aiLL 09/23/23 - Exploring the State of AI Tools Like ChatGPT, Midjourney, and More

Description

Chapters

Transcript

0:03 mm-hmm
0:08 [Music]
0:24 all right let's see
0:52 welcome everybody
0:54 how can I use AI to find clients well
1:00 I don't know it depends
1:02 what's your business what are the
1:04 clients
1:06 what do you mean by find clients
1:09 um
1:12 you know they're like it might give you
1:14 strategies for for looking on things
1:17 like um
1:19 LinkedIn groups or Facebook groups or
1:25 strategies for for Twitter
1:28 um
1:34 yeah I'm not sure companies need help
1:37 Staffing oh companies that need help
1:39 Staffing
1:43 I don't know I would I mean I I would I
1:46 would use General sort of chat GPT for
1:48 that to um
1:50 to
1:54 give you strategies and then you might
1:57 experiment with Bing or Bard that are
2:01 connected to the internet
2:02 and see if it can find you
2:05 companies that have posted something
2:08 that indicate that they're
2:12 you know
2:13 hiring
2:15 but you could probably you could
2:17 probably do some some paid stuff with
2:19 LinkedIn I'm not I'm not anything less
2:22 than 15 000 people a dramatic
2:24 disappointment
2:26 if you weren't here last night we were
2:29 going along fine and it was 50 and then
2:31 it was 75 and then it was like I don't
2:34 know 100 we broke a hundred and then
2:38 thanks Stevo and then all of a sudden
2:40 hate Tobias
2:41 um all of a sudden I look up and it's
2:43 like 4.6 K and then it was like 7K and
2:47 it was like 10K we hit a peak of 17 700
2:51 viewers
2:53 so so
2:54 one of our uh one of one of the
2:57 Irregulars from South Africa decided
2:59 that that the AI learning lab needed a
3:03 boost last night so I was just like holy
3:07 [ __ ]
3:08 um what was amazing was was afterwards I
3:10 looked at the stats and the stats the
3:12 the engagement of those 17 000 people
3:15 was very very low like the average view
3:17 from them was like uh four seconds or
3:20 something like that so I don't it must
3:22 be it must be some sort of service where
3:24 they're they're literally just clicking
3:25 into lives clicking out clicking into
3:27 lives clicking out I don't know who
3:29 knows anyway that's kind of fun
3:32 um oh right uh looks like Silver Fox
3:35 shared this life so yeah if you'd be
3:36 willing to share the live
3:38 um let people know that you're a weirdo
3:42 you're not weird you're in a regular
3:44 that that you're curious about this AI
3:47 [ __ ] and if they're like oh AI isn't
3:50 that stupid aren't you weird yeah let
3:52 them let them think you're weird let
3:53 them think you're weird go ahead and do
3:55 it do it see if I care see if I care
3:58 Paulie Z what's happening thanks for the
4:00 input last night great info awesome
4:01 appreciate it
4:03 they know I'm weird already see this is
4:06 so you have to at some point you just
4:09 have to embrace the uh
4:11 the oddness that is you and and you know
4:14 quite frankly with with all the stuff
4:16 that's coming with this AI stuff they're
4:17 gonna be a lot of weirdness there's
4:19 gonna be a lot of weirdness
4:21 so I think I think we're all just trying
4:24 to figure it out so why not figure it
4:26 out and have a little fun that's what
4:27 this channel is all about uh my name is
4:30 Kyle Shannon this is the AI learning lab
4:31 and that's Ali the AI learning lab get
4:34 it it's it's a play on words a lab is
4:39 like a laboratory like like this well
4:42 you might think this just looks like a
4:44 home office but no it's a laboratory
4:47 and then also a labrador you see like
4:50 it's just I don't because I just I'm
4:53 clever I'm clever a lot of times people
4:55 come in here and they're like sir what
4:57 are your qualifications watch all your
4:59 own qualifications huh
5:02 well I can do word play huh lab and lab
5:07 it wasn't my idea I think it was
5:10 actually Nancy's idea she's one of the
5:11 Irregulars yeah she came in one night
5:14 and she said uh make a graphic with a
5:16 yellow dog on it you'll thank me I was
5:18 like oh a lab the lab oh I see
5:22 so anyway that's why we have the
5:24 Irregulars here by the way they're the
5:25 ones that are
5:27 here every night with me they're odd
5:30 Tobias I'm so sorry that 17K freaked me
5:33 out I bailed on you I figured I figured
5:36 some hack was happening
5:39 that's really funny you're like I'm
5:41 going down they're going to take all the
5:43 money out of my bank account through my
5:45 Tick Tock that's [ __ ] awesome
5:49 was it thirteen thousand nine hundred
5:52 thousand Bots it was 17 700 at the peak
5:56 that I saw
5:57 um
5:59 I don't know that it was Bots I it was
6:02 because a lot of the a lot of the
6:04 Irregulars were saying it was a bunch of
6:06 people from Vietnam I don't know how
6:07 they knew that but
6:10 um it was probably you know engagement
6:12 farms in Vietnam is my guess but yeah
6:15 but it was it was it was just crazy it
6:18 really was crazy
6:21 I I wasn't you know as funny as I was
6:24 okay we're up to 114 people now so we're
6:26 good so yeah last night we had a crazy
6:29 influx I didn't really talk about the
6:31 salon and the Discord because like I had
6:34 no idea who the people were so I didn't
6:36 want to necessarily invading invading
6:39 you know that sort of ongoing space you
6:41 know [ __ ] up alive is fine
6:44 I'm I'm exploring using it for I don't
6:47 wait for identifying learning outcomes
6:49 for student readings
6:51 oh that's really interesting
6:54 for identifying learning outcomes I
6:56 don't what does that mean brother 52
6:59 I don't know what that means identifying
7:01 learning outcomes like like analyzing
7:03 their work or or analyzing what they
7:06 learn and then how do you do that is
7:08 that a is it like a a form
7:10 Oh I thought I was it was just all
7:13 Librarians I told you about finally
7:15 tuning in yes yeah it was they put us up
7:19 on the uh the Nash the big screen at the
7:22 at the stadium event for librarians in
7:24 Detroit
7:26 everybody logged in with their phone
7:28 could have been that it could have been
7:31 a bunch of librarians I don't know
7:34 oh that's really funny uh welcome
7:36 everybody my name is Kyle Shannon this
7:38 is the AI learning lab
7:39 and uh a little bit about my background
7:42 I'm an entrepreneur
7:44 um I've been I've been uh digging deep
7:47 into this AI stuff for about two years
7:49 this channel is about kind of that that
7:51 inflection point that started on
7:53 November 30th 2022 when Chachi BT was
7:56 launched uh you did great and and not
7:59 flustered I would have froze I was
8:00 someone noticed that that I blushed I
8:03 was like what do I do with this and like
8:05 every time I looked up there was another
8:06 5 000 people so I just kept restarting
8:09 like you know my little my little
8:11 starting Spiel that I do
8:13 so but thank you for the kind words
8:16 um anyway I've been digging into this uh
8:18 this this generative AI stuff for a
8:21 while and and
8:22 that moment of when chat GPT launched
8:26 and then went from zero to 100 million
8:29 users in six weeks changed everything
8:31 and it's this inflection point that
8:33 prior to that inflection point machine
8:35 learning and artificial intelligence was
8:37 reason really the purview of researchers
8:40 and scientists and engineers and things
8:41 like that
8:42 after that day
8:45 it's the rest of us
8:47 the rest of us get to play with these
8:49 tools so so we get to we get to we're
8:52 lucky enough to be living at a time
8:56 when we can just go to this thing called
8:58 Chachi PT
8:59 chat.openai.com or this thing called
9:01 Claude or this thing called Bard and we
9:05 can just ask it questions and it answers
9:07 them and it's and it's not a Google
9:09 search it it may seem on the surface
9:11 like a Google search but if you have not
9:14 played with Chachi BT go there right now
9:16 go to chat.openai.com start playing with
9:18 it
9:19 and you'll see it's there's there is a
9:22 there's a
9:24 there's a something you gotta play to
9:26 win you know what I'm saying you just go
9:28 there you start playing you gotta play
9:31 to win
9:31 I actually wrote an article today on
9:34 LinkedIn
9:35 um
9:37 about lateral play and and what I mean
9:40 by lateral play is like you know I've
9:42 got my skills that are you know I've
9:44 spent
9:45 you know my my five plus decades honing
9:49 and lateral play is like playing outside
9:53 of that like I was never good at drawing
9:55 or art so me going to mid-journey is is
9:58 lateral play it's it's off to the side
10:00 of what I do but I really enjoy
10:02 creativity and I enjoy visuals and I'm a
10:05 photographer like there's a lot of
10:06 things I do that if I had the skill to
10:09 draw I would do good stuff with it
10:13 and now with mid-journey I can I can
10:16 take my wordy brain and make it a
10:19 picture brain or I can go over and play
10:22 with code where you know I've got add
10:24 and I never really learned code Beyond
10:26 you know scripting in HTML and [ __ ] like
10:28 that and now I can go write python right
10:31 so lateral play is this idea that
10:35 we can use chat GPT to
10:38 improve our productivity at work
10:42 but all of that lateral play feels like
10:44 what I wrote in the article is is that
10:47 lateral play kind of felt like I was
10:49 cheating on my business like like this
10:52 is too much fun like I'm having fun over
10:53 here that's not really work though but
10:55 what I learned is it was work and it was
10:58 contributing and it was contributing to
10:59 the quality and the productivity and in
11:02 ideation and all sorts of stuff like
11:04 that so so a big part of what this
11:06 channel is about is just [ __ ] play
11:08 just go there right now don't you don't
11:10 have to be technical
11:13 I have a degree in acting all right
11:18 you know I spent years as a Tom Arnold
11:20 impersonator look at this it's [ __ ]
11:23 uncanny
11:24 he and I are like we're best I've never
11:27 met him but that's not the point okay
11:30 I made as much as ten dollars an hour
11:33 impersonating Tom at malls at mall
11:35 openings but they there aren't malls are
11:38 all closed now but it was a good gig for
11:41 the week that I did it it was I anyway
11:44 back to AI
11:48 he seems like kind of a dum-dum I I
11:52 don't know I went on this place it's
11:53 called AI learning lab and I thought oh
11:55 it's going to be a scientist and I was
11:56 expecting him to wear you know like a
11:58 like a white coat like a lab coat and
12:01 then and it was gonna be like there's
12:02 gonna be like computer screens behind
