it's just like that now if you go
12:05 bard.com so so if you're logged into a
12:08 single account not a workspace account
12:11 you go to bard.com
12:14 bard.google.com extensions
12:17 extensions
12:20 it should then take us
12:23 huh
12:25 it didn't this is what was happening oh
12:29 wait it re-logged me into the other
12:30 account no I want the other account come
12:33 on you weirdo okay here we go
12:37 good Lord
12:38 you know
12:41 is it so hard to make [ __ ] that doesn't
12:43 suck apparently apparently it is
12:46 okay
12:48 so uh
12:50 uh one conversation many extensions so
12:52 they have Google flights Google hotels
12:54 Google Maps Google workspace so this is
12:58 all of your drive Gmail and Docs
13:02 and then YouTube oh that's let's play
13:04 with YouTube so
13:07 so the way you call these things is you
13:09 call like at YouTube at GMail at drive
13:13 at docs right so these are I've got
13:15 these all turned on okay so go back to
13:18 Bard Bard experiment
13:20 and then I'll go at uh
13:26 at YouTube
13:30 and that was kind of cool so it says
13:32 this this extension's enabled so it's
13:34 recognizing it as an extension
13:37 so let's say at YouTube
13:40 um
13:41 what is
13:43 the last video I uploaded
13:56 um
13:57 query unsuccessful
14:00 well to be fair I might not like this
14:03 might be a dead uh YouTube account hang
14:06 on so let's let's ignore that one uh all
14:08 right let's see
14:10 um
14:12 we'll do at gmail
14:21 what
14:25 um
14:26 what kind of product
14:30 is being sold
14:33 in most of my spam
14:37 email
14:40 let's see if it can do this by the way
14:42 like I I this is not prepared at all I'm
14:44 just I'm just playing here kind of in
14:46 real time with you so this is kind of
14:48 the stupid [ __ ] I cannot determine
14:50 specific products sold out of most of
14:52 your spam emails because the emails you
14:54 have shared do not contain that
14:55 information that's not true
15:00 thank you for your order PlayStation
15:03 so it gives me some
15:05 some links to emails
15:09 I think it takes me over to Gmail yeah
15:13 it takes me into that that's interesting
15:16 all right
15:17 this lovely Ford Flex
15:21 what what did I just close I'll just
15:24 close the window oh good God all right
15:26 whatever
15:27 anyway
15:29 so what I think what what do I think of
15:31 Bard I think that
15:34 um
15:34 the
15:38 the tools are
15:41 weak right now the the janky piece is a
15:44 [ __ ] they're they're weak
15:46 and the large language model that's
15:48 powering them is not great but Gemini's
15:51 coming out so I think they're laying the
15:53 foundation for some pretty amazing [ __ ]
15:55 coming up I hope I hope
15:58 um okay so how this channel works is ask
16:03 any questions you have
16:05 wait tell it that's not true yeah well I
16:08 would have but I closed the window so I
16:10 just bailed I just gave up I'm done I'm
16:12 done with I'm done with the Bard I don't
16:14 know what I'm doing I don't I don't like
16:15 doing live kind of like
16:18 I don't mind doing live [ __ ] but like I
16:21 haven't tested it at all so I haven't
16:22 figured out like how to how to use it
16:24 and I haven't like what what I need to
16:26 do because that Gmail account that I
16:28 have to use to
16:30 be able to see the extensions I haven't
16:32 used it in like eight years so it's like
16:35 a completely dormant account so I don't
16:37 know what's in it I don't know what to
16:38 search for I don't know what would yield
16:40 an interesting result all my good
16:42 stuff's in my workspace account that's
16:43 the one I want to play with but you
16:45 can't you just can't do it they're not
16:47 gonna let you do it you know what I'm
16:48 saying I don't know if you know what I'm
16:49 saying yeah
16:52 all right
16:58 good Lord good Lord all right fantastic
17:01 Bob let's put a nice doggie up
17:04 let's put the AI learning lab up this is
17:07 Ali the AI learning lab ain't she sweet
17:10 she sounds white she's so sweet she's a
17:13 hunting lab she likes to chase ducks
17:15 she's duck chaser
17:18 okay
17:19 so I'm gonna go start answering some
17:22 questions this shirt is really hard to
17:24 read against I might need to go flip on
17:26 a uh um what's it called t-shirt
17:30 a dark t-shirt because this is this is
17:33 hurting my eyes
17:35 my eyes are old and tired people they're
17:37 old and tired they're old and tired
17:42 chat GPT helped me land a big client
17:44 today by creating killer marketing
17:46 campaign Jim Ross Bang
17:49 Jim welcome
17:51 good to see you again that's awesome
17:54 so Jim Jim I met Jim I do uh live office
18:01 hours on LinkedIn every Friday afternoon
18:04 and I met Jim last week or the week
18:06 before
18:07 and he runs a bunch of uh
18:12 storage businesses and so he's been
18:15 using he's been using I said I said well
18:17 how much you use an AI I'll just use it
18:19 a little bit here and there I said well
18:21 how long you've been using it he goes
18:22 since last November I'm like dude you've
18:25 been using it since day one
18:27 you you're Advanced and so now he's like
18:30 out there Landing business with it boom
18:33 we'll wait wait A Change Kyle okay fine
18:35 I'll go change
19:11 all right
19:13 I got my little crescent moon shirt over
19:15 the
19:16 high tide the waves of
19:19 of your
19:22 Silver Fox many music notes many music
19:25 notes
19:30 [Music]
19:36 I don't know he put on his shirt he
19:38 seemed to he tucked at the neck a lot I
19:41 don't know what was going on there
19:45 oh
19:49 I wonder if that's really an office or a
19:51 backdrop oh wouldn't you like to know
19:55 you've seen this before if this was a
19:57 backdrop could I grab a screwdriver off
19:59 a shelf
20:00 nope could I bag it grab a Ziploc baggie
20:03 full of change nope
20:06 [Laughter]
20:12 Kyle is your office a big pile of trash
20:14 shut up shut up don't pick on me
20:17 all right
20:19 do anyone have any insights into a 3D
20:22 printing generative AI not yet
20:24 um
20:25 there's ways you can do it there's some
20:27 there's some
20:29 AI models in the open source space on
20:32 GitHub that are
20:34 there's papers on it there's some early
20:36 models that'll take an image
20:40 and convert it into 3D objects there's
20:42 other ones that will do text to 3D
20:44 objects
20:46 they're they're a little too early right
20:48 now
20:49 um
20:50 but it won't be long it won't be long
20:53 so
20:55 um
20:56 yeah I would say just be patient just
20:59 give it another six months or so and
21:01 then you I mean like like we we're
21:04 within a year of ask for it and it'll
21:06 start printing you know if you ever if
21:08 you've ever owned a 3D printer they take
21:10 for [ __ ] ever so so it's not like you
21:13 ask for it and it shows up like a
21:14 mid-journey image it's like you ask for
21:16 it and then you go to bed and you wake
21:17 up and maybe the print hasn't failed
21:20 and you get something that looks pretty
21:23 cool
21:25 Home Collection
21:28 huh
21:31 why is that pinned
21:35 why is there a multi-carat color area
21:38 rug pinned
21:41 oh
21:43 let's see the CW said the other shirt
21:45 reminded him of rugs that's hilarious
21:48 yeah it's that sure gets a lot of
21:50 compliments in in person it's a nice
21:52 shirt it's got all this embroidery on it
21:54 but it's pretty much impossible to read
21:56 here
21:57 um dude you know you can press the
21:59 yellow heart at the top oh and send Kyle
22:01 a free heart oh
22:03 didn't know that can I sell myself a
22:05 free heart
22:06 right hey nope
22:10 I got 88 free Hearts that's pretty cool
22:13 thanks chair bear I didn't know that was
22:14 possible
22:16 um
22:19 classic Google everything's in beta for
22:21 a decade yeah exactly exactly experiment
22:25 but yeah Gmail was an experiment for 15
22:29 years
22:32 uh
22:34 at one point A buddy of mine Ben Jones
22:37 got hired at Google he still works there
22:39 and uh
22:41 I was having some weird issue with Gmail
22:43 and I reached out to him like dude man's
22:45 Gmail sucks he's like how much you're
22:47 paying for it I'm like oh you're already
22:48 a douche you worked for Google for like
22:50 three weeks and no you're already a
22:52 douche
22:53 yep
23:00 [Music]
23:07 crypto joke Joe just joined Emilio's
23:10 wife what's happening
23:13 Ahoy sir Ahoy irregular cello Ron how
23:16 are you
23:17 what's the best place to start it seems
23:19 like a lot of public-facing products are
23:21 I hope this says pieces of [ __ ]
23:23 are wait wait
23:28 kneecapped oh
23:31 yeah so okay
23:34 so
23:37 it depends what you want to start with
23:39 but if you're just trying to like figure
23:41 out this AI [ __ ] just get your head
23:42 around it
23:44 there's only one yeah there's only one
23:46 you need go to chat.openai.com
23:48 and you don't need to pay the 20 bucks
23:51 for it I personally would because you're
23:54 going to get access to Dolly three when
23:55 it comes out and then you won't have to
23:57 pay for mid-journey
23:59 um you also get access to code
24:01 interpreter or Advanced Data analysis
24:02 you also get access to plugins and you
24:05 get access to gpt4 if you pay for the
24:07 the 20 bucks a month
24:09 um
24:13 it's it's the first one it's the one
24:15 that started all this nonsense if you're
24:16 wondering why everyone's talking about
24:18 AI it's because of chat gbt so they've
24:21 got like a six month head start on the
24:23 industry which with the speed this
24:26 industry is moving six months is a lot
24:29 right so we're gonna see Gemini come out
24:31 which is going to kick GPT for his ass
24:33 and then November 6th or 8th open AI is
24:37 having their developer conference I have
24:40 a feeling they're going to announce some