12:04 him and it was just like a basket and
12:06 some magazines and they seemed like kind
12:08 of a dumb dumb and then and then he kind
12:11 of made fun of people asking what his
12:12 qualifications were and I was sitting
12:13 there thinking well what are your
12:14 qualifications
12:15 what are you qualificating I don't even
12:17 know
12:18 how long did you watch it I watched it
12:20 for I I don't know I watched it for like
12:22 three hours or so you spent three hours
12:24 on Tick Tock with some idiot well yeah
12:26 yeah did you learn anything I kind of
12:29 did I kind of I kind of I think they say
12:32 I think it's going to be big
12:33 it could be bigger than the internet
12:35 that's what he said he said it could be
12:37 bigger than the internet and I and I
12:38 think the internet's big I got it
12:40 installed on my phone and it's am I I'm
12:43 all out of I'm all out of room on the
12:45 phone now because I think I put all the
12:46 internet on it you know what I mean it's
12:48 just [ __ ] this is crazy it's just
12:50 crazy anyway
12:52 um do you want some corn pie
12:55 I made a lovely corn pie today would you
12:57 like a slice okay let me go get that for
12:59 you all right and scene all right
13:02 trying to learn how to advertise with AI
13:05 where's the best play to go to learn
13:06 place to go to learn so so what I would
13:09 do he
13:12 surmise that I'm frustrating to a lot of
13:15 people
13:16 um my answer to almost every question
13:19 right now and probably for the next 18
13:21 months is there's no place to go to
13:25 learn
13:26 how to do advertising with AI
13:29 because all of this stuff is changing so
13:33 fast now now that doesn't mean you can't
13:35 learn with AI about advertising
13:38 but what I would do is and I why don't
13:41 we why don't we go do it right now let's
13:42 let's go play
13:45 um
13:55 and I sing Hawaii Five-O songs
14:00 yeah all right so we are at chat GPT
14:04 this is chat GPT if you've never seen it
14:06 this is it it's you're like that's it
14:10 why is everyone talking about it doesn't
14:12 look very exciting it's not it's a janky
14:14 piece of [ __ ] right now so down the left
14:17 hand side here are your little uh
14:19 your your history
14:22 these are little conversations I've had
14:24 with chat GPT these things here are just
14:27 uh suggestions like if you literally
14:29 can't think of anything you just click
14:30 it just go click and it'll make a prompt
14:33 for you
14:35 um
14:36 there there are two different kind of
14:38 modes of chat EBT there's the free mode
14:40 which you get access to chat CPT 3.5 and
14:43 then there's paid mode which is called
14:45 plus and you get access to gpt4 and to
14:48 plugins and to I don't know a bunch of
14:50 stuff
14:51 so
14:53 so let me show you one other thing
14:55 because I I'm kind I I'm not copping out
14:58 on the answer wait why oh no
15:03 I think Microsoft Edge just ate my
15:06 memory
15:08 because I can't go you know and quit
15:10 Microsoft Edge hang on we'll come back
15:11 to this
15:16 okay I'm gonna quit
15:19 leave
15:25 Microsoft Edge is full of memory leaks
15:29 yeah if you've got too many tabs open
15:32 because you have the add
15:36 Edge will crash on you
15:39 every three hours or three
15:43 and it looks like it hung my machine
15:46 yay you did it yay yay Microsoft you're
15:52 a winner keynote not responding
15:55 uh good Lord force quit
16:00 I know
16:02 hold please
16:05 well he you know someone asked him his
16:07 qualifications and then he said he kind
16:10 of made fun of it like well he doesn't
16:11 need to have qualifications and then he
16:13 couldn't get his computer to work and
16:15 it's like I just felt like I was
16:16 watching a geriatric trying to make a
16:18 computer work and he kept asking about
16:20 how does tick tock work
16:22 he I think his wife's name is Marge or
16:25 something he goes Marge how's this thing
16:26 work anyway it it was sort of funny but
16:30 then I just wondered well what are his
16:32 qualifications
16:33 what what were his qualifications and I
16:36 I still don't know I still don't know
16:38 but you watched it for three hours well
16:40 that's not it's not really really I had
16:42 a Coco I had a little cocoa and I was I
16:45 was making the corn pie all right I
16:47 gotta start I gotta restart my machine
16:48 people
16:50 it's it's just not in a happy place
16:53 yeah
16:56 QuickTime Player interrupted it try
16:59 again
17:00 do I want to save that why is that
17:02 distorted there's just weirdness going
17:05 on
17:09 I'm gonna say don't save there uh I'm
17:12 gonna say don't save there
17:14 I will say oh that's recovered
17:18 hang on sorry about this people
17:22 um did you find out why so many people
17:23 were in the in the live last night
17:26 um yeah it was we're down we're down to
17:28 50. we're down to a manageable 55
17:30 tonight the uh the peak last night was
17:33 17 700.
17:35 um
17:37 um
17:38 what's his name James John uh wait hang
17:41 on
17:42 one of the one of the Irregulars the
17:44 dude from South Africa
17:47 whose name James
17:49 kamua
17:52 he's from he's from South Africa he's
17:54 from uh nehru Nairobi County Kenya
17:58 according to Firefly Firefly did some
18:00 some sleuthing
18:02 um
18:05 no I don't want to force quit
18:08 um
18:08 I'm gonna have to
18:11 yeah boy yeah this this machine is not
18:15 happy ignore okay
18:20 hang on hang on people it's gonna be so
18:24 worth it
18:24 [Laughter]
18:30 oh man so
18:33 I think so James said last night that it
18:36 was him that that he liked he liked the
18:38 live and so he just sent like 5 000
18:41 people and I I like I don't know how you
18:45 um get like thousands of people to view
18:48 alive I assume there are farms like
18:50 there are engagement farms in China a
18:53 lot of the people last night said it was
18:54 uh Vietnamese a lot of Vietnamese people
18:57 so
18:58 it might have been some engagement Farm
19:00 in uh in Vietnam
19:03 that that he used
19:05 um but it was it was really exciting I
19:07 mean it was just [ __ ] insane we went
19:09 from like 100 people to like 17
19:12 000. and I was just like huh what do I
19:15 do
19:17 um what was wild though is that it um
19:19 boy this thing is just this is bad I
19:22 think I'm just gonna do a hard restart
19:24 on this thing
19:26 um
19:28 [Music]
19:28 um
19:29 am I
19:31 doing
19:38 yeah I went and I looked at the um I
19:41 looked at the engagement after
19:43 the live was over I went for like an
19:46 extra hour because I didn't know what
19:48 was happening I didn't know if it was
19:49 tictographers James or what it actually
19:51 meant and I looked at the engagement for
19:54 the um
19:56 you know for the live and it it wasn't
19:58 dramatically different than you know
20:01 when we just have like 50 or 75 people
20:03 in here so
20:05 so that's why I think it was engagement
20:08 farming which is just basically like P
20:10 and the the average view of
20:12 non-followers of the channel was four
20:14 seconds and the average view of
20:16 followers was like 18 minutes so so
20:19 there was definitely a difference
20:21 between people who knew about the the
20:23 lab and those that didn't
20:26 um okay there's Poe
20:30 all right so it restarted all that stuff
20:32 let me restart Keynote
20:34 I'll be with you let's go seriously this
20:36 is gonna be really good what's your
20:37 opinion on the famous authors taking
20:39 opening out of court for copyright
20:41 um I understand why they're doing it I
20:44 don't think they understand how
20:46 chat GPT works or how large language
20:48 models work
20:52 here here's this is my real opinion on
20:55 it
20:57 if mid-journey or stability AI or open
21:00 AI or Google or any of them
21:05 went to all copyright holders first and
21:07 said hey we're thinking about training
21:09 up a model on all of the internet which
21:11 includes some of your [ __ ] do we have
21:13 your permission everyone would have said
21:14 no
21:15 they would have been stuck in legal
21:17 battles for decades so so
21:21 we are in a situation where they all
21:24 basically had to break some eggs and ask
21:28 for forgiveness rather than permission
21:30 and that's that's where we are right now
21:33 I equate it if you understand how
21:35 embeddings work and the fact that the
21:37 original documents are not being stored
21:39 in some big database they're they're
21:41 turned into vectors they're turned into
21:42 mathematical numbers in a thousand
21:44 dimensional Vector space
21:47 where all these words are represented in
21:49 semantic clusters now those semantic
21:52 clusters are derived from all that
21:55 copywritten work but it's not like when
21:58 you search Chat GPT it's like oh let me
22:00 go into silver Sarah Silverman's book
22:02 and copy out a passage and paste that
22:05 into your answer that's not how they
22:06 work it's more it's more equivalent to
22:12 someone goes to college for four years
22:14 and they study the literary works of
22:17 stand-up comics and they come out after
22:19 four years and they know all these
22:21 different references from all these
22:23 books from the papers they've written
22:24 and what they've studied and all the
22:25 stand up they watched and this and that
22:27 and you ask them to write something and
22:30 they quote something from Sarah
22:31 Silverman's book that they love like
22:33 it's more like that it's more like
22:35 recall from some nebulous
22:39 mathematical brain of all these Concepts
22:43 and semantic clusterings of words than
22:45 it is copywriting but
22:48 it was trained on some of that work and
22:53 mid-journey was trained on copywritten
22:55 images and but it's not stealing them
22:57 it's not how the technology works so I
22:59 think so it's so it's a very gray area I
23:01 understand why they're doing it
23:03 um
23:04 and uh
23:07 I don't fault them for it but I I don't
23:10 think they're going to hold up in court
23:13 what's likely going to happen is open AI
23:17 is going to cut
23:18 um you know 100 million dollar deals
23:21 with big Publishers like the New York
23:23 Times and time Inc and they already did
23:26 one with the AP the Associated Press
23:29 and so all of these all of the big
23:31 companies are going to cut deals with
23:33 big companies and then basically below
23:36 the 100 million dollar deal threshold
23:39 it's going to be the legal equivalent of
23:41 go [ __ ] yourself is is what I think now
23:43 does that mean that people aren't going
23:46 to keep suing them oh no people are
23:48 going to keep suing them again and again
23:50 and again and again and again and at