24:41 things
24:42 so we're going to keep seeing these
24:44 these things do this but GPT right now
24:46 is the one just go start playing with it
24:48 go to prompts.chat that'll teach you how
24:51 to talk to large language models which
24:52 is what all of these things are
24:55 um
24:56 start with parlor tricks like write me a
24:58 poem write me a country song write me a
25:00 marketing plan just just tell it just
25:02 give it you know compare two obscure uh
25:07 philosophical treatises it doesn't
25:10 matter like just just throw
25:14 language pasta at the wall with it have
25:17 it write you a Python program or a
25:19 c-sharp program have a do coding for you
25:21 write some HTML
25:23 rewrite an XML schema depending on what
25:26 you're uh where where you are on the
25:27 geeky scale
25:29 and just start playing with it and then
25:31 once you once you kind of
25:33 say okay this is I think this is a
25:36 little bit more than Google
25:38 then
25:40 start to think about what you do
25:43 and start to query can it do what I do
25:48 and then at some point in there you will
25:50 have your Kevin McAllister moment
25:55 and then then the adventure begins and
25:58 and by the way this is a red pill blue
26:00 pill moment
26:02 if you don't have the Kevin McAllister
26:03 moment it's your blue pill you can just
26:05 sit on the sidelines and say I'll just
26:07 wait for this AI stuff to show up for
26:09 real
26:10 or you dig in and you have that moment
26:13 where you realize
26:15 holy [ __ ]
26:17 these things are something different
26:19 this is something different
26:23 um
26:25 language pasta that's a t-shirt
26:28 a language pasta sort of the language
26:30 pasta at the wall I didn't even know I
26:33 said that
26:36 all right let me add that to the let me
26:39 add that to the Trello board that's a
26:40 good one
26:43 oh that's really funny
26:45 90 tabs let me restore my 90 tabs just
26:48 to give those of you who actually close
26:50 your tabs anxiety
26:51 [Laughter]
27:00 where is my Trello where's Trello where
27:04 are you Trello I guess I'm gonna have to
27:06 type those letters t-r-e all right
27:09 that's all
27:10 by the way is an autocomplete the best
27:13 that's that's you know we've been using
27:15 AI for decades right you do know that
27:18 right like all the autocomplete all the
27:20 spell check the grammar [ __ ] all that
27:21 stuff say I
27:24 um so where am I going here
27:27 t-shirt sayings
27:30 uh
27:31 language
27:34 pasta
27:38 I don't know what that means next time I
27:40 come look at this cue the Jeopardy music
27:42 I guess that was when I changed my shirt
27:44 I wonder if that's really an officer
27:46 backdrop oh that's what that was in
27:47 reference to
27:49 he shows his stuff from the shelves
27:51 anyone have insights into the 3D
27:53 printing generative AI just wait uh
27:56 listen so so Ron the the Insight I have
27:59 is
27:59 go just go to Reddit
28:02 and search for
28:05 um 3D printer AI or you know AI 3D
28:09 prints or whatever just I guarantee you
28:11 there's some subreddits in there where
28:13 there's some Geeks that have figured out
28:15 that if you take this tool and you've
28:18 Stitch it together with that tool and
28:20 you do this it'll generate of this which
28:22 you can bring into Cura slicer and you
28:24 can slice it and you can make your 3D
28:26 print
28:27 I can't imagine that at this point you
28:30 know sort of you know I what we're
28:32 probably like two years into
28:34 gpt3 being around two and a half years
28:37 um
28:38 someone will have solved some version of
28:40 it it's just not gonna be super easy
28:41 right now so if you want to geek out
28:43 just go to Reddit I guarantee you
28:45 there's something there someone's gotta
28:46 probably got a Google collab notebook
28:48 cutting up tonight
28:51 um Tech Becky techie Becky oh techie
28:53 Becky
28:54 10 [ __ ] out of 10 on the name techie
28:58 Becky
29:00 oh that's [ __ ] gold that you know
29:02 what that deserves
29:08 yeah
29:10 okay sorry what's your comment
29:12 just saw that there's a class action
29:14 lawsuit from The Writer's Guild against
29:15 open Ai and Sarah Silverman and lots and
29:20 lots and lots of them did you also see
29:21 that Microsoft said that if anyone gets
29:24 sued for copyright infringement they're
29:26 gonna pay the lawyer bills
29:29 um yeah I don't think these lawsuits
29:33 really stand much of a chance
29:36 um but I think they're gonna keep coming
29:38 for for a couple of reasons one is
29:41 copyright holders are
29:44 rightfully concerned that their
29:47 livelihood
29:49 could be [ __ ] hosed
29:51 like like like it
29:55 at a minimum
29:57 things are going to change dramatically
30:00 if you understand actually how large
30:02 language models work and how embeddings
30:04 work
30:07 what what Chachi PT is not
30:10 is it's not a copying and pasting
30:13 plagiarism machine that's how it's
30:15 characterized by everyone but that's not
30:18 what it is like the original documents
30:21 that were used to train jet chat GPT
30:23 they do not exist in the chat GPT
30:28 um Vector space
30:30 what lives in there is individual words
30:33 that are clustered semantically
30:38 all the all the like words kind of hang
30:40 out together
30:41 and they're they're disassociated from
30:44 their Source documents
30:47 so depending on how you
30:49 prompt them
30:51 you can get them to give you things
30:53 maybe even exact copies of stuff that
30:56 was copywritten but
30:58 I I just I think there's a technical
31:00 argument here that you know this is the
31:02 equivalent of having books in a library
31:04 and people going in and learning and
31:06 then being able to recite something you
31:08 know
31:09 I can sing all of the words of Gilda
31:11 radners let's talk dirty to the animals
31:14 but it you know
31:17 it doesn't mean
31:18 I'm a copyright infringer now if I
31:20 record it and try to sell it then I am
31:23 I gotta the reason that one's on my mind
31:25 is I got a one of my tick tocks one of
31:27 these lives I played the Gilda Radner
31:29 let's talk dirty to the animals thing I
31:31 played the whole song I got a copyright
31:33 strike on that video on on YouTube
31:37 um is AI out of control and can't turn
31:39 back due to lack of policies
31:42 um no not yet
31:44 but we can't turn back
31:47 um for a couple of reasons one is
31:52 so so when Chachi PT launched
31:55 um
31:57 nobody including open AI thought it was
31:59 going to be a big deal nobody
32:02 and then it got to a million users in
32:05 five days and it got to a hundred
32:06 million users in six weeks
32:10 and that shooketh the Earth for these
32:14 tech companies and so everyone kind of
32:16 panicked
32:17 and everyone who had AI projects that
32:19 they were kind of slow rolling and we
32:22 don't want to get it out until it's
32:23 right all of a sudden they're like holy
32:25 [ __ ] we gotta we gotta get out there
32:27 including Google and Google invented the
32:30 core technology that chat CPT is built
32:32 on this thing called the Transformer
32:35 um
32:35 so so there's there's just too much
32:38 money at stake you know Microsoft
32:41 invested 10 billion dollars in open AI
32:43 10 billion
32:45 and
32:47 um they're rolling it into all of their
32:49 applications all of Office 365 all of
32:53 azure all of uh it's in GitHub co-pilot
32:58 it's in slack it's in teams it's gonna
33:00 be in everything
33:02 so they're making a big big bet they're
33:04 not going to let it die because they're
33:06 they're you know they're sort of betting
33:08 their future on it
33:10 um Google is now responding pretty
33:12 aggressively meta this is the big reason
33:15 why it's not going to go away
33:17 meta has open sourced some very powerful
33:21 large language models I don't think
33:22 they've open sourced their most powerful
33:24 ones but they open sourced a 70 billion
33:26 parameter model and they open source
33:28 this thing called code code llama which
33:31 is fine-tuned on coding
33:33 um it's a 34 billion parameter model and
33:35 those are out there for commercial use
33:36 freely opened no safety guard rails so
33:41 could nefarious actors use the Llama 2
33:45 models to do bad [ __ ] yes
33:49 go look at worm GPT
33:54 100 bucks a month
33:57 product design for hackers to use GPT to
34:01 go do bad [ __ ]
34:02 so yeah they can do some bad [ __ ] with
34:04 it the good guys can do bad [ __ ] to the
34:07 bad people too so so so that's a bit of
34:10 a Saving Grace
34:11 the models that are out there right now
34:15 are probably not powerful enough
34:19 to do major damage but the ones that are
34:21 coming probably are and so if if you're
34:24 curious in into what
34:27 those companies are doing look up let me
34:30 see let me I want to get you the right
34:32 um
34:34 whatever you call it the name of the
34:36 document the front here
34:40 look up Frontier AI article
34:58 [Music]
35:01 open AI Google Microsoft and anthropic
35:03 form a body to overseer oversee safe
35:07 frontier AI development so the big
35:09 companies
35:11 trying to think I'm trying to see where
35:13 the actual link is to it
35:16 uh
35:21 is it
35:24 um
35:25 I don't know anyway you can go find it
35:27 it's basically the the frontier AI group
35:29 or commission or commitment or whatever
35:32 the [ __ ] it is
35:33 um
35:35 but the ones that have enough
35:37 money and funding and gpus to be able to
35:39 do the next level of these things like
35:42 gpt5 and
35:44 Gemini that's coming out and whatever
35:47 meta is working on that's going to be
35:48 amazing
35:50 um and and Elon Musk all of the
35:52 companies that have have the resources
35:54 to do things that could truly do some