23:53 some point they'll probably counter Sue
23:55 um Microsoft put up put out a statement
23:58 last week or the week before
24:00 that they're going to pay the legal
24:04 bills of anyone who gets sued because
24:07 they used
24:08 chat GPT to generate content basically
24:11 so that says to me that their customers
24:13 are worried about it right and their
24:15 customers are saying well we can't use
24:16 this if there's any risk we're going to
24:18 get sued so Microsoft said we'll pay for
24:20 your lawyer bills
24:22 um that says to me if I were a lawyer
24:25 I'd be like let's go sounds like class
24:29 action it sounds like class action it
24:32 sounds like class action let's go yeah
24:36 all right
24:40 see you Denver okay
24:44 all right now I've got keynote back to
24:46 where it was let me get edge up and
24:49 running
24:51 thank you to Microsoft for absolutely
24:53 smashing
24:55 the memory in my Mac to the point that
24:57 it was non-recoverable that's an awesome
25:00 feat
25:01 it's actually hard to get Max to crash
25:03 that hard
25:05 very accomplished if if opening I lose
25:09 this case though what would it mean
25:11 would they have to retrain on copying so
25:13 here's the thing
25:15 they can't
25:18 you can't
25:19 so so like I said it's not like there's
25:22 a giant database with all this data in
25:24 it what they do is they take all the
25:26 data they found and they put it into
25:29 um into into a form that that it can be
25:31 embedded right it can be turned from
25:34 words into mathematical vectors
25:38 and and they do this across billions if
25:41 not trillions of documents right at
25:44 least trillions of words
25:46 probably billions of documents trillions
25:49 of words and what goes into the vector
25:51 database are all these numbers that are
25:53 tied to all these words
25:56 and
25:58 there there's literally no way to go
26:01 well you know let me go into the Sarah
26:03 Silverman section and pull that data out
26:05 if she if she won that lawsuit so the
26:08 only option is they shut the [ __ ] down
26:11 and start over and I just I don't think
26:14 that's gonna happen I like it's I I
26:16 don't think there's a a judge or like I
26:19 don't know how how it would get
26:21 adjudicated but I I just
26:23 the the stifling effect that would have
26:27 on on the industry it just I I just
26:30 can't imagine it happening right I think
26:32 I think all of those lawsuits are going
26:34 to settle
26:35 um outside of it
26:38 all right let me see
26:42 what was that gonna show oh I was gonna
26:45 show let's go to uh what's it called
26:50 chat.openai.com so if you're new here
26:52 welcome got a few new people in here my
26:55 name is Kyle Shannon this is the AI
26:57 learning lab if you've not played with
26:58 chat GPT
27:00 go
27:02 right there
27:04 chat.openai.com that's what I'm doing
27:05 right now I'm going to chat.openai.com a
27:09 while ago before where before I found
27:12 out my computer was hard crashed
27:15 um someone asked me about what AI would
27:18 I use to
27:20 um
27:21 to learn advertising and I said well
27:24 there's not really an AI that you can go
27:27 to to learn advertising there might be
27:30 that URL on the very bottom
27:31 futurepedia.io
27:34 is like thousands I don't know four
27:37 thousand forty five hundred AI tools
27:40 so you can go there and you can search
27:42 for advertising but I do not recommend
27:44 it
27:46 my stock answer to any very very
27:49 specific request like how do I learn
27:52 this how do I learn that how do I you
27:54 know how do I get AI to do you know this
27:57 specific thing for me there is probably
28:00 some sort of tool at futurepedia.io most
28:03 of those tools
28:05 are just tools that someone figured out
28:07 a prompt that generated a cool output in
28:11 chatgpt and then they turned it into an
28:13 app they're called wrapper apps they're
28:16 basically just doing calls to open AI
28:18 with with a clever prompt or a hundred
28:21 and then they charge you 39.95 a month
28:24 and it's all stuff that you could do
28:26 anyway so
28:28 so what I wanted to do
28:30 let's go show everyone
28:32 um how I would attack the problem so
28:34 here we are at chat gbt this is my chat
28:37 history
28:38 um there's Chachi PT 3.5 and 4. you have
28:41 to be paying the 20 bucks a month to get
28:43 to gpt4 I think it's worth it but you
28:45 don't have to so we'll just do 3.5 for
28:47 now these things are just you know uh if
28:51 you literally can't think of anything
28:52 you click on those and it'll write a
28:54 prompt for you
28:56 um
28:56 so what I what I'm going to do here is
28:58 I'm going to say
29:06 oh come on
29:08 why is that not working now
29:13 uh
29:15 it's serious messing with me again
29:21 Yeah Siri I thought so
29:25 stop Siri go away
29:31 Siri shut down
29:34 uh
29:36 I don't know how to turn her off
29:40 I'll just return Syria this is really
29:44 annoying
29:56 um
29:56 [Music]
30:00 oh good God okay I suppose I could type
30:03 but I don't want to type
30:05 I want to use my microphone
30:19 oh something happened
30:22 hello
30:24 ah this is this is maddening this is
30:27 just Madness you know it's got to be
30:29 worse than what I'm going through is you
30:31 watching this there we go
30:34 okay now it's working
30:37 I want you to act
30:40 like David Ogilvy
30:45 and
30:47 help me understand
30:51 the world of advertising period
30:54 I'm just getting started and want to
30:56 learn advertising and I want you to
30:59 serve as my mentor
31:01 period
31:04 do you understand what I'm asking for
31:07 and what do you think is the best way
31:09 for us to begin
31:13 so and and so now it's it's just popping
31:16 up an answer now why did I do that I
31:18 could have just said teach me about
31:20 advertising the way the large language
31:21 models work is if you just don't give it
31:25 any context it basically pulls from
31:28 everything
31:29 so
31:30 there are idiots that talk about
31:32 advertising and there are ignorant
31:34 people that talk about advertising and
31:36 there are brilliant people that talk
31:38 about advertising and there are clients
31:40 that talk about advertising so if you
31:42 just say tell me about advertising it's
31:44 kind of pulling from all these different
31:46 data sets at the same time and they just
31:48 kind of get mushed together and what you
31:50 end up getting is very very generic [ __ ]
31:52 so by me starting out saying I want you
31:54 to act like David Ogilvy who's you know
31:56 one of the most famous admin in all of
31:58 the land
32:00 um
32:02 and say that I want David to Mentor me
32:05 and ask things like do you understand
32:08 what I'm asking you it's providing
32:10 what's called In The Biz context right
32:13 so it's kind of narrowing the window
32:15 where chat GPT is going to focus when
32:18 we're having our conversation so the
32:20 answers tend to get much much better
32:22 so let's see what it said
32:25 of course I understand your request I'd
32:27 be delighted to guide you through the
32:28 world of advertising much like David
32:30 Ogilvy would David Ogilvy was a true
32:32 advertising Legend known for his
32:33 principles
32:35 to begin your our mentorship in
32:37 advertising we should start with some
32:39 fundamental concepts understanding your
32:41 audience the big idea research creative
32:45 execution headlines and copywriting
32:46 testing and measurement media planning
32:48 brand image and continuous learning
32:52 um so then I'd say
32:54 um
32:57 well I just want to make some social
32:58 media ads should I use Facebook or
33:00 Twitter
33:02 now this should slap me back a little
33:06 bit and say well you should start with a
33:08 strategy it should do that it shouldn't
33:10 just let me go right to a tactic but
33:12 let's see
33:15 yeah choosing between Facebook and
33:18 Twitter for your social media depends on
33:20 your target audience goals and the
33:21 nature of your product or service both
33:23 platforms have their strengths and
33:24 weaknesses so let's consider some
33:26 factors to help you decide audience
33:28 demographic demographics Facebook
33:30 Facebook has a broader and more diverse
33:33 user base Twitter attracts a younger
33:35 audience with a significant presence of
33:37 18 to 34 year olds although I don't know
33:40 that that's accurate these days but it
33:43 might be
33:44 um so anyway so so this is what I would
33:48 do is if if
33:51 if I were trying to figure out
33:53 essentially any problem right now I
33:56 would start with having a conversation
33:58 with chat gbt now or or Claude or Bard
34:01 it doesn't matter which tool you use
34:04 um start by having a conversation with
34:06 one of these things because it will
34:08 educate you really quickly and then you
34:11 can start to go explore specific areas
34:14 and things like that so anyway have you
34:16 seen chat Dev yet no I haven't
34:18 I haven't even heard of it let's go see
34:20 what chat dev's all about
34:23 I assume it's a coding bot is it built
34:26 on top of code llama
34:27 chat Dev
34:30 GitHub chat Dev create stands uh
34:38 the primary objective of Chet chat that
34:40 was creating easy to use highly
34:42 customizable extensible framework which
34:44 is based on large language models
34:46 oh I see so this is this is a oh
34:49 llama I got myself a llama from Winston
34:53 thank you Winston hello Silver Fox I am
34:56 llama you guys are awesome
34:59 um yeah it looks like this is a thing to
35:02 to allow you this is this is probably
35:03 not just similar to uh what the hell is
35:06 it called GPT for all
35:09 there's a project called GPT for all
35:11 where you can install
35:13 um an application on your Mac or Linux
35:17 or Windows that you can then install
35:19 large language models in and create your
35:21 create your own chat Bots you can also
35:22 do that in zapier they've got this thing
35:24 called interfaces now
35:27 um
35:28 so
35:31 let me look over on the Irregulars
35:33 Channel okay everything's good there
35:36 turn Siri off by going into settings
35:38 scroll down to Siri and search top three
35:41 buttons thanks Nancy you know I think
35:43 I'm going to keep it on because the
35:45 comedic gold that Siri gave us when she
35:47 got jealous of Bing I don't want to lose
35:50 that but boy is it annoying when she
35:52 takes over my microphone
35:58 uh oh I don't have my photos thing up
36:01 hang on let me get photos up so I can
36:03 get all that pretty pictures back like
36:05 like Ali the AI learning lab there she
36:08 is she's so cute
36:13 all right
36:16 okay