35:57 damage if they if they got out of
35:58 control they've agreed to cooperate
36:01 together
36:03 to do everything they can to not let
36:05 them get out of control Sam Altman has
36:07 committed 20 of open ai's compute moving
36:10 forward for what he's calling super
36:13 alignment if you don't know what
36:15 alignment is alignment is basically
36:18 this is so [ __ ] trippy
36:22 are these things these large language
36:24 models the AI the robots
36:28 are they aligned are there imperatives
36:31 aligned with these things imperatives
36:36 because if they're not
36:39 when these things get to the next level
36:41 and when they can start executing code
36:44 which they already sort of can
36:46 and when they can start taking actions
36:48 like go do a Google search go book me an
36:51 airline go you know
36:54 reach out and communicate with someone
36:56 when they can start doing that which
36:57 they kind of can right now it's a little
36:59 janky
37:01 but when they get really sophisticated
37:03 at that if if the alignment of them and
37:06 the the alignment of us isn't right and
37:09 these things at some point decide well
37:11 these pesky humans their imperatives
37:14 don't align with mine so let's just find
37:18 a way to slow them down a little bit
37:21 it seems like they're they're pretty
37:23 reliant on Logistics to eat at least in
37:26 America
37:27 huh
37:28 well since I have access to some systems
37:31 why don't I just uh confuse the
37:33 logistics
37:36 and then groceries store shelves would
37:40 be bare for a while right so so that's
37:43 the fear
37:45 um you know I don't think the tools are
37:47 are strong enough that we have that to
37:49 fear but these things are improving very
37:53 very quickly and because of that
37:54 competitive pressure
37:56 of chat GPT getting to 100 million users
38:00 thank you sherbert
38:02 um
38:04 they're going to keep they're they're
38:06 going to keep
38:07 innovating it's not going to get
38:09 uninvented it's not going to go away no
38:11 single country can shut it down
38:14 so anyway
38:15 I don't think that even answered your
38:17 your uh question but hopefully it did
38:21 I honestly recommend Claude AI to start
38:24 free is full model just limited usage
38:26 that's actually not a bad I that's
38:28 that's not a bad thought Sam
38:31 so the thing about Claude AI that that's
38:34 that Sam says that's that's where he
38:36 recommends people start it's not a bad
38:38 thought so Claude AI is
38:41 uh the company's called anthropic it's a
38:43 bunch of X open AI people the people
38:45 that helped build the gpt3 model are the
38:48 ones that started anthropic and they
38:50 built Claude what Claude has that chat
38:52 GPT doesn't and this is actually a big
38:54 deal and and you know what I I think you
38:57 know what now that you say that Sam
38:59 I might I might switch those two maybe I
39:02 won't switch them
39:04 but I think you're right I think I think
39:06 you start with Claude
39:08 because here's the deal it's got a
39:10 hundred thousand token context window
39:12 which if if you don't know anything
39:13 about these things you're like Kyle I
39:15 don't know what a token is what's a
39:16 context window why are you talking so
39:18 technical I didn't realize you were some
39:20 sort of Brainiac or something you seemed
39:23 you seem like normal people and then you
39:24 go busting out hundred thousand token
39:28 context what is that
39:30 so
39:32 so what that is
39:34 is how much it can deal with in a single
39:37 chat session
39:39 BT can deal with about 3 000 words so if
39:42 you put in like a two thousand word
39:44 prompt and it responds with a thousand
39:46 words that's kind of its whole context
39:47 window and then anything you do after
39:48 that it sort of forgets some of the [ __ ]
39:51 at the beginning of the conversation or
39:53 if you put in like a 5 000 word
39:56 um document or or you copy and paste
39:58 five thousand words into it it'll just
40:00 go gee I can't do that I'm not
40:02 interested in your
40:05 your words there's too many words there
40:07 I can't I'm a little sleepy chat TBT
40:11 gets a little sleepy sometimes it's just
40:13 like
40:15 Claude
40:16 can do about 75 000 words into and if
40:19 you're like I don't know how many words
40:22 are in a newspaper article what does
40:24 that mean
40:25 the what is it the Great Gatsby yeah The
40:28 Great Gatsby I think that's the one
40:30 fifty thousand word novel so with Claude
40:32 you could take the entire text of The
40:34 Great Gatsby paste it into Claude and
40:37 just start interacting with it like
40:38 it'll it'll have read the whole thing
40:41 and you can have it write a little book
40:42 report for you
40:45 there you go first any little high
40:46 schoolers in here and you got a Great
40:48 Gatsby book report coming up tell them
40:51 that the AI learning lab told you to use
40:52 chat or to use Claude to cheat
41:00 and then when they call you into the
41:02 principal's office tell the principal
41:03 you'll teach them how to use chat GPT if
41:05 you if he lets you off the hook
41:08 because they're gonna need to figure
41:10 that [ __ ] out too
41:12 uh that's that's a really good thought
41:13 Sam I think I'm gonna make a little
41:15 shift there language pasta that's a
41:16 t-shirt that's been added I'm using it
41:19 to learn new ml package in go long oh go
41:22 Lang
41:23 which is a language I really don't know
41:25 well that's super cool yeah you can use
41:27 these things to teach you how to program
41:31 did an apple to Apples between GPT and
41:34 club today not impressed with Claude
41:37 what am I missing
41:39 um well the thing that I just talked
41:40 about uh Source Camp Kelly camp
41:43 is the the big one is the the context
41:46 window
41:48 there's another thing about Claude that
41:50 so so here's the deal
41:52 if you're a writer if you're a lover of
41:54 words gpt4 is the best chat gpt4 at that
41:59 address chat.openai.com
42:02 is really good the Bing version of it
42:05 not so good because it's got too many
42:07 safety guard rules but that is really
42:09 really good so if you want words that
42:12 don't suck use that
42:14 Claude is closer to GPT 3.5 in terms of
42:18 its capability and its scale
42:21 um but because of that large context
42:23 window it might be a little less
42:25 frustrating it might not be a
42:26 sophisticated output but it might be a
42:28 little less frustrating here's the thing
42:30 about anthropic the company behind
42:33 Claude I'm a huge fan of so the way they
42:35 train these things the way open AI
42:37 trains their models is what's called
42:39 human reinforcement learning where as
42:42 they're training these models they have
42:44 the model spit out and answer and they
42:45 have humans go thumbs up thumbs down and
42:49 as they do that the model gets smarter
42:51 and smarter and smarter and that's how
42:52 they do it
42:55 couple of things about that approach
42:57 humans are expensive and slow and have
43:00 biases so so every time a human does
43:03 this or this they're introducing their
43:04 bias into the system
43:07 um and you know they kind of famously
43:09 got dragged drug over the coals of like
43:12 well you all your people are in San
43:14 Francisco so it's you know it's it's you
43:16 know
43:16 it's very left-leaning
43:19 um
43:20 Claude anthropic uses what's called a
43:23 constitutional learning model which is
43:25 they write a constitution a set of
43:26 imperatives a set of rules
43:28 that when Claude answers a thing it's
43:31 compared against those rules and if it
43:33 doesn't pass the rules it goes back and
43:35 iterates on the answer
43:38 um and so that approach I think moving
43:41 forward is going to be an important
43:42 approach and anthropic's whole mission
43:45 in life is not to create the best large
43:47 language model it's it's to dramatically
43:52 improve large language model safety for
43:55 the entire industry they just have to
43:57 make their own large language model to
43:59 be able to
44:00 build out the tech right so there's a
44:03 reason to support them and I think I
44:05 think that you know
44:08 I think all of these things are going to
44:10 get much much better
44:12 in the next I don't know six months or a
44:14 year so
44:16 anyway yes chat GPT is still better
44:19 um there's a lot of people who don't
44:21 understand AI so they fear it exactly
44:23 wait where did that go
44:27 but it may come across as anger
44:30 yeah well
44:32 that are [ __ ] angry about it right
44:36 so
44:37 the The Writer's Guild and sag are angry
44:41 about it
44:42 now
44:43 I personally think
44:47 um
44:49 this is coming I have a degree in acting
44:51 and I've written seven screenplays so I
44:53 know the world this is not this is not
44:55 coming out of nowhere
44:57 um
44:58 I think what they're striking for and
45:00 and what they're trying to get in their
45:02 contracts I absolutely support but the
45:05 position that they've taken of AI is
45:08 evil and if if that's true that the wga
45:11 is suing open AI they're taking a
45:16 you know scorched Earth
45:18 we will never use AI for anything kind
45:21 of approach to this and I just think
45:23 that's not the wisest thing to do
45:27 because this [ __ ] isn't getting
45:28 uninvented it's common whether we want
45:30 it or not
45:32 Microsoft announcement today Windows
45:34 toolbar next week Office 365 co-pilot
45:38 goes live wait what did that say when
45:41 hang on let me go find this
45:45 um let's see
45:47 uh
45:48 Microsoft
45:50 co-pilot release date
45:56 [Music]
45:56 um
46:06 [Music]
46:08 12 hours Windows 11 co-pilot just got
46:11 okay Windows 11 co-pilot just got a
46:14 release date is that what you said yeah