36:17 so welcome everyone uh if you're new
36:20 here pop any questions you have about AI
36:22 in the comments below I'm happy to talk
36:24 about them I'm happy to talk about my
36:26 background I'm an entrepreneur I was an
36:28 entrepreneur in the early days of the
36:29 web
36:30 um the parallels between the early days
36:33 of the web and the early early days of
36:36 this AI stuff are uncanny and wild and
36:40 exciting and
36:42 um
36:43 I feel lucky to be alive
36:47 at the beginning of one of these things
36:49 for the second time in my life
36:51 um I just missed the PC one
36:55 but um but I got the World Wide Web one
36:57 and I'm here for this one so anyway it's
36:59 it's exciting times
37:02 do
37:09 [Music]
37:11 all right so we don't have a ton of
37:13 questions maybe did I miss a bunch I may
37:15 have
37:16 it might have jumped me to the bottom
37:17 Tick Tock jump into the bottom
37:20 you talk
37:22 Camelot
37:28 oh yeah I missed a lot of questions okay
37:30 if you have questions pop them in the
37:32 comments below I also do have some mods
37:34 in here they are they are formed from
37:36 the Irregulars so the Irregulars are
37:38 people who show up here regularly
37:40 um if you're interested in modding and
37:42 you show up here a lot just let me know
37:44 DM me and I'll find a way to get you
37:47 hooked up
37:48 uh yeah there's turn Siri off I saw that
37:51 earlier I'm I'm Gonna Leave Syrian just
37:53 for the comedic value
37:55 [Laughter]
37:59 um
38:05 Kyle seems so excited finally the
38:07 crowd's showing up for the play yeah
38:10 well I I mean I was excited because I
38:12 thought it might have been
38:14 like Tick Tock just blasting the
38:17 algorithm but it wasn't it
38:19 it was engagement farming and it you
38:23 know when it went from like
38:26 a hundred to forty six hundred and then
38:29 from that to 10 000 and then from that
38:31 to 17 000 that's that's not you know I
38:35 could see if Tick Tock did it it would
38:36 probably go to like 100 then 200 then
38:39 you know 500 maybe but it's not going to
38:41 jump you know 10 you know thousands at a
38:44 time I'm hosting my first AI seminar in
38:47 two weeks friends signed up six people
38:49 then told me
38:52 um wait why is this not working
38:55 told me what she'd done oh that's funny
38:57 I'm I'm doing a uh I'm doing a workshop
39:00 too in two weeks starting on October 1st
39:03 or 4th October 4th so here let me let me
39:06 let me pull up the info for it
39:09 and I'll go through I'll actually I'll
39:11 I'll start with the the workshop that
39:13 I'm doing so I got invited by a woman
39:17 named Rhonda Phillips who's got a uh
39:19 she's a creative coach and a voiceover
39:21 actress and she's got a whole community
39:23 of creative people so we're doing this
39:26 thing called the AI launch pad
39:28 um where we're doing uh three two-hour
39:31 sessions a week apart so October 4th
39:34 11th and 18th if you go to
39:36 rhondasvoice.com events you can see that
39:38 there
39:40 um
39:40 and yeah it's a similar sort of thing
39:42 where it doing workshops right now I've
39:45 I've real mixed feelings about it I
39:47 think it's a good thing but everything
39:49 is changing so fast that I'm really
39:52 reticent to put too much down in writing
39:55 to say this is a good thing so so what
39:57 I'm going to try to focus on as much as
39:59 possible is
40:00 the craft of of playing with AI as
40:05 opposed to here are six prompts that are
40:07 perfect and and things like that so
40:09 anyway it should be really good
40:12 um
40:14 and yeah so that should be fun
40:16 um okay so if you're new here let me let
40:20 me
40:20 show you a couple of things these are
40:23 these are my kind of recommended things
40:25 to go play with if you're brand new to
40:27 AI if you haven't played with chat gbt
40:29 before that's it
40:32 it's not chatgbt.com it's
40:34 chat.openai.com go there right now and
40:36 start playing or go to Claude
40:39 or go to Bard either one of these all of
40:41 these are have a free option Claude is
40:44 really powerful because you can put in
40:46 giant prompts and it doesn't choke so
40:48 you can copy and paste really big
40:49 documents into it and and just start
40:51 interacting with it in fact it's a it's
40:54 most of the way that I use Claude is if
40:56 I've got a document that's bigger than a
40:59 page or two because I know chat CPT will
41:01 choke on it Claude won't
41:04 so just go play with those if you don't
41:05 know what to say there at all if you're
41:08 just like I don't know what to do Kyle
41:10 it's just a box what do I put in the box
41:14 go to prompts.chat go to that thing down
41:16 there and that'll tell you what to put
41:19 in the box
41:21 and you you all you always want to make
41:23 sure you put good stuff in the Box you
41:25 can put good stuff in the Box you get
41:27 good stuff out what is corn pie
41:31 you ain't never had corn pie oh sweetie
41:35 you ain't live till you had some of my
41:37 corn pie now you got to get to Del Monte
41:39 you get to Del Monte whole sweet corn
41:42 don't get the store brand don't get the
41:46 store brand you want the Del Monte okay
41:50 and you got to get the sweet one
41:52 because the one they got the one in
41:54 Spring Water you don't it's it's
41:56 flavorless it's like eating it's it's
41:58 like eating low beads of Styrofoam no
42:01 you want the sweet whole corn Del Monte
42:04 that's this that's the secret that's the
42:06 whole secret and I wanna I'm not even
42:08 gonna get into my pie crust okay
42:11 all right
42:13 um anyway
42:16 we do skits here too it's like Saturday
42:19 Night Live if Saturday Night Live we're
42:21 even less funny
42:22 [Laughter]
42:25 uh okay
42:28 um so those are the three this is what I
42:30 call the adult table this is the this is
42:32 the adult table these are these are the
42:33 big three that I use pretty regularly
42:35 this is the Kids Table you'll notice
42:38 Bing is barely readable because because
42:41 I dimmed I dimmed Bing because Bing
42:44 pissed me off because the safety guard
42:45 rails on Bing are so egregious right now
42:48 that it makes it essentially unusable
42:50 but it's still really cool conceptually
42:52 it's connected to the internet it can
42:54 generate images it's going to have Dolly
42:56 three in it really soon which is gonna
42:58 be super amazing
43:00 um it can analyze images it's it's just
43:02 there's there's things about it that are
43:04 right and then it but it's it's in
43:07 timeout right now so that's why it's
43:09 there
43:10 um perplexity.ai pi and Poe are all fun
43:13 to play with okay so those are those now
43:15 for the image things if you if you're
43:17 like but Kyle I don't care about words I
43:20 like images I am a purveyor of fine
43:24 images and I would like to understand
43:26 how I make images with
43:30 um with these tools these are the these
43:31 are the ones mid Journey midjourney.com
43:34 will take you to a website and when you
43:36 say join the beta it will take you to a
43:39 whole nother application called Discord
43:41 and if you've never been on Discord it
43:43 will confuse the ever living [ __ ] out of
43:46 you wait wasn't I just in a website yes
43:48 you were and now I'm in a thing that
43:51 looks like slack but it's kind of a
43:52 shitty version of slack yes it is and
43:56 and now I have to go into slack and I
43:58 have to make an account there yes you do
43:59 and then you have to go into these weird
44:02 Channels with these these young people
44:05 and they're
44:07 saying things and making slash commands
44:10 yes and then I have to type command line
44:13 interfaces to make images yes
44:17 but it's the best so if you want the
44:20 really good images you go to Mid Journey
44:23 um ideogram not as good images but it
44:25 can do text so
44:28 that AI learning lab there that was done
44:30 with ideogram ideogram ideogram idea
44:32 Grant idea Graham janky pieces of [ __ ]
44:35 all these tools right now
44:37 these are all I'm not weird I'm
44:39 irregular
44:40 laughs
44:45 uh these are all ID gram okay
44:48 um leonardo.ai is a is a website that
44:52 you can make images at it's really good
44:54 it's built on top of stable diffusion
44:55 hey Jen I'll come back to Adobe Firefly
44:58 and Photoshop generative fill are out of
45:00 beta now which means you can use those
45:02 images commercially those are if you're
45:04 in a company that's probably your best
45:07 option because it hasn't been trained on
45:09 copywritten stuff
45:10 and then openart.ai promptbook will
45:15 teach you about how to talk to these
45:17 these things
45:19 I'm going to show you one example
45:24 of
45:28 something that hey gen does
45:31 that will blow your mind
45:37 hey look more me
45:38 [Laughter]
45:42 okay I'm not really Tom Arnold this is
45:45 Tom Arnold okay
45:48 um
45:49 hey Jen if you go to labs.hagen.com
45:53 and then click on the video translate
45:55 button
45:56 and if you upload a video of yourself
45:59 speaking I assume English if you're
46:01 listening to this in English
46:03 and you can pick one of I think it's 12
46:05 or 14 languages now that you can
46:07 translate it into watch this so I I
46:09 actually cut together some of me
46:11 speaking English with some of me
46:13 speaking Italian check this out
46:17 where is it
46:19 um
46:20 so we've heard from a number of clients
46:22 that modular content is important to
46:24 them getting more efficient with how we
46:26 create content and how we leverage
46:27 contents
46:35 behind me you see these little green
46:38 boxes um that's how we think about
46:39 stories at storyline we break stories
46:41 down into little atoms right little
46:43 modules
46:52 I don't say storyline like that I say
46:55 Story vine and and this makes me say
46:57 esta Divine uh it's amazing it's sinking
47:01 the lips it's syncing my voice it's
47:04 matching the Cadence and the inflection
47:06 of the English perfectly the timing to
47:10 all of my many hand gestures because I'm
47:12 a hand talker
47:14 um are all there that tool is
47:17 labs.hagen.com and it's their video
47:18 translate tool
47:20 um they've got a backlog right now that
47:23 is gargantuan so if you want anything
47:26 um done quickly you're gonna have to pay
47:27 for it
47:29 um but that that's a game changer in my
47:31 opinion and the uh the uh what's his
47:34 name one of the co-founders
47:37 um was talking about they're not that
47:39 far from that thing being able to be
47:41 essentially real time so so we are
47:45 probably within a year or 18 months of
47:49 true real time
47:53 translation
47:56 in video
47:58 so imagine being on a zoom call with
48:00 someone in another language in another
48:02 country that doesn't speak your language
48:04 but hearing you in your voice in your
48:08 Cadence
48:09 speaking their language fluently and you
48:12 hear them speaking yours fluently
48:15 we are essentially a year out from that
48:20 odd
48:21 so anyway that's cool
48:23 um the other thing that that might be of
48:25 interest if if you're into this AI stuff
48:29 um I think it's an adventure what this
48:31 channel is about is about kind of
48:32 leaning into the adventure of whatever
48:35 the hell this stuff's gonna be I created
48:37 a community called the AI Salon we meet
48:39 every other week we've got a meeting
48:41 coming up Tuesday night we've got a
48:43 Discord server
48:45 um we've got a a special little section
48:47 in there for the Irregulars to hang out
48:48 so if you start hanging out here you can
48:51 hang out there as well and talk you can
48:53 make mid-journey images there you can
48:56 ask questions there you can share news
48:59 it's it's a it's a good community so
49:02 um so that's what that's about
49:03 storybinds my company that's my social
49:06 media tag so anyway
49:08 that's it that's
49:12 that's some of that stuff
49:14 um by the way Irregulars you're you're
49:16 up to 20 21 000 likes you you're like
49:19 you're out like machines tonight
49:21 have a hot Saturday Saturday night date
49:24 with Kyle just going click click click
49:26 click click click click what'd you do
49:28 Saturday night well I don't want to talk
49:30 about it it's a little embarrassing
49:36 how do people create the AI videos I've
49:39 seen okay Kim Let's go do that since I
49:42 have the the thing down there
49:44 let's go make some videos
49:47 I'll show you two different ways to make
49:49 them let's see
49:52 all right we'll start we'll start with
49:54 Runway ml there there are a couple of
49:56 things there's one now called Pika I
49:58 think it's called
49:59 but there's also Runway ml so we're
50:01 gonna start with Runway ml because
50:03 that's kind of the the first one to get
50:05 it get it kind of right
50:06 and
50:10 um it's pretty slick
50:14 so it's it's uh
50:18 app.runwayml.com so they so what they
50:21 say here is bring your image to life
50:23 with Gen 2 and then you can start with
50:25 an image or you can start with text
50:28 and if you start with text you can
50:30 literally just describe something well
50:32 let's we'll do both of these so start
50:34 with text
50:40 um
50:50 a man and a woman holding hands walking
50:53 through a field of wild flowers
50:57 on a breezy day
51:01 I don't know that's good
51:05 generate
51:08 four seconds so
51:11 um I haven't spent a lot of time doing
51:13 text to video but for the most part it
51:16 will give you what you ask for the thing
51:19 I have done is the next thing I'll show
51:20 which is image to video so so one of the
51:23 nice things about image to video is you
51:26 can go to Mid journey and make all sorts
51:28 of cool images
51:29 and then you come here and you just
51:31 upload the image and basically it takes
51:33 that that first frame is the start frame
51:35 of the video and then it makes a four
51:37 second video after that and you can
51:39 extend that four different times you can
51:41 make up to an 18 second video clip from
51:44 a single start image or from a single
51:47 um text prompt like this so what the
51:50 people are doing that are doing this is
51:53 um you know they're they're probably
51:56 generating you know
51:59 25 videos for every every single clip
52:02 that they put in one of those hand
52:04 edited
52:05 um sequences
52:07 um
52:07 but so this looks like it gave us the
52:10 woman but that not the band but let's
52:12 see if the man mysteriously emerges
52:17 no
52:20 but that's actually not bad I mean you
52:22 know
52:23 with the exception of the fact I wanted
52:25 a man and a woman
52:27 um it looks like a woman walking you
52:29 know down a trail
52:31 uh you know about to have a hay fever
52:33 attack so this this could this could be
52:36 the the good first scene from a uh from
52:39 an allergy medicine commercial
52:43 okay so that's so that's starting with
52:46 text and and now I'm going to start with
52:49 an image and so let's see I've got here
52:54 upload a file select from assets I've
52:56 got a bunch of assets already uploaded
53:00 where are they
53:03 why are they not showing up there we go
53:06 because they were loading
53:20 oh that's cool
53:22 all right so I got I've got I'm a
53:25 photographer so this is a like a sunset
53:27 Cloud thing I I took a picture of so
53:32 one of the things you can do here is
53:35 let's see uh
53:38 [Music]
53:39 is it this thing
53:48 image plus description okay motion so I
53:51 can I can change my motion how much
53:53 motion I want so I'm going to put that
53:55 kind of high so I'm going to move that
53:56 up to eight
53:58 and then that's duration
54:02 I thought there was some oh Advanced
54:03 Camera controls okay I I thought there
54:06 was something here so there's Advanced
54:07 Camera control so I'm gonna I'm gonna
54:09 open these Advanced Camera
54:11 controls so there's horizontal vertical
54:14 zoom and roll
54:16 and then the speed so I'll leave the
54:18 speed at 0.5
54:20 and I guess we'll go we'll pan it to the
54:23 right so let me go horizontal to the
54:25 right
54:26 and so now I say generate so now it's
54:29 going to give me movement of 8 out of
54:31 ten it's going to pan it to the right
54:33 and it's going to use my cloud photo as
54:35 the start image
54:36 one of the things I've noticed about the
54:39 videos that are created from images is
54:41 that the the sharpness of the image
54:44 degrades really quickly it makes it very
54:46 blurry but it does create for the most
54:50 part stuff that is interesting and
54:53 usable so so this is absolutely one of
54:56 those tools you know I kind of jokingly
54:58 talked earlier about oh all these tools
55:01 are janky pieces of [ __ ]
55:04 um they are they're all janky pieces of
55:06 [ __ ] right now because all this stuff is
55:08 so new so the art of playing with them
55:11 right now is kind of learn where their
55:14 boundaries are what are they good at
55:15 what are they not good at right now and
55:17 then figure out ways to work around that
55:21 and don't get too married like if you
55:23 find the perfect way to make Runway do
55:26 in you know videos better
55:28 um
55:29 don't get too married to that because
55:31 within months they're going to come out
55:33 with a new version that's probably going
55:35 to solve all the problems that you've
55:36 created fancy workarounds for in fact
55:39 one of the things that the the runway ml
55:42 people were doing
55:44 um that were making some of those longer
55:45 form movies is that they would
55:48 they would they would render a video
55:51 then they would take the last frame of
55:53 the video use that as the start frame of
55:55 a new video so they they were
55:57 artificially extending the length of the
55:59 videos by doing that and then Runway
56:02 just came out with now you can extend
56:04 these up to
56:06 um up to 18 seconds so so all of those
56:09 workarounds that people created uh
56:11 Runway just launched a new version and
56:13 it did that
56:15 so let's see what it looks like
56:18 oh that's really good that's really good
56:23 that's way more coherent than
56:28 I've seen before now what I would do is
56:31 I would actually start this at the end I
56:33 would start it where it's bright and
56:35 then end it where it's more Sunset right
56:38 so I would sort of I would flip this in
56:39 post if I wanted to use it
56:42 because this seems like it's getting
56:44 earlier it's it's like the uh
56:47 I guess it could be sunrise couldn't it
56:50 yeah this could be sunrise
56:52 um but I took it at night but that's
56:54 there you go that's pretty good and then
56:56 you can extend it for four seconds and
56:58 extend it again
56:59 and so
57:01 from an image so that was from a
57:02 photograph I took but that could just as
57:05 easily have been mid-journey
57:07 so that's that wild huh isn't it crazy
57:09 isn't it crazy it's crazy it's crazy
57:11 crazy crazy crazy crazy crazy crazy
57:21 all right let's see
57:23 Share Bear hello Irregulars good day to
57:27 you good day do they milk the finances
57:30 wait this could be so brilliant for
57:32 b-roll yeah exactly I mean so so think
57:36 about this
57:37 okay okay yeah big future of work thing
57:40 I don't know if and funding fundraising
57:43 is in here but she wanted me to talk
57:44 about future work future of work stuff
57:50 18 months ago the image generation tools
57:54 were so bad that they were laughable
57:57 stable diffusion mid-journey
58:00 you'd say make me a horse and it would
58:02 make you some sort of four-legged thing
58:04 that looked like it was a genetic
58:05 mutation of something from another
58:08 planet that kind of resembled the horse
58:12 um
58:13 within 18 months mid-journey now is
58:16 doing you know full photographic realism
58:18 for the most part they fixed the hand
58:20 problem
58:22 um they haven't quite dealt with the
58:25 with with letters and words but that's
58:27 coming in Dolly three
58:29 if I need an image for a presentation or
58:34 for a concept
58:36 I historically would have gone to stock
58:38 photo houses I'd have gone to Google I'd
58:41 have gone to you know I'd go to whatever
58:43 the resources are that I go to
58:46 and I would have spent
58:49 hours
58:51 trying to find an image that was
58:54 acceptable for the thing I wanted and
58:56 then if I actually wanted to get
58:59 something that was really close to what
59:01 I wanted I would have to go do a photo
59:03 shoot
59:04 right I have to hire a freelancer
59:07 now
59:10 if I'm facile with mid-journey or stable
59:13 diffusion or adobe Firefly I go in and I
59:16 type exactly what I want and it gives me
59:18 exactly what I want and if you see
59:20 what's coming from Dali it's going to
59:22 give you exactly what you want because
59:24 it can take all of the context of a big
59:26 long conversation in chat GPT and
59:29 incorporate that into an image with text
59:32 that's coming in like a week or two
59:35 and then you look at Runway ML and
59:37 you've got the same exact thing if you
59:39 wanted to comp up to mock up a video of
59:42 hey here's here's the the ad concept
59:45 that we've got and you want to put
59:46 together a bunch of pieces of b-roll and
59:48 you know record a nice voice over and go
59:51 pitch that to a client
59:53 you would have had to go pay for Adobe