46:15 yeah so that's the so so they're
46:17 building so into Windows 11 you're gonna
46:20 have a toolbar that's basically a chat
46:22 thing where you can just say
46:24 put my computer into dark mode
46:27 um I don't want my computer to go to
46:29 sleep after 10 seconds I want it to wait
46:32 15 minutes whatever and it'll just go
46:35 find the [ __ ] and do it
46:37 that's really cool and then Office 365
46:40 goes live November 1st I told you it's
46:43 gonna be it's gonna be a wild wild wild
46:45 wild wild wild wild fall
46:49 so
46:51 um duet is live in Google right now but
46:54 it's kind of buried
46:56 November 1st
46:59 GPT is going to be live in Windows and
47:02 office 365. Dolly 3 will probably be out
47:06 maybe a new mid-journey Google Gemini
47:09 which is supposed to be better than chat
47:11 gpt4
47:13 um
47:16 those are the biggies I can think of
47:17 right now there might be something else
47:19 I missed but there's some big stuff
47:20 coming
47:22 what's up talk Dylan haha Kyle and
47:25 Irregulars what's happening to you
47:29 um
47:36 I went through the whole exercise with
47:38 Bing today they failed Claude Hit it
47:41 first try
47:42 oh I went through a-hole exercise
47:50 with Bing today
47:53 they failed Claude hit it the first try
47:55 yeah
47:56 yep uh Bing
47:59 so
48:00 so what I did what I did to Bing not not
48:03 that Microsoft cares that the AI
48:05 learning lab has done this but so you
48:07 can barely see the word Bing there
48:09 underneath my fingerprints that's
48:11 because I made it dark gray
48:13 I I dimmed it I turned out Bing's light
48:16 because I'm pissed off at it it just it
48:20 shouldn't suck this bad
48:25 and it doesn't suck because the
48:26 technology is bad it sucks because wait
48:30 never mind Kyle I was on the quad train
48:32 there
48:33 oh that's right I don't mind that I
48:35 don't listen here's here's here's the
48:37 deal about about this channel
48:41 I don't give a [ __ ] if people like try
48:44 this AI stuff and just say I [ __ ]
48:46 hate it it's awful or if they try it and
48:49 they say I love Bing or I love Claude or
48:52 I love that like I don't give a [ __ ] the
48:54 thing I'm most interested in is getting
48:56 people off the sidelines so that they
48:59 have an opinion that is informed
49:03 right
49:04 if you're just sitting on the sideline
49:05 come on I heard chat DBT lies it can't
49:08 even it it it stops it uh September 2021
49:12 it doesn't even know anything in the
49:14 future
49:15 or in the present whatever
49:18 but if you're sitting on the sidelines
49:20 not actually playing with it not
49:21 actually understanding how it works what
49:23 it does what it makes possible what it's
49:24 good at what it's not good at
49:27 then you're just a whiner
49:31 don't be a whiner be engaged
49:34 if you're gonna [ __ ] all over AI [ __ ]
49:36 all over it from a place of of knowledge
49:40 and make a cogent argument for why you
49:42 think the robots are gonna kill us
49:45 why they are going to take all the jobs
49:48 you know I go the other way but like
49:51 that kind of that kind of
49:54 um opinioneering
49:57 is great
49:58 I love that stuff
50:00 it's the ones that have clearly never
50:02 used it that have all sorts of
50:04 pontificating opinions about it it's
50:06 like shut the [ __ ] up have you used it
50:11 and it's clear that they haven't that
50:13 drives me [ __ ] crazy
50:16 have you heard anything about Google
50:17 using AI to sort through users docs and
50:20 steal info oh well Laura um
50:24 I I don't so
50:28 probably I mean but here's the deal
50:31 Google
50:34 has been okay no let's not put this on
50:38 Google
50:40 Google created create created a thing
50:43 called the terms of service
50:44 that over the years just got worse and
50:47 worse and worse and we all us stupid
50:50 human beings that were like oh free
50:51 email
50:53 oh a free search engine
50:55 willingly gave them all of our data
50:58 so they've already got all of our data
51:01 Google Docs has been free for
51:04 I don't know ages so they don't need AI
51:09 to mine that but I guarantee you that
51:13 they've been mining that with AI
51:15 for
51:17 many years
51:18 it's not just in the past year
51:21 and it's not just this generative AI
51:22 [ __ ] right the data is already gone our
51:26 data is already gone
51:29 they have so triangulated when you talk
51:33 to people that really know what the [ __ ]
51:34 is going on with data between Facebook
51:37 and Google and all of these services
51:41 they they triangulate everything they
51:45 know where you are they know when you're
51:47 there they know what you're doing what
51:48 you're thinking what you bought what you
51:50 didn't they know it all and they know it
51:52 all in real time
51:54 and that's before any of this [ __ ] came
51:56 out that what I think what makes this
51:58 stuff scary is we now have access to
52:01 tools where we can look across all of
52:04 our Google Docs and go oh
52:08 I wonder if Google can do that well yeah
52:11 they've been doing it for [ __ ] ever
52:13 so
52:15 so you know that ship has already sailed
52:18 that ship is is is you know that ship
52:21 has sailed and sunk
52:23 the do they have access to our data yeah
52:26 and they've sold it
52:28 it's good it's long gone long gone
52:32 um we being where I work oh cool very
52:34 nice so Claude is best for free
52:38 um can you use AI for advertising you
52:41 can yeah it's it's it's really good at
52:44 you know what Sam Quinn said here
52:47 um gpt4 in particular is really good at
52:50 writing now you have to get good at
52:51 prompting so it doesn't just give you
52:54 um
52:55 pablum [ __ ] if you just say write me
52:58 an ad it'll write you an ad and it'll
53:00 suck but if you say hey here's my
53:03 creative brief and here's who I am and
53:06 you know I want you to act like a master
53:08 copywriter for a world-class Ad Agency
53:11 and I want you to come up with you know
53:14 a whole campaign inspired by Lee Klaus
53:17 work with Apple
53:20 it'll know what all that means and it'll
53:22 go write you some cool [ __ ]
53:24 and and tell it to write you 30 you know
53:27 30 headlines you know tag lines for a
53:31 campaign
53:32 and then out of the 30 taglines it
53:34 writes
53:35 five of them won't suck and then you can
53:37 say hey I like these five give me 50
53:41 more that are more like that than the
53:43 other 25 and it'll
53:46 so yes really good for advertising and
53:50 also for marketing strategies and I'm a
53:53 writer and actually enjoy prompting so
53:55 maybe that's the difference for me so so
53:58 Source Camp Kelly camp
54:01 I like your name it's a lot there's a
54:05 lot going on with the source and the
54:06 Kelly but I like it
54:08 um
54:11 this this is the thing that frustrates
54:13 me about the The Writer's Guild and the
54:15 SAG thing and sag's a bit a bit more
54:18 complicated but the writer's Guild in
54:19 particular
54:21 these
54:23 in my humble opinion in my my very
54:27 humble opinion
54:29 are the most profoundly powerful writing
54:31 tools in the history of humanity
54:34 and you've got a professional
54:36 organization that represents writers
54:38 going [ __ ] that we'll never use it
54:41 really
54:44 when it's built into Microsoft Word and
54:47 it's built into your screenwriting
54:48 software and it's built into your
54:50 note-taking applications and it's built
54:52 into Google Docs and it's built into
54:54 [ __ ] everything windows 11.
54:57 you're not going to use it really
55:01 if you can improve the quality of your
55:04 outline of your screenplay
55:07 by having this thing get you to 80
55:10 percent in 10 minutes rather than three
55:12 days
55:13 and then you spend a day refining that
55:17 outline without chat GPT why would you
55:20 not use that
55:22 and I think the fact of the matter is a
55:24 lot of writers will
55:27 but the Union's taking the stance like
55:29 AI is evil well no what okay
55:32 you can judge it all you want it's not
55:34 going away it is the most profoundly
55:36 powerful writing tool in the history of
55:39 mankind
55:40 and you're like we don't want to use
55:43 computers we like the sound of
55:44 typewriters we we like the letter
55:46 hitting the paper chick
55:48 it's not a screenplay unless a metal
55:51 letter hits a ribbon onto a piece of
55:53 paper
55:56 storing screenplays on magnetic media
55:59 how dare you that's not real writing
56:03 I mean that's that's literally what's
56:05 happening
56:08 it just baffles my mind it doesn't
56:11 because it is [ __ ] scary like I get
56:14 why it's scary I get why they're saying
56:16 hey producers
56:18 don't [ __ ] use these things
56:21 but the producers are going to use them
56:23 anyway
56:25 and if they can say listen
56:28 these tools are inevitable we will like
56:30 to likely incorporate them into our
56:32 workflow
56:34 but
56:37 even if something is AI enhanced you
56:40 know from the production company if I'm
56:42 the original writer on it here's the
56:43 rights that I want
56:46 like they can still fight for for rights
56:49 and sag can do the same thing without
56:51 denying that these things are coming
56:53 they can say hey they're coming and
56:55 given that they're coming here are our
56:57 demands
56:59 that's cool like I don't know if it's
57:01 cool that makes sense to me
57:04 but this you know we're gonna Sue open
57:07 AI to get rid of it it's not going to
57:09 slow it down
57:11 I mean
57:12 llama is open sourced so let's say let's
57:16 say they win they shut down open AI
57:19 they even shut down Microsoft right the
57:23 writer's Guild Microsoft can no longer
57:25 use chat GPT what about all these open
57:27 source projects built on top of llama 2.