59:55 stock photo you know stock video
59:58 it wouldn't have been close to what you
59:59 want you'd have to you'd have to spend
1:00:01 hours looking for the right thing and
1:00:03 then tweaking it and color grading it
1:00:05 and all that sort of stuff
1:00:07 and now you can just ask for it
1:00:10 you can take the images from a
1:00:13 storyboard and use them as the start
1:00:14 frames of Runway ml videos
1:00:17 I told a bunch of fellow students at fit
1:00:20 30 years ago that this day would come
1:00:23 Winston Grant you are correct so
1:00:27 um
1:00:29 um I was 30 years ago oh share bear
1:00:32 thank you that's my favorite of the uh
1:00:34 of those thingies the dangly lights I
1:00:35 love that
1:00:37 um
1:00:38 I was I I was a co-founder of agency.com
1:00:40 back in the early days of of the World
1:00:43 Wide Web so I was I I I loved that
1:00:46 neighborhood where fat was but um
1:00:49 but yeah yeah this the day's here the
1:00:52 day's here it's like like we knew in the
1:00:55 early days of the web that this stuff
1:00:56 was that that that some version of this
1:00:59 was coming right but the tools of of the
1:01:02 early web were just they were so
1:01:04 primitive right it was literally like
1:01:06 click on a hyperlink and it will take
1:01:08 you to a document and what you could
1:01:10 kind of Imagine at that point was this
1:01:13 interconnectedness right that that came
1:01:15 to be right instant Communications
1:01:19 um you can you can be anywhere in the
1:01:21 world you know instantly
1:01:24 um
1:01:27 and our ability to connect with one
1:01:29 another is is you know seamless now it's
1:01:31 now just the fabric of society
1:01:35 for the most part
1:01:37 and we kind of knew something like this
1:01:39 was coming but I think all of us
1:01:41 including a lot of people in machine
1:01:43 learning an artificial intelligence
1:01:45 assumed that the stuff that we get to
1:01:48 play with today
1:01:50 that stuff as powerful as chat gpt4
1:01:55 would be
1:01:57 30 years from now 50 years from now
1:01:59 probably not in my lifetime
1:02:03 and it just like
1:02:04 kind of materialized
1:02:07 it's amazing
1:02:09 it's amazing right and it's you know
1:02:11 it's not that it's not that chat gbt is
1:02:15 this new thing it is a new thing but
1:02:17 it's just an interface on a foundation
1:02:20 that's being built for 30 or 40 years
1:02:22 neural networks and machine learning and
1:02:25 Transformers in 2017 and gpt1 and gpt2
1:02:29 and gpt3
1:02:31 and then GPT 3.5 they build this proof
1:02:34 of concept kind of thing called chat GPT
1:02:36 so they have something to show when
1:02:39 Microsoft makes their big announcement
1:02:40 of their investment
1:02:44 and the world lost their mind I'm
1:02:46 spending less time in Adobe stock and
1:02:48 now creating magazine covers for my
1:02:50 business this is cool Lauren right
1:02:56 in mid-journey yeah
1:02:58 it's amazing
1:03:00 it's that yes
1:03:05 I'm curious what kind of magazine what's
1:03:07 what's the business that's really cool
1:03:11 um
1:03:17 do you have parallels
1:03:20 I've installed parallels occasionally in
1:03:23 my life it's just maybe it's easier than
1:03:25 to use than it used to be
1:03:28 um
1:03:29 I have a
1:03:32 I don't want to maintain two operating
1:03:34 systems it's rough enough maintaining
1:03:35 one
1:03:40 if I did install parallels then I could
1:03:42 run The Tick Tock livestream thing and
1:03:44 maybe I could stream this to more places
1:03:46 than just Tick Tock I don't know I'm
1:03:51 hives as they are right now these more
1:03:54 intimate things you know just just just
1:03:57 a 17 000 people hanging out if you
1:04:00 weren't here earlier last night we had
1:04:03 an infiltration of of uh
1:04:07 it was first five and then ten and then
1:04:09 15 and then 17 000 people and then it
1:04:12 dropped down to 10 and then it went back
1:04:14 up to 15 and it was crazy
1:04:16 but it just ended up being some
1:04:17 engagement farming which was interesting
1:04:19 this is cool
1:04:20 but can it make you money oh I don't
1:04:24 know print on demand can it
1:04:27 [Laughter]
1:04:30 if you're wondering how to make money
1:04:32 with chat EBT I have opinions on that
1:04:34 it's one of my side businesses a monthly
1:04:36 publication called the dance calendar oh
1:04:38 super cool that's that's cool
1:04:42 um oh yeah Winston Make It Rain
1:04:47 thank you appreciate that
1:04:50 will there ever be an AI that can turn
1:04:52 your script idea into a TV episode or
1:04:55 movie yes
1:04:57 it exists it's right here
1:05:01 wanna see I've written seven screenplays
1:05:04 I I've these things are [ __ ] amazing
1:05:06 Claude
1:05:08 Claude you can just oh so here let's
1:05:11 just watch
1:05:13 these this is just stupid
1:05:16 [Music]
1:05:17 um
1:05:19 we'll do gpt4 since it's the best
1:05:26 um
1:05:28 [Music]
1:05:32 I have an idea for a screenplay
1:05:35 and I want to make sure that I write the
1:05:38 best screenplay possible
1:05:40 can you recommend the five best story
1:05:43 Frameworks
1:05:45 used by successful screenplays
1:05:49 written by Hollywood a-listers
1:05:57 oh sorry
1:06:01 so
1:06:07 um the 3x structure the hero's journey
1:06:11 little Joseph Campbell action mono myth
1:06:15 I've never I've never heard that phrase
1:06:17 the mono myth that's cool
1:06:21 and there's that save the cat that's
1:06:23 another famous One
1:06:27 the sequence structure divides the
1:06:29 screenplay into eight sequences about
1:06:31 about 10 to 15 minutes each
1:06:34 and then the 5x structure tips when
1:06:37 structuring
1:06:38 while structures are helpful as guides
1:06:41 it's important to let your unique voice
1:06:43 go through okay that's great
1:06:45 um so I'm going to ask
1:06:47 um
1:06:50 all right
1:06:53 what's the advantage of sequence
1:06:55 structure and are there examples of
1:06:58 films that were made where this is the
1:07:00 primary framework
1:07:07 manageability pacing each sequence has
1:07:10 its own tension and resolution which is
1:07:12 cool cool I kind of like that so so
1:07:15 notice What's Happening Here is there's
1:07:18 two things happening as I'm talking with
1:07:20 this I'm narrowing the context of what
1:07:23 what
1:07:24 Chachi PT and I are talking about
1:07:27 so it's understanding more and more what
1:07:29 the nature of the conversation is
1:07:33 um
1:07:35 I'm also getting an education
1:07:37 I'm also getting ideas
1:07:40 um so I might not have thought about
1:07:42 like the sequence structure wasn't
1:07:44 really one that I dug into and heard
1:07:46 about Star Wars is one North by
1:07:47 Northwest Raiders of the Lost Ark
1:07:50 Silence of the Lambs Pulp Fiction
1:07:52 okay so I'm gonna say
1:07:55 this is great
1:07:57 my screenplay idea
1:08:00 is in the style of Pulp Fiction
1:08:03 so that's exciting period can you create
1:08:06 a framework for me that combines the
1:08:09 save the cat
1:08:10 framework with the sequence framework
1:08:13 you just described
1:08:18 [Music]
1:08:20 if you're going for a style similar to
1:08:22 Pulp Fiction you might be interested in
1:08:24 a non-linear narrative or intertwining
1:08:26 storylines the sequence structure can
1:08:28 help by breaking down these individual
1:08:31 story lines while save the cat Canadian
1:08:33 ensuring each sequence for the overall
1:08:35 story is a compelling development of
1:08:37 satisfying Resolution Act One sequence
1:08:39 one sequence two
1:08:43 right so now I've got a framework
1:08:47 and now I can just say
1:08:50 all right
1:08:52 [Applause]
1:08:54 that's terrific can you now give me an
1:08:58 outline based on that framework
1:09:00 about
1:09:03 a movie about a hedgehog named Larry
1:09:08 who gets killed early in the film
1:09:11 and his brother Lenny
1:09:14 seeks revenge
1:09:18 and it turns out
1:09:21 that Lenny's girlfriend Laura
1:09:24 was the real culprit all along period
1:09:29 [Laughter]
1:09:33 I'll show I'll show you the inspiration
1:09:35 for that
1:09:36 the title of this movie revenge of the
1:09:39 quills
1:09:45 um
1:09:46 so so sequence Larry the Hedgehog
1:09:49 enjoying his life in the Woodland
1:09:51 oblivious to the lurking dangers
1:09:53 introduction of Lenny Larry's brother
1:09:55 and Laura Lenny's girlfriend
1:09:57 theme stated the deceptive nature of
1:10:00 appearances and the pursuit of Truth
1:10:03 inciting incidents Larry is mysteriously
1:10:06 killed
1:10:08 Lenny discovers Larry's body and vows to
1:10:11 find the killer Lenny faces an internal
1:10:14 conflict the Woodland creatures debate
1:10:16 whether vengeance is the right path
1:10:18 right
1:10:19 so
1:10:21 why am I why am I starting here there
1:10:24 are probably if you want to
1:10:27 go look for
1:10:29 things that are specifically tied to
1:10:32 screenwriting with AI if you go to
1:10:35 futurepedia.io there are probably tools
1:10:38 at futurepedia.io that are like we'll
1:10:40 write your screenplay for you here's the
1:10:42 problem
1:10:43 all of those tools all of them
1:10:46 are basically what I just did here but
1:10:49 they're pre-prompted so someone like me
1:10:51 wrote a bunch of those prompts stuck
1:10:53 them in a framework and put a button in
1:10:56 front of it so you press that button and
1:10:58 it's actually making calls to open AI or
1:10:59 one of these other large language models
1:11:01 and it's doing what we just did there
1:11:03 it's probably broken it down into a
1:11:05 bunch of steps here's the problem with
1:11:07 it
1:11:11 you may have a very specific way that
1:11:13 you want your thing to work
1:11:18 the way they set up that program
1:11:20 probably isn't that so what ends up
1:11:23 happening is you spend more time working
1:11:25 around the limitations that their
1:11:27 framework that their software
1:11:30 handcuffed you with then you would just
1:11:33 exploring chatgpt
1:11:36 so here's the other thing that I'll say
1:11:38 about chat GPT and and it might not be
1:11:40 the best tool
1:11:42 for long conversations
1:11:46 Claude is better for that
1:11:48 but Chachi BT is is by far the best at
1:11:51 writing so what I would do is I would
1:11:53 chunk up if rather than imagining that
1:11:57 Chachi GPT is just going to write you a
1:12:00 whole screenplay it's not
1:12:02 for a bunch of reasons it can't it's
1:12:05 only got a context window of 4 000
1:12:07 tokens which is about three thousand
1:12:09 words so that's kind of its memory
1:12:12 so very very quickly the early stuff
1:12:16 that you talked about with it it will
1:12:17 forget and it'll kind of lose its mind
1:12:20 Claude is another tool like chat EBT
1:12:24 that won't lose its mind because it's
1:12:27 got a hundred thousand token context
1:12:29 window or about 75 000 words Claude
1:12:33 AI
1:12:35 um
1:12:37 let me see did I
1:12:39 the hell was the movie I got did I get
1:12:42 Godfather
1:12:46 no
1:12:47 did I get Scarface
1:12:55 no
1:12:57 oh what was the movie that I got
1:13:01 um
1:13:05 well anyway it doesn't matter
1:13:08 you can upload
1:13:11 you could upload an entire screenplay
1:13:13 into Claude
1:13:15 so one of the ways that you could
1:13:18 and and
1:13:20 if you don't think that screenwriters do
1:13:22 this you're wrong
1:13:26 because even like Star Wars like
1:13:28 everything's based on everything else
1:13:29 right there's only nine stories or
1:13:31 whatever it is right there's only nine
1:13:32 story types
1:13:35 you could take us an existing screenplay
1:13:38 that's that's in the in the hood of what
1:13:41 you like upload it here tell it to
1:13:44 abstract out the the framework of that
1:13:47 screenplay and then fill it in with your
1:13:50 story
1:13:51 you could also have Claude do an
1:13:54 interview a back and forth interview
1:13:56 with you so let's let's try that
1:13:59 um
1:14:00 so let's see
1:14:05 oops
1:14:10 I want you to act as a screenwriting
1:14:12 coach
1:14:13 like Sid field
1:14:15 and
1:14:22 interview me
1:14:24 over a series of questions
1:14:29 to get enough information about my
1:14:32 screenplay idea so that you can create
1:14:34 an outline with me period
1:14:37 do you understand
1:14:42 yes I understand let's let's begin the
1:14:45 session by discussing your story could
1:14:47 you please provide a brief premise of
1:14:49 your main storyline
1:14:51 my story is about an animated Hedgehog
1:14:54 named Larry
1:14:56 who dies early
1:14:58 in the movie
1:15:00 and his brother Lenny vows to
1:15:05 avenge his death only to realize that
1:15:09 his girlfriend Laura
1:15:11 was the real culprit
1:15:15 Hedgehog love triangles hit hard period
1:15:27 uh okay so so check what it's doing here
1:15:31 it's doing what I hoped it would which
1:15:33 this is not what I want it to do
1:15:35 here are some follow-up questions to
1:15:37 help develop the outline what genre is
1:15:40 it where does the story take place what
1:15:42 are the main characters the antonym
1:15:45 as someone with add the thought of
1:15:48 filling out that [ __ ] set of
1:15:50 questions makes my head hurt right now
1:15:52 like it makes me want to just turn this
1:15:54 thing off and go away but we don't have
1:15:56 to live with this [ __ ] anymore because
1:15:58 this is this is what you do if you get
1:16:00 Sid Field's book he asks all these
1:16:02 [ __ ] questions and then you got to
1:16:03 sit there and go well I don't want to
1:16:05 answer all those questions I just want
1:16:06 my screenplay idea to be brilliant I
1:16:10 just want to start writing this seems
1:16:12 like work okay so how are we gonna fix
1:16:16 that what we're going to do is we're
1:16:17 going to say
1:16:19 while those are great questions I want
1:16:22 this to be an interactive experience
1:16:23 period can you please ask me one
1:16:26 question at a time and don't move on to
1:16:29 the next question until I've answered
1:16:31 the previous and you're clear that
1:16:33 you've gotten all the information you
1:16:35 need for that question period
1:16:41 absolutely the interactive approach is
1:16:44 very helpful for developing a strong
1:16:45 story outline let's go through this one
1:16:47 at a time what genre is the film is it
1:16:49 an animated comedy drama or something
1:16:50 else
1:16:53 I'd like it to have the look and feel of
1:16:56 an animated comedy but the hard-hitting
1:16:59 story of an animated drum like an animal
1:17:02 documentary
1:17:06 see you know what we're gonna do one
1:17:09 night in here we're gonna inadvertently
1:17:11 create something [ __ ] brilliant and
1:17:13 then we're gonna go make a movie and and
1:17:15 everyone's gonna go see our movie
1:17:17 okay that gives me a good sense of tone
1:17:20 and style we're envisioning an animated
1:17:22 film that visually looks like a comedy
1:17:24 but has a weighty narrative of a drama
1:17:25 like it reinterpreted and mirrored that
1:17:28 back to me and it did it well it's good
1:17:30 with some documentary style narration as
1:17:33 well that's an interesting genre blend
1:17:36 right so it understands that what I just
1:17:38 asked for is not a standard thing
1:17:40 moving on where does the story take
1:17:42 place is there a particular setting or
1:17:44 location that's important
1:17:47 the setting is the woods
1:17:50 near Larry's house
1:17:53 which is now abandoned because Larry's
1:17:56 dead
1:18:01 got it the woods near Larry's house
1:18:05 that helps set the scene next question
1:18:08 what are the main characters Laura Lenny
1:18:10 and oh what are the main characters
1:18:13 Larry
1:18:14 Lenny and Laura like what other
1:18:18 personalities backgrounds and
1:18:19 motivations
1:18:24 Larry was a door-to-door grain salesman
1:18:28 while Lenny fancied himself
1:18:32 an entrepreneur
1:18:34 although he never actually started a
1:18:37 company period he only talked about it
1:18:42 Laura Laura was the true ambitious one
1:18:45 in the bunch
1:18:53 and it seems like things just got done
1:18:55 when she was around
1:18:58 so I don't need to keep going on this
1:19:00 it's just kind of a silly exercise but
1:19:03 with Claude because it's got such a long
1:19:05 context memory
1:19:08 you can now just have this thing
1:19:10 interview you right you could also do
1:19:12 this with pi so if Pi Pi is another one
1:19:16 Pi dot a I or the PI app pi
1:19:19 on the App Store
1:19:21 and this one down down there
1:19:25 if you go to the App Store search for pi
1:19:27 personal assistant
1:19:30 and the thing about Pi is it's connected
1:19:32 to the internet so if you wanted to do
1:19:34 research on the internet you could do
1:19:36 that but you can put it into
1:19:38 conversation mode where you just talk to
1:19:39 it
1:19:40 and it listens and then it'll talk back
1:19:42 to you so you can actually have a
1:19:44 conversation like I'm having there but
1:19:45 instead of having to type or or you know
1:19:48 use the the microphone feature on a Mac
1:19:51 you can just have a conversation with pi
1:19:53 and then it remembers all the [ __ ] you
1:19:55 talked about so
1:19:57 and then you know what I would do is you
1:19:59 know out of this I would have it
1:20:01 generate you know a character sheet
1:20:03 really explaining that I would have it
1:20:05 create log lines I would start having it
1:20:08 create dialogue dialogue is going to be
1:20:10 the toughest thing
1:20:12 um because it's just so nuanced it it
1:20:15 you know I would say gpt4 is going to be
1:20:18 way better at dialogue than Claude will
1:20:20 like I would say Claude would be good to
1:20:22 do all your all your framework and your
1:20:25 outline and then I would use gpt4 to do
1:20:28 any any sort of
1:20:30 um writing
1:20:33 um and I would assume that for
1:20:34 screenplay dialogue it will get the
1:20:37 structure right
1:20:39 the scenes will tend to be too short and
1:20:42 they'll tend to be too patent
1:20:44 so I feel like with the outline part you
1:20:47 can probably get it to 90 pretty quickly
1:20:50 with the dialogue part I wouldn't expect
1:20:53 to get it better than sixty percent of
1:20:55 what you want so assume you're going to
1:20:57 be rewriting a lot of the dialogue if
1:20:59 not writing it you know wholesale
1:21:02 um and that's how I do it and and here's
1:21:04 the deal like
1:21:06 there may be a tool out there that does
1:21:08 this really well but I doubt it what you
1:21:11 will learn in the process of figuring
1:21:13 this [ __ ] out for yourself is invaluable
1:21:16 it will serve you in all other parts of
1:21:19 your life so anyway
1:21:21 go to the go go do that
1:21:24 what's the app name uh I don't know what
1:21:27 app I was showing
1:21:30 um did the Mandarian come through yet
1:21:34 um I don't know what the mendarian is
1:21:36 Emilio's wife anyone know the name
1:21:38 what's the name of the site again can
1:21:40 someone name the app oh my God
1:21:42 don't you know any other language well I
1:21:45 do I I oh the Mandarin oh that's what
1:21:48 you're asking Emilio's wife no the
1:21:49 Mandarin I'm only up to like 60
1:21:52 something thousandth in line hang on so
1:21:56 what these two are asking for is they're
1:21:58 tired of they're tired of hearing me
1:21:59 speaking Italian
1:22:01 and I did I went to hey Jen and
1:22:06 um
1:22:09 did a mandarin version of me I'm up to
1:22:12 63 000 in line so
1:22:15 I'm gonna have to I'll I'll probably
1:22:17 I'll probably buy a subscription
1:22:19 tomorrow I don't really want to buy a
1:22:21 subscription but I think I will
1:22:25 because that will accelerate me in line
1:22:36 [Music]
1:22:38 came up with the name
1:22:40 we're not a big Library I didn't want to
1:22:43 take Kyle's name
1:22:45 that's cool
1:22:47 uh
1:22:49 um I'm going to use this to reach out
1:22:53 wait I'm going to use this to reach out
1:22:55 to VP of sales cool yeah I mean these
1:22:59 tools to help you get your thoughts
1:23:01 together to write letters to
1:23:05 um you know one of the things that you
1:23:07 can do Franklin is Franklin Tech SDR one
1:23:11 of the things that you can do with these
1:23:12 tools is go copy
1:23:16 um
1:23:17 copy like the work experience
1:23:20 or what someone has written on their
1:23:22 LinkedIn profile of someone you want to
1:23:24 reach out to pop it into Claude
1:23:27 and then have it write you a
1:23:29 personalized letter to that person
1:23:32 referencing stuff in their work history
1:23:34 about the thing that you want to talk to
1:23:36 them about and it will
1:23:39 figure it out
1:23:40 and it'll write you something that'll
1:23:42 take you know give it a little piece of
1:23:43 writing about your [ __ ] give it some you
1:23:47 know background on that person like can
1:23:49 literally be copy and pasted out of
1:23:51 LinkedIn
1:23:52 and then have it write a very
1:23:53 personalized letter and then if it's too
1:23:56 long say I wanted in three short
1:23:57 paragraphs but keep keep the most
1:23:59 important things that will build the
1:24:01 relationship whatever you might want to
1:24:03 do it's brilliant at that [ __ ] brilliant
1:24:05 at it
1:24:07 um
1:24:08 I tried to use basic logo today with
1:24:11 ideogram and it was bad real bad EA
1:24:14 Source Kemp I don't know if you've heard
1:24:16 this here at the AI learning lab but all