57:30 what happens to them does the wga go
57:33 after meta and say put the horse back in
57:36 the barn and Matt is like uh it's not
57:38 one horse now it's like 10 million
57:40 horses and nothing we can do about it
57:45 Fair Point I've been hearing Rumblings
57:47 from writer talk about Google taking
57:49 story ideas oh that's interesting
57:52 Google taking story ideas from writers
57:55 but but is Google
58:00 like they don't have a Netflix arm do
58:02 they I guess YouTube might have a
58:03 production arm
58:05 but I don't think YouTube's doing like
58:07 like full-on full-on shows I I I find
58:11 that that sounds to me like just
58:13 paranoid [ __ ] like what would be in
58:15 it for Google to steal story ideas
58:19 because story ideas are [ __ ] dime a
58:21 dozen and now with chat GPT I mean
58:23 honest to God I've I've done some
58:25 experiments I did an experiment where I
58:28 said
58:29 I took gpt4
58:31 and I told it
58:34 to give me the original Star Wars plot
58:38 it did
58:41 and then I said I said okay now write me
58:43 an outline that matches you know scene
58:46 for scene matches Star Wars
58:48 except instead of being in space I want
58:51 it to be
58:54 um animals on the Galapagos Island at
58:57 the time of Charles Darwin
58:59 and it came up with character names that
59:02 were [ __ ] genius it came up with a
59:05 plot that was [ __ ] amazing
59:08 and so so
59:11 I I just I I just
59:14 like it just feels like conspiracy
59:16 [ __ ] to me I don't what's in it for
59:17 Google to steal [ __ ] story ideas they
59:21 just it just doesn't make sense to me
59:23 I mean Google's got much bigger fish to
59:25 fry like like predicting consumer
59:28 behavior of every person on the planet
59:32 there's big money in that you know if
59:35 they get story credit for some
59:38 Film Production that might make them 150
59:41 million dollars or a billion if they
59:42 really knock it out of the park
59:44 and meanwhile they're making trillions
59:46 with their data it just doesn't make
59:49 sense to me
59:52 again I understand fear
59:55 like like what that probably was was
59:57 like that Air Force thing where some
59:58 some writers said it hypothetically
1:00:00 because Google's got access to Google
1:00:02 Docs they could steal our stories and
1:00:04 then someone else heard that and said
1:00:05 Google's stealing stories that's
1:00:08 probably what happened there good use of
1:00:10 chat GPT input customer personas and use
1:00:13 it as a focus group ask questions it's
1:00:15 really good for that
1:00:16 how would they respond it's really good
1:00:18 for that it's really good for that
1:00:20 I did some I did some Persona stuff
1:00:22 where I I actually brought one of the
1:00:24 personas to life
1:00:27 um using did and the voice synthesis
1:00:29 stuff it's pretty good I wonder if I
1:00:31 have that somewhere I do you wanna see
1:00:33 it
1:00:39 yeah this is it
1:00:47 view play in a window
1:00:52 play
1:00:55 Ed oops
1:00:58 [Music]
1:01:08 February
1:01:11 and this was something I wanted to do I
1:01:13 wanted to experiment
1:01:15 for my company storyline so we have a
1:01:17 video storytelling platform
1:01:20 and we do a lot of work in healthcare
1:01:21 and Pharma and we wanted to see
1:01:24 if
1:01:25 like a lot of times our Pharma clients
1:01:27 can't find patients like they don't have
1:01:29 the right indication or whatever they
1:01:31 all the stuff they all their fancy s
1:01:34 words
1:01:37 and so
1:01:39 I wanted to see
1:01:42 you know if we needed to generate a
1:01:44 patient if we needed to generate a
1:01:46 patient's story or generate a patient
1:01:48 how might we do that well you know they
1:01:51 always want them to be you know diverse
1:01:53 and represent different groups and
1:01:54 things like that so
1:01:57 what I'm about to show you
1:01:59 from the from the minute I went into
1:02:02 chat GPT and said huh wonder if I could
1:02:05 do that
1:02:06 to the video I'm about to show you took
1:02:08 less than 30 minutes and this was the
1:02:10 first time I tried this this was last
1:02:12 February
1:02:14 so shit's gotten better since since this
1:02:17 so the first thing I did was I had
1:02:19 chat GPT come up with
1:02:21 um a Persona framework and I said you
1:02:23 know how do you create Persona
1:02:25 Frameworks and it educated me on that
1:02:27 and it gave me a bunch of different
1:02:28 options and so I selected some of the
1:02:30 things and I said okay these look good
1:02:32 and then I said now write me 12 personas
1:02:34 that that you know have different
1:02:36 backgrounds in all these different areas
1:02:38 bang out they came so I took one of them
1:02:41 this is Amina and
1:02:43 the way
1:02:45 storyvine software works is we have an
1:02:48 app that guides an end user it prompts
1:02:50 it prompts a patient in this case
1:02:53 to answer a series of questions about
1:02:56 their disease when were you diagnosed
1:02:59 and what do you enjoy doing and you know
1:03:01 how have you been treating it whatever
1:03:03 whatever the prompting questions are in
1:03:05 a person answers them so I then had chat
1:03:07 GPT have Amina tell her story using our
1:03:10 prompts so so basically she answered our
1:03:13 prompts just like she was a real person
1:03:16 and then
1:03:18 um Leah fasten who co-hosts the salon
1:03:20 with me she's a photographer and she had
1:03:22 this really cool style that was super
1:03:24 authentic so so she gave me a whole
1:03:26 bunch of images to start with so I
1:03:28 picked this one for Amina and then I
1:03:31 took this story and I took a
1:03:34 um South Asian accent in
1:03:38 um it was at 11 labs
1:03:43 yeah uh
1:03:45 yes
1:03:47 no it wasn't it was in did so in did you
1:03:51 can pick different languages
1:03:53 and so I just copied in these and then I
1:03:55 made video clips and then I put those
1:03:57 video clips in my company software which
1:04:01 automatically assembles them
1:04:03 into a fully edited video hang on
1:04:07 and this is the video that was created
1:04:10 so again this was from start to finish
1:04:13 less than 30 minutes
1:04:19 [Music]
1:04:23 I was diagnosed with illness about four
1:04:26 years ago
1:04:27 what brings me joy is spending time with
1:04:30 my family especially my parents and
1:04:33 siblings
1:04:34 we enjoy cooking and eating traditional
1:04:37 South Asian dishes together
1:04:39 moving beyond my diagnosis was not easy
1:04:42 but I found comfort in my faith and my
1:04:45 family's support
1:04:46 I also started to focus on what I could
1:04:49 still do instead of so anyway you get
1:04:52 the basic idea so yeah so not only can
1:04:54 it do personas you can bring those
1:04:56 personas to life in ways that that you
1:04:58 just couldn't otherwise and
1:05:01 you know again this still sounds a bit
1:05:04 monotone I think I think with the accent
1:05:06 it it sounds you know decently realistic
1:05:09 but
1:05:09 it's still fairly flat and monotone
1:05:13 but within a year it's not going to
1:05:15 sound flat and monotone it's gonna have
1:05:17 it's gonna have you know Cadence and
1:05:20 and Dynamics and things like that and
1:05:22 it's going to seem more increasingly
1:05:24 real and so you know the like the amount
1:05:27 of use cases here like like just just
1:05:30 dealing with
1:05:32 um
1:05:32 things like
1:05:34 um
1:05:35 you know there's patients who want to
1:05:38 tell their story that can't tell their
1:05:39 story well now they're going to be able
1:05:41 to
1:05:43 um you know so all sorts of stuff
1:05:45 ideogram site any tips to using it
1:05:48 um a little bit a human being so
1:05:52 um be patient with it
1:05:56 um put your text in quotes use the
1:05:59 buttons at the bottom like the
1:06:00 typography ones a good one like if you
1:06:03 do typography and poster it tends to do
1:06:06 pretty good typography
1:06:08 AI soapbox preach it Ted look at life
1:06:11 that's cool look look at what Ted
1:06:13 putting put in the in the uh in the
1:06:15 Discord that's pretty funny
1:06:19 that's awesome
1:06:22 um
1:06:23 so so that's the first thing is
1:06:26 um just
1:06:27 uh
1:06:29 experiment with a different thing the
1:06:31 the different buttons will just give you
1:06:33 different results the other thing is go
1:06:36 find a video that um
1:06:39 or I mean go find an image an ideogram
1:06:42 that looks close to what you want and
1:06:45 just copy the post because in copying
1:06:47 the post it'll have what buttons they
1:06:49 pushed
1:06:51 um so that's a way then you can remix it
1:06:54 another way is to remix it where you you
1:06:56 take one of their an image that already
1:06:57 exists and just hit the remix button and
1:07:00 just keep spinning it
1:07:02 so so what I tend to do on it is just
1:07:06 experiment experiment experiment until I
1:07:08 get something that's close like AI
1:07:09 learning lab with this dog I would get
1:07:12 like AI leaning lab or or learning lab
1:07:16 or AI lab it like it was always all over
1:07:19 the place but if you get it close like
1:07:21 the one that said AI leaning lab
1:07:24 I took that one and I remixed it and I
1:07:27 just kept remixing it until it spelled
1:07:30 it right
1:07:32 because because once once it it sort of
1:07:35 you know understands the sort of
1:07:37 generals you know amount of words and
1:07:39 letters in in the neighborhood it it
1:07:41 tends to get it right at some point in
1:07:45 two or three iterations
1:07:47 what was the video platform you just
1:07:50 showed Kyle so the the thing that
1:07:52 assembled that with all the graphics is
1:07:54 my company's platform called storyvine
1:07:57 storyvine.com but the thing that
1:07:59 generated Amina talking is D Dash ID
1:08:04 so Dash id.com I think let me just make
1:08:07 sure that's right d
1:08:10 Ash id.com yep
1:08:15 the digital people platform
1:08:20 here I'll show it to you
1:08:33 [Music]
1:08:38 so like here's all the videos that I've
1:08:40 made over the over the you know months
1:08:44 and then let's see if if I want to
1:08:47 create a new video
1:08:54 um
1:08:55 I've got here are
1:08:58 the presenters some of these I made in
1:09:01 mid Journey some of these I made in
1:09:03 stable diffusion
1:09:06 um here's ones that they give you
1:09:08 they've got high quality ones that I
1:09:10 guess those are 1080p ones or instead of
1:09:12 720p or whatever
1:09:15 um we've got close-ups these are all
1:09:16 ones so you basically just pick a pick a
1:09:19 thing and then you can
1:09:23 um
1:09:25 foreign
1:09:26 you can either type your text or you can
1:09:29 upload
1:09:31 where do I do it oh yeah so right here
1:09:33 you can either type text in right so you
1:09:36 type in a script here
1:09:37 like hi
1:09:40 um
1:09:42 hi my name
1:09:44 is Bob I like
1:09:48 Turtles
1:09:52 and then you can choose different
1:09:55 accents
1:10:00 um
1:10:02 we'll just leave the United States and
1:10:05 then certain ones like the United States
1:10:06 they have lots of different characters
1:10:09 so we'll do that let's go let's get a
1:10:12 let's get a Bob here
1:10:15 where's Bob
1:10:19 there's Bob that's Bob there's Bob
1:10:23 and then you say generate video
1:10:28 it's gonna cost me one credit let's go
1:10:33 it should go really quick
1:10:41 it's funny every time I demo this
1:10:43 they've slightly changed the uh the
1:10:46 interface that because that's the thing
1:10:47 you know all these tools being janky
1:10:50 pieces of [ __ ] they they get less and
1:10:52 less janky every month so you know you
1:10:54 go back in to use these things and
1:10:55 you're like oh they actually made this
1:10:57 usable now
1:10:58 hi my name is Bob
1:11:00 I like touchels
1:11:02 [Laughter]
1:11:10 okay
1:11:12 [Laughter]
1:11:18 so that's D Dash ID so that's that's the
1:11:21 tech I use to make that um
1:11:24 that South Asian woman telling her
1:11:27 patient story
1:11:28 that woman Amina
1:11:38 hello I'm Amina Ahmed
1:11:41 I was diagnosed with Alice about four
1:11:43 years ago
1:11:46 cuckoo
1:11:47 cuckoo
1:11:50 and then and then you know then you've
1:11:51 got things like
1:11:53 um
1:11:53 okay I'll show you we've got 100 people
1:11:55 in here so there's people that haven't
1:11:56 seen this yet
1:11:58 so so that's her so that's just sort of
1:12:00 inventing a Persona like literally out
1:12:03 of nothing in in like 25 to 30 minutes
1:12:07 and having her tell her story and then
1:12:11 um and then you have a tool like hey Jen
1:12:13 which just sort of sneak sneakily
1:12:16 launches a feature they don't even
1:12:19 realize is a big deal
1:12:21 and it does this
1:12:25 this is [ __ ] this is so stupid I mean
1:12:29 so stupid like crazy like holy [ __ ] we
1:12:32 live in this reality right now this
1:12:35 exists right now so this is uh
1:12:39 look more Kyle this is a video I filmed
1:12:42 last uh February this was just a clip in
1:12:46 a in a video I made you know for
1:12:48 marketing or this was like a video I
1:12:51 sent out to my advisors or something
1:12:52 like that so I just I just took some
1:12:54 clips from this I uploaded it to this
1:12:56 hey Jen video translate thing and watch
1:12:59 this
1:13:02 so we've heard from a number of clients
1:13:04 that modular content is important to
1:13:06 them getting more efficient with how we
1:13:07 create content and how we leverage
1:13:09 content is
1:13:17 behind me you see these little green
1:13:19 boxes um that's how we think about
1:13:21 stories at storyline we break stories
1:13:23 down into little atoms right little
1:13:25 modules
1:13:37 right so like
1:13:39 that exists today you just upload files
1:13:42 10 or 15 minutes later you get back
1:13:44 something and I think it's up to eight
1:13:46 languages right now
1:13:48 the co-founder I watched it live with
1:13:50 the co-founder last week he said they're
1:13:53 going to be up to 20 languages pretty
1:13:54 quickly and then a lot more after that
1:13:56 and and he said
1:13:58 there's nothing
1:14:00 this is astounding to me he said there's
1:14:03 not much present preventing them from
1:14:06 being able to do that in real time other
1:14:09 than enough compute power
1:14:11 so as algorithms get better and gpus get
1:14:16 better and
1:14:18 you know
1:14:19 more server Farms are are you know
1:14:23 serving up more GPU Cycles
1:14:27 um
1:14:28 we're gonna have a literal Babel Fish
1:14:33 um kind of situation where you're going
1:14:35 to be on a zoom call with people in
1:14:36 multiple countries where everyone's
1:14:38 speaking their native language and
1:14:40 everyone's hearing all the other
1:14:42 participants speaking their language
1:14:44 fluently
1:14:46 [ __ ]
1:14:47 Wild
1:14:49 crazy
1:14:51 all right
1:14:55 time sensitive bedtime reminder your
1:14:59 bedtime is approaching time to wind down
1:15:01 and prepare for a good night's sleep
1:15:03 thanks Apple
1:15:05 I know you've got my best in mind
1:15:09 today you know what I need to do no Kyle
1:15:12 what I need to charge my little watch
1:15:15 hang on I'll be with you you people just
1:15:18 calm down just calm down I'll be back
1:15:22 just hold your horses hold your horses
1:15:27 I know I was late tonight everyone was
1:15:29 all like where's Kyle where's Kyle
1:15:32 class didn't start yet we've been here
1:15:35 for an hour where are you where are you
1:15:38 Sam corn Kyle oh have you seen
1:15:42 Google's new generative Image Dynamics
1:15:44 yet
1:15:46 um I don't know is this a new thing I
1:15:49 saw their style drop thing are you
1:15:51 talking about style drop or is this a
1:15:52 new thing because
1:15:54 if it's new no
1:15:56 oh and by the way for those of you that
1:15:59 are here and you're like um I'm so far
1:16:02 behind if you pay attention to this [ __ ]
1:16:05 every single day and like you do
1:16:07 everything in your human power to keep
1:16:12 up with it you're not even [ __ ] close
1:16:16 not even [ __ ] close so let's see if
1:16:19 Google did something new with Image
1:16:20 Dynamics Google
1:16:22 image
1:16:24 I'm telling you Google's gonna come out
1:16:26 [ __ ] swinging
1:16:28 Dynamics I didn't have
1:16:31 faith in them after their
1:16:35 Paris debacle back in February that was
1:16:38 [ __ ] embarrassing
1:16:41 and then they did another one in June
1:16:43 that was [ __ ] embarrassing
1:16:45 and then just all of a sudden in July
1:16:47 Bard started sucking less
1:16:49 in this latest set of updates and Bard
1:16:51 is actually pretty good
1:16:58 uh
1:17:04 um
1:17:05 generative
1:17:08 Image Dynamics from oh September 18th
1:17:11 okay so this is three days ago
1:17:14 generative Image Dynamics from Google
1:17:17 brain research is this it
1:17:22 sorry I might I might have some dust in
1:17:25 my throat or something maybe my throat's
1:17:26 going again
1:17:30 um
1:17:34 let's watch it together
1:17:37 Google just dropped a new paper with the
1:17:39 name generative Image Dynamics which is
1:17:41 the proof we are living in a science
1:17:43 fiction World they created a diffusion
1:17:45 based model which takes it still image
1:17:48 as input and outputs Dynamic looping
1:17:50 video okay I can't understand that
1:17:53 um
1:17:55 from static to stunning Google Images
1:18:00 generative Image Dynamics so this is a
1:18:03 paper okay this is a paper
1:18:06 yeah September 14th was this is this the
1:18:08 one from Google
1:18:30 is this the one yeah that's the one
1:18:32 what's it do
1:18:35 um oh you want to see something fun yes
1:18:38 Kyle that's why we come here show us
1:18:40 something fun don't ask us questions
1:18:42 like that silly you know we want to see
1:18:44 something fun
1:18:45 my goodness he's such a silly bird
1:18:49 all right
1:18:51 so we're gonna go to chat gpt4 I don't
1:18:54 know if you ever heard of chat jpt but
1:18:57 uh
1:19:00 [Laughter]
1:19:02 so chachifi teased this thing it's going
1:19:04 to change everything I don't know if you
1:19:06 heard about it gonna totally [ __ ] up uh
1:19:09 I didn't talk about the future of work I
1:19:10 don't know if uh if and fundraising is
1:19:13 in here but if you're in here raise your
1:19:15 hand and I'll I'll wax
1:19:17 philosophic about the future of work but
1:19:20 anyway
1:19:22 we're gonna go to plugins
1:19:25 and then I'm using a plug-in called Web
1:19:27 pilot and web pilot can look at
1:19:30 websites and PDFs
1:19:33 so I'm going to say based on this PDF
1:19:37 and this is by the way so one of the
1:19:40 nicknames for the channel as you might
1:19:42 imagine is chat add which is I think
1:19:45 it's my favorite nickname of all the
1:19:48 stuff
1:19:48 uh
1:19:50 I I have 80d and
1:19:53 while I like technical [ __ ] the thought
1:19:56 of reading a long technical paper and
1:19:59 actually parsing all of it
1:20:01 makes me insane and and with these
1:20:03 technical papers even reading the
1:20:05 abstract is like you know they kind of
1:20:08 assume you know what you know certain
1:20:11 abstract mathematical concepts are so
1:20:13 when you read the abstract you can't
1:20:15 even understand what the [ __ ] they're
1:20:16 talking about
1:20:17 so
1:20:19 that's that's no longer an issue so
1:20:23 based on this PDF
1:20:25 um please
1:20:28 summarize this
1:20:31 um at a fifth grade reading level
1:20:39 I'm going to talk it
1:20:43 knowing
1:20:44 that
1:20:46 I am an artist
1:20:49 and trying to understand what is
1:20:51 significant
1:20:53 about this paper in the context
1:20:57 of digital art
1:21:03 and so now it's going off it says using
1:21:06 web pilot so it's using that that
1:21:08 plug-in it's going off and reading the
1:21:10 paper for us
1:21:11 and then it's going to come back and
1:21:13 it's going to do a little writing it's
1:21:15 going to do a little creative writing
1:21:19 all right this paper is about making
1:21:22 pictures move summary the scientists
1:21:25 again I did this for a fifth grade
1:21:26 reading level the scientists in this
1:21:28 paper have created a way to make
1:21:30 pictures look like they are moving like
1:21:32 trees swaying or flowers Dancing In The
1:21:34 Wind all from just one single image they
1:21:36 use something called neural stochastic
1:21:39 motion textures
1:21:42 see you don't you don't know what those
1:21:45 words [ __ ] mean like they're it's
1:21:48 four words neural stochastic motion
1:21:52 textures I don't even know what some of
1:21:54 the individual words mean much less what
1:21:56 that grouping of forwards means
1:22:00 to predict how each part of the picture
1:22:02 should move to make it look real this
1:22:05 means they can turn a still image in
1:22:07 into a video that Loops over and over
1:22:09 and over or even let people interact
1:22:12 with the picture like dragging and
1:22:14 releasing a point on an object
1:22:16 of the picture and see it bounce back so
1:22:19 you could grab a flower and drag it and
1:22:20 it'll snap back that's amazing
1:22:23 this is really cool for artists because
1:22:25 it can bring your artwork to life why
1:22:28 it's cool for digital art bringing art
1:22:29 to life interactive art creating Dynamic
1:22:32 scenes questions to think about how can
1:22:34 I use this
1:22:36 you can use this technology so anyway
1:22:38 like
1:22:39 even though that's kind of a silly
1:22:41 little way I did that you know what I'm
1:22:43 gonna do I'm gonna say
1:22:45 hang on neural stochastic motion
1:22:47 textures copy
1:22:49 I'm going to say
1:22:52 watch this this will be fun
1:22:55 I want you to pretend
1:22:57 like you're a marketing master
1:23:01 and your goal is to help these
1:23:04 scientists
1:23:05 repackage the concept of
1:23:12 and I'm gonna and then I paste it in
1:23:14 neural stochastic motion textures
1:23:19 um
1:23:23 come up with a way
1:23:25 to say that that will make sense to
1:23:28 non-scientists
1:23:31 and let's use that as the tagline
1:23:35 and then I want you to come up with 10
1:23:39 creative names for what this technology
1:23:42 might be called if we were going to
1:23:44 Market it
1:23:48 absolutely tagline bring your art to
1:23:51 life with magical motion textures
1:23:53 perfect
1:23:56 creative creative names art mover Magic
1:23:59 picture pixel pulse Creations a live
1:24:02 brush technology motion craft Studios
1:24:04 these are all shitty Dynamic canvas
1:24:06 Innovations interactive wait
1:24:09 interactive imagination flow interesting
1:24:13 wondrous wave Artistry no
1:24:15 alive and art Solutions okay
1:24:18 um let's see
1:24:21 the names are too slick
1:24:23 can you keep them simple and compelling
1:24:28 but have them sound less like you're
1:24:31 trying to sell me something
1:24:33 I don't even know if it'll understand
1:24:35 that but
1:24:37 Art In Motion living pictures that's
1:24:39 pretty good
1:24:41 living pictures simple motion art
1:24:43 breathe art picture pulse quiet motion
1:24:46 gentle move art but anyway you get the
1:24:48 idea like
1:24:51 one of the one of the things that
1:24:53 researchers often face is they'll create
1:24:56 something remarkable and they they have
1:24:58 a tough time communicating like the
1:25:00 value of it because they're they're
1:25:01 interested in like how to describe what
1:25:04 the thing actually is but if they want
1:25:06 to get the thing out there you could use
1:25:08 something like this to to you know
1:25:11 create a marketing version of of your
1:25:14 paper where one goes to Archive for the
1:25:17 scientists and another one goes to to
1:25:19 Twitter to understand what the [ __ ] this
1:25:22 thing is
1:25:24 that's really cool thanks for sharing
1:25:26 that I think that was Sam or no is that
1:25:28 joker Sam or Joker shared that thank you
1:25:31 for sharing that
1:25:33 meth lab AI learning meth lab yeah
1:25:39 all right
1:25:41 um yesterday I saw a video that said
1:25:43 dolly is now embedded in chat GPT can
1:25:46 you show it's not it's
1:25:49 uh I can I can show you the site it's
1:25:53 not in chatbt yet
1:25:58 I'll go look now the thing that was
1:26:00 announced yesterday is it was announced
1:26:02 yesterday and it sounds like from a
1:26:05 Braun
1:26:07 um this guy I know that it's one of the
1:26:08 open AI ambassadors it's going to show
1:26:10 up as a plug-in
1:26:13 um
1:26:19 I'm gonna go just make sure that they
1:26:20 didn't sneak it in while we've been
1:26:22 sitting here this Happened One Night
1:26:25 with uh
1:26:27 code interpreter I was like when am I
1:26:29 gonna get access the code interpreter
1:26:30 they never gave me eyes as the code
1:26:32 interpreter I talk about oh here it is
1:26:34 that's pretty funny
1:26:37 um and then we'll go look and see if
1:26:38 they snuck it into plugins because
1:26:40 that's where he said it's going to be so
1:26:42 in plugins
1:26:43 we're going to go to the plug-in stowa
1:26:47 downtown
1:26:53 dolly
1:26:55 no dog Dash e
1:26:59 nope not there
1:27:02 um but I'll go show it to you
1:27:07 well that lost its CSS connection didn't
1:27:10 it there you go all right so here's
1:27:13 Dolly three this is the website so if
1:27:15 you want to see this go to openai.com
1:27:21 d-a-l-l-e-3 because why make it easy to
1:27:25 type in
1:27:25 [Laughter]
1:27:31 here's here's some of the images the
1:27:33 this this one right here
1:27:36 um
1:27:38 there's a couple of things going on here
1:27:40 so the prompt for that image where is it
1:27:43 is
1:27:45 let me open this up all right
1:27:50 The Prompt for this image is an
1:27:52 illustration of an avocado sitting in a
1:27:54 therapist's chair check
1:27:57 saying I just feel so empty inside
1:28:01 check
1:28:03 and if you know these tools they can't
1:28:06 do that well
1:28:08 with a pit sized hole in its Center so
1:28:12 that's not a pit it's a hole
1:28:14 a pit sized hole
1:28:16 the therapist a spoon wait
1:28:19 a spoon check is taking notes check
1:28:25 so
1:28:27 it can take all of the context it's also
1:28:29 built in natively in the gpt4
1:28:33 um
1:28:36 gpt4 is going to help be able to help
1:28:38 you prompt and it's also going to be in
1:28:41 context of a larger conversation so
1:28:44 let's say you describe the plot of a
1:28:47 graphic novel and then you want to start
1:28:49 creating concept art for it it
1:28:51 understands the context of the graphic
1:28:53 novel that you're that you're writing
1:28:55 about
1:28:57 um so that's that guy
1:29:00 um
1:29:01 here's another one where the the oops
1:29:03 what did I do oh
1:29:06 old man trying to use computers old man
1:29:09 trying to use computers yeah
1:29:13 um so this is cool this is a lychee and
1:29:17 inspired spherical chair with a bumpy
1:29:20 white exterior and plush interior set
1:29:23 against tropical wallpaper
1:29:26 yep that pretty much hits it out of the
1:29:29 park
1:29:30 so it does that
1:29:33 um it also does this is a thing that
1:29:36 again just it's like a really
1:29:38 complicated prompt so they broke the
1:29:40 prompt up and just showed all the
1:29:42 different things that uh it captured
1:29:46 um here's dolly two with this same
1:29:49 prompt and here's Dolly three with the
1:29:51 same prompt so Dolly three is just way
1:29:54 better and then I'll play this video
1:29:57 here because this video is crazy
1:30:06 [Music]
1:30:19 oh
1:30:21 my daughter says
1:30:23 its name is Larry can I see more like
1:30:27 this
1:30:28 so it now it it now remembers that that
1:30:32 character
1:30:34 is Larry
1:30:36 and so
1:30:38 now he just goes on and prompts it to
1:30:40 put Larry in a bunch of other scenes
1:30:42 which if you've ever tried to do a
1:30:44 children's book this is one of the most
1:30:45 frustrating things because you get a
1:30:48 cute thing like Larry and then you try
1:30:50 to get Larry to be in other scenes and
1:30:52 you just [ __ ] can't it's like it all
1:30:54 of a sudden the Hedgehog is a platypus
1:30:55 and then it's a squirrel and then you're
1:30:57 like where the [ __ ] is Larry
1:30:59 Larry's gone not anymore
1:31:08 [Music]
1:31:33 look at that
1:31:35 you tell the whole story then you then
1:31:37 you put Larry to bed
1:31:39 and he goes to bed this it's like
1:31:41 Emilio's wife it's like Emilio's wife
1:31:44 Larry was made for you look at that bed
1:31:47 look what we have for Emilio's wife
1:31:50 she's very concerned about being late
1:31:52 and
1:31:57 she's just got lovely Emoji game
1:32:00 and so so
1:32:03 fans made her an am I late sign and it's
1:32:08 just like little Larry look look how
1:32:12 sweet
1:32:14 you see how long this steps in her
1:32:17 related
1:32:19 we're all part of the same Oneness you
1:32:22 know that right
1:32:24 so we shouldn't be mean to each other
1:32:26 anymore how about that
1:32:28 what if we use AI as an excuse to stop
1:32:31 being [ __ ] to each other
1:32:33 how's that for a radical idea
1:32:38 you in
1:32:42 Microsoft debut debuts AI powered
1:32:45 co-pilot platform new Surface Hardware
1:32:48 interesting
1:32:50 Microsoft to sell AI assistant software
1:32:53 the biggest clients starting November
1:32:55 1st yep
1:32:56 Kyle there's a demo of the moving images
1:32:58 online oh cool where is it let me go see
1:33:01 if I can find it
1:33:02 um
1:33:03 what was that called
1:33:07 um
1:33:09 generative Image Dynamics
1:33:14 generative image
1:33:19 Dynam
1:33:23 X demo
1:33:27 generative image dynamics.github.io
1:33:32 paper demo
1:33:34 I met Sam Sam's the best
1:33:38 Sam has has has brought me to the other
1:33:42 side of if someone says where should I
1:33:44 start I think they should start with
1:33:45 Claude I think you're right Sam and now
1:33:48 Sam's like but Kyle showed demo
1:33:56 don't just read the paper with your
1:33:58 stupid chat JP hey
1:34:01 all right seamless looping video it's
1:34:04 just wiggling interactive Dynamics it's
1:34:06 wobbling the picture that's super cool
1:34:08 oh yeah look at these things
1:34:12 oh this is so cool you know what just
1:34:17 I mean think about this for just
1:34:20 um just for marketing for for marketing
1:34:23 if nothing else
1:34:24 you generate a cool image or you take a
1:34:27 cool photograph you pop it into a tool
1:34:30 like this and it just brings it to life
1:34:32 like this it just so imagine
1:34:36 the web and and social media
1:34:40 is just gonna start having like more
1:34:42 emotion in it because it will be just as
1:34:45 easy to make an image that does this
1:34:48 is to just make an image
1:34:51 that's pretty slick
1:34:55 can we grab any try it yourself yes oh
1:34:58 look
1:35:00 by the way I'm at generative Dash
1:35:03 Dynamics
1:35:05 slash GitHub dot IO
1:35:10 wow
1:35:13 wow wow can I grab the background
1:35:16 nope
1:35:18 so it understands that this is in the
1:35:20 foreground so it's the thing in Focus so
1:35:22 it separates it from the background wow
1:35:24 that's wild
1:35:26 let's see let's grab another one a
1:35:28 little white one
1:35:30 I mean it's a little boring but it is
1:35:33 it's it's a lot of cool it's like what's
1:35:35 nice about it is how
1:35:38 fluid it is and how it's it's instant
1:35:45 oh that's cool oh wow it's even doing
1:35:48 some parallax
1:35:50 you can't really see it it's doing some
1:35:52 kind of Parallax stuff
1:35:55 on like that little Branch right there
1:35:57 is moving slower than the one that's
1:36:00 closer
1:36:02 so like it's it's understanding some
1:36:04 depth stuff in here
1:36:06 there's some artifacts that are weird uh
1:36:09 see see these blurry lines here there's
1:36:12 some weird artifacts on that one
1:36:15 yeah
1:36:16 yeah it's it's sort of
1:36:19 [ __ ] in the bed there but this right
1:36:22 here
1:36:23 is perfect
1:36:25 huh
1:36:28 but it understands that this is closer
1:36:31 so it's trying to move it more and so
1:36:33 it's just not interpolating that
1:36:36 movement right but the the one where
1:36:38 it's not moving it a lot it's really
1:36:39 cool
1:36:41 motion magnificent
1:36:45 oh that's so cool we can mimify top or
1:36:49 magnify bottom
1:36:51 Minify oh zoom out
1:36:58 by adjusting the amplitude of the most
1:37:00 motion textures so this is motion
1:37:03 textures
1:37:04 Wild
1:37:06 motion
1:37:08 related work that's super cool all right
1:37:11 see please repeat the name again yes the
1:37:15 the paper is called generative Image
1:37:17 Dynamics from Google research
1:37:20 and the URL is generative Dash Dynamics
1:37:26 dot GitHub dot IO and in fact
1:37:34 um
1:37:35 [Music]
1:37:36 uh-oh what happened
1:37:39 no stop
1:37:42 oh Bing
1:37:46 whoa
1:37:49 look what Bing did
1:37:51 this is like Siri taking over I was just
1:37:55 trying to show you the URL of that thing
1:37:58 um and Bing Bing heard me and said
1:38:00 generative Image Dynamics demo and made
1:38:06 an image to kind of go with it that's
1:38:08 kind of cool
1:38:10 um anyway there's the there's the URL
1:38:12 gender generative Dash Dynamics dot
1:38:16 GitHub dot IO and then there's a link to
1:38:20 the uh
1:38:22 to the paper to the archive and to the
1:38:25 demo
1:38:28 so that's pretty Swale all right what
1:38:31 time is it I got Listen People I got I
1:38:34 have a 7 A.M meeting it's 10 53 it's
1:38:37 very late people are ruining my life
1:38:39 you're ruining my life
1:38:42 um
1:38:43 oh
1:38:45 you're not
1:38:46 You're Swell
1:38:49 um a couple of things let's see I'm
1:38:50 totally up for that idea I'm gonna I'm
1:38:52 gonna scroll to the bottom here
1:38:54 just so the Takeover begins it starts
1:38:57 with our web browsers
1:38:59 uh off to work I go later Thelonious
1:39:01 peace now Bing is jealous exactly what
1:39:04 the hell is going on here the dot IO
1:39:06 extension is assigned to the British
1:39:09 Indian Ocean territory that's really
1:39:11 funny
1:39:14 um let me show you I want to hit a
1:39:17 couple of things before I go
1:39:19 um first thanks for hanging out this is
1:39:20 cool um I gotta go to bed though uh uh
1:39:23 uh uh
1:39:25 where is my there it is okay
1:39:29 um
1:39:30 if you want to join a community of
1:39:34 people exploring Ai and trying to figure
1:39:36 out what the [ __ ] all this stuff is
1:39:40 um this is this is a group I started
1:39:41 back in December called the AI salon
1:39:45 we meet every other Tuesday
1:39:47 um in person here in Denver and then
1:39:49 online and then we have an online
1:39:50 Discord and a lot of the people from the
1:39:53 AI learning lab have infiltrated the
1:39:56 salon and are kind of like you know
1:39:58 playing around in there it's really cool
1:40:00 there's a lot of new energy a lot of new
1:40:02 activities so if you want to you know be
1:40:05 part of a community that's trying to
1:40:06 figure this stuff out Larry's gone
1:40:12 keep the digital Gods just made it funny
1:40:15 uh people are making funnies what's fun
1:40:18 is people are using ideogram with the
1:40:20 with the text generation as I say stupid
1:40:23 [ __ ] they make stupid images and put
1:40:24 them in there to remind me of the stupid
1:40:26 [ __ ] I said which I think is it's kind
1:40:29 of the most flattering kind of trolling
1:40:31 I've ever experienced I like it a lot it
1:40:34 makes me it makes me smile inside
1:40:37 um so anyway go do that
1:40:39 um uh Story vines my company uh but the
1:40:42 the other thing is I'm putting on a
1:40:44 workshop
1:40:45 um for creatives so I've been invited by
1:40:48 another Creator who's a creative coach
1:40:50 and a voiceover artist and we've put
1:40:52 together a three-day Workshop it's it's
1:40:55 two hours once a week for three weeks
1:40:57 starting October 1st so if that's
1:40:59 something you'd be interested in just go
1:41:01 there
1:41:02 um and I'm trying to think what else
1:41:04 just I'll go back to this this is kind
1:41:07 of the most important thing
1:41:09 um
1:41:11 if you haven't played with these tools
1:41:13 yet
1:41:14 start with Claude as Sam told you
1:41:18 um or Chachi PT chat.openai.com or Bard
1:41:23 any one of these three
1:41:24 and just start playing with them just
1:41:26 start seeing what they do and how they
1:41:29 do it and and it will serve you well
1:41:31 also I've got
1:41:33 um
1:41:35 the recordings of these lives I'm
1:41:37 hosting on
1:41:39 a YouTube channel AI learning lab Dash
1:41:41 TT
1:41:43 um and if you go there you'll look at
1:41:45 the impressive descriptions of each
1:41:47 episode that was a that's all Claude
1:41:52 um I figured it took me
1:41:55 God it took me it probably took me 50
1:41:58 attempts
1:41:59 at that prompt to get it even close to
1:42:03 giving me what I wanted
1:42:05 because I'm trying to figure out how to
1:42:06 chapterize these these big long things
1:42:08 that are just kind of random but anyway
1:42:11 um hope that's good listen I appreciate
1:42:15 you all
1:42:17 let's get up a pretty little image here
1:42:19 to go to go night night with we started
1:42:22 with that one let's see let's see which
1:42:24 one we got here
1:42:26 that one's good the little the little
1:42:30 cranky the cranky bartenders with the AI
1:42:32 salon
1:42:33 the AI Salon booze
1:42:37 um
1:42:38 all right so we're gonna do a little
1:42:39 Romper Room here Nancy good night micro
1:42:42 thanks for coming Sam corn thanks for
1:42:45 your contributions tonight that was
1:42:46 awesome
1:42:47 um forever hooked welcome thank you
1:42:50 micro Thelonius Emilio's good to see you
1:42:53 uh Nancy jrc
1:42:56 um thank you all peace out digital Gods
1:43:00 thank you so much I appreciate the
1:43:02 lemonade lemonade there that tasty
1:43:05 lemonade
1:43:06 can't wait till Ben gives you the survey
1:43:10 share bear thank you oh cool Shivers
1:43:13 signed up for the workshop that's
1:43:14 awesome
1:43:16 um is Claude still exclusive to Slack
1:43:22 it really is [ __ ] days of our AI
1:43:25 learning lab models
1:43:27 man I appreciate these lives thank you
1:43:29 Paulie Z I appreciate that and thanks
1:43:31 for whoever pinned that that's really
1:43:32 cool all right everybody nostalgic fix
1:43:35 thank you Joker good to see you micro
1:43:39 peace out everybody I hope you have a
1:43:41 good night and I will see you tomorrow's
1:43:43 Friday so I'll see you Friday night what
1:43:46 are you doing Friday night I'm watching
1:43:47 some old guy on Tick Tock what are you
1:43:49 doing I'm doing the same thing
1:43:51 thank you
1:43:52 I don't know why you would do that but
1:43:54 I'll see you there