1:24:18 these tools today they're janky pieces
1:24:20 of [ __ ]
1:24:22 anything anything that requires
1:24:26 precision
1:24:28 these diffusion models are not great at
1:24:30 right now it looks like Dolly three
1:24:34 is gonna be good at Precision prompting
1:24:38 and it's it you're gonna it's going to
1:24:41 allow you to use chat gpt4 to write the
1:24:44 prompts for you and it's gonna it's like
1:24:46 I think I think it's gonna be good
1:24:48 whether or not it can do logos well is
1:24:50 another thing
1:24:51 I learned of a thing last night called
1:24:55 um Trace Journey not Vector Journey like
1:24:57 that I called it last night but there's
1:25:00 there's a tool called
1:25:02 Trace Journey
1:25:04 that you
1:25:06 install into Discord
1:25:10 in a channel that you've got mid-journey
1:25:12 in and then when you make a mid-journey
1:25:14 image
1:25:15 Trace Journey you can create vectors out
1:25:18 of it so if you can get mid-journey to
1:25:20 give you a decent logo
1:25:23 and
1:25:25 vectorize it then take it into so mid
1:25:29 Journey won't do text well
1:25:32 but it will do images really well it'll
1:25:35 do logos well and with this
1:25:37 vectorization tool you can vectorize the
1:25:39 logo just pop it into illustrator and do
1:25:41 your own text in illustrator
1:25:43 or
1:25:46 make
1:25:49 a janky shitty version of the logo in
1:25:52 ideogram that's got the typography that
1:25:55 you want use that as a source image in
1:25:57 mid-journey
1:25:59 generate the image based on that Source
1:26:02 image then vectorize it then take it
1:26:04 into illustrator and redo all the font
1:26:06 but you'll at least be following a
1:26:08 design then
1:26:10 um that's how I would do that
1:26:12 um we'll we'll see how good Dolly three
1:26:14 is at
1:26:16 um
1:26:17 at logos and stuff
1:26:20 the the minute see
1:26:26 without doing licensed deals with like
1:26:30 the NFL like if you want NFL logos
1:26:33 without doing license deals with the NFL
1:26:35 teams
1:26:36 I don't think any of these
1:26:39 um image generation tools are
1:26:42 I don't think they're gonna work that
1:26:43 hard
1:26:44 at doing logos of existing companies
1:26:47 well because there's too much liability
1:26:49 there but we'll see I mean at some point
1:26:51 someone's going to cut deals licensing
1:26:53 deals with with big IP holders so that
1:26:57 you can do something that's like
1:26:58 mid-journey but you want a Denver
1:26:59 Broncos logo on it or you want the
1:27:01 Colorado Buffaloes on it you can do that
1:27:06 so yep janky pieces of [ __ ] right now
1:27:09 just do the best you can work around be
1:27:12 patient understand
1:27:14 that
1:27:16 these tools are going to change
1:27:18 everything and it's all brand new it's
1:27:21 all brand new
1:27:23 stability uh AI unveiled a model called
1:27:27 stable audio for controlled audio
1:27:28 generation yeah they did the the the the
1:27:32 um
1:27:33 all the audio things that I've seen
1:27:35 right now are
1:27:37 they're kind of like do you remember
1:27:39 when when digital audio first came out
1:27:42 and and all the song like you you could
1:27:45 hear songs in digital
1:27:47 and then they were also over compressed
1:27:49 they all had these compression artifacts
1:27:51 and they were just bad for a long time
1:27:52 that's kind of where all these uh all
1:27:55 these audio things are including stable
1:27:57 audio but we can go we can go look at it
1:27:59 and play with it
1:28:01 uh
1:28:03 last time I tried to play with this it
1:28:05 was it was over busy it would not let me
1:28:08 play
1:28:09 all right let's see
1:28:12 um
1:28:24 let's see
1:28:34 melancholic
1:28:36 Hedgehog
1:28:39 trance music
1:28:43 with a thumping Jungle Beat
1:28:52 so so this is text to audio and they're
1:28:55 doing this is this is diffusion so this
1:28:57 is an almost identical technology to how
1:29:01 stable the fusion generates images
1:29:03 through the diffusion model so it's
1:29:05 using the same basic technology that's
1:29:07 one of the remarkable things about
1:29:10 these tools is that they work on any
1:29:14 kind of data they're kind of data
1:29:16 agnostic in some sense that basically if
1:29:19 you train them on a certain data set
1:29:21 then it'll it'll find patterns in that
1:29:23 data set and if you put it in another
1:29:25 one it'll find patterns in that data set
1:29:27 is this ready oh yeah
1:29:30 no
1:29:33 [Music]
1:29:43 [Music]
1:29:47 I don't know that that's melancholic or
1:29:50 Trancy I guess it's Trancy it's not
1:29:53 melancholic
1:29:54 that's pretty awful okay here's another
1:29:57 one I did called Moody finger style
1:29:59 guitar
1:30:02 [Music]
1:30:10 okay the compression kind of artifacts
1:30:12 was like wow wow you get that swelling
1:30:14 wow
1:30:18 [Music]
1:30:20 but again here's the thing even though
1:30:27 they're that
1:30:31 this is as bad as it will ever be
1:30:35 and if you go back 18 months to what
1:30:38 stable diffusion was doing with images
1:30:40 this is the audio equivalent of that
1:30:42 it's it's it's decent it's not bad
1:30:46 but it's kind of like in the
1:30:49 neighborhood of Music 18 months from now
1:30:52 this will be full on
1:30:55 orchestrated layered no no compression
1:30:58 artifact music probably with coherent
1:31:02 lyrics and singing and harmony
1:31:09 seriously seriously people listen I'm
1:31:13 very disappointed there's only 73 people
1:31:15 in here last night we had 17 000 people
1:31:18 in here tonight it's like 70 what's
1:31:20 going on people
1:31:21 have you not told your friends last
1:31:23 night you told your friends and we had
1:31:24 17 000 people in here do you understand
1:31:26 what I'm saying I don't think you do
1:31:28 I don't think you understand the gravity
1:31:30 of the situation
1:31:32 um I decided that I was a a celebrity
1:31:35 last night and now all of a sudden it's
1:31:37 not looking so good okay
1:31:40 it's a little awkward it's a little
1:31:42 awkward okay so I don't know what you're
1:31:43 doing
1:31:44 but I've got responsibilities you've got
1:31:47 responsibilities and uh where are my 17
1:31:50 000 peeps
1:31:51 they're not here are they oh we're not
1:31:53 special in that special
1:31:56 I wonder who could have brought those 10
1:31:58 000 people who could that have been
1:31:59 maybe Satan
1:32:00 that was actually James
1:32:03 all right and my ten thousandth caller
1:32:07 exactly we're gonna give away a Toyota
1:32:09 in the ten thousandth caller
1:32:13 this Saturday call in
1:32:17 wesn radio hard rock and hip hop
1:32:20 we're giving away the Country Store
1:32:22 you're going to want to come
1:32:24 all right see you at the county fair
1:32:27 Saturday
1:32:29 [Laughter]
1:32:35 my ha moment I made a get well card for
1:32:38 a friend that had a hip replacement GPT
1:32:41 nice of you to make a card GPT made it
1:32:45 amazing oh that's awesome
1:32:47 that that was your uh
1:32:50 Kevin McAllister moment oh by the way so
1:32:52 look at the article that I wrote this is
1:32:54 this is a nod to all of you
1:32:57 uh I'm gonna go to LinkedIn
1:33:02 LinkedIn bam
1:33:12 so this is the article that I wrote and
1:33:14 this is a cool this was cool I did this
1:33:16 in mid-journey
1:33:17 and then
1:33:18 it was
1:33:20 um like these these two dudes were there
1:33:23 and then those people were there but
1:33:25 this was just kind of this empty space
1:33:27 but the the desk was there but there was
1:33:29 no person so I went into Photoshop
1:33:31 generative Phil and I selected that area
1:33:33 and I put a woman sitting at a desk with
1:33:35 a computer screen and so that was kind
1:33:38 of cool so that was that but then I I
1:33:40 talk about
1:33:41 um
1:33:42 uh every Everything used to be about uh
1:33:46 productivity and AI is gonna help with
1:33:48 productivity but then I talk about my
1:33:51 Kevin McAllister moment and look
1:33:57 laughs
1:34:01 [Laughter]
1:34:03 uh and then I the the caption is what it
1:34:07 feels like every day working with
1:34:09 generative AI
1:34:13 so so if you want to read this article
1:34:15 I'm actually really happy with this
1:34:16 article the basic idea that that
1:34:19 one of the things I talk about a lot in
1:34:22 here is just go play with AI just go
1:34:24 play with it
1:34:25 and I think people are a little weirded
1:34:27 out by that like well oh thank you very
1:34:29 much you know I think I I have to learn
1:34:31 it it's a technology it's a tool it's AI
1:34:33 it's it's this important thing and I
1:34:35 gotta no no just go play
1:34:37 just go play because everything that you
1:34:40 learn will be some augmentation to who
1:34:44 you are and some amplification of who
1:34:46 you are and it's and it will feed your
1:34:49 work in non-intuitive weird ways
1:34:52 this stuff is
1:34:55 I it's we're just we're we're entering a
1:34:58 weirdness
1:35:00 and so
1:35:03 by by enter by by engaging with AI with
1:35:08 a playful curiosity
1:35:11 um you won't take it too seriously
1:35:12 because it's gonna get surreal very very
1:35:14 fast
1:35:15 and the more Nimble you can be with sort
1:35:18 of bouncing between different tools and
1:35:20 learning how to use them and playing
1:35:22 like joyfully between these different
1:35:25 things and kind of forgiving the
1:35:27 imperfections of it
1:35:29 um it it will serve you so that's what
1:35:31 that article is about so
1:35:35 um
1:35:38 there's a nice little balloon that's a
1:35:41 little balloon
1:35:42 oh you know what I'll show you while I
1:35:44 have the uh well I have the screen up
1:35:46 here
1:35:47 we a couple of folks were asking about I
1:35:50 don't get the Larry the Hedgehog
1:35:52 reference
1:35:53 um
1:35:55 open AI so if you go to openai.com
1:35:59 dolly2 or doll
1:36:02 d-a-l-l Dash E-3
1:36:07 Dolly three D A openai.com
1:36:14 d-a-l-l-e-3 that'll take you to this
1:36:17 page
1:36:18 so this was announced three days ago
1:36:20 it's just going to show up if you're a
1:36:23 chat GPT plus member if you're paying
1:36:25 the 20 bucks a month you're gonna get
1:36:26 access to this first
1:36:28 this will also show up in Bing's image
1:36:31 generation so right now if you go to
1:36:33 Bing and you say make me an image of a
1:36:35 hedgehog it'll do it it'll be fine but
1:36:37 that's Dolly too and Dolly 2 sucks this
1:36:40 is not good you know Bing continues to
1:36:43 disappoint and Dolly too is the is their
1:36:46 image generation engine but it's gonna
1:36:48 have this in it this looks remarkable so
1:36:51 a couple of things about this that I'll
1:36:53 show you right now and then I'll show
1:36:54 you Larry the Hedgehog at the bottom
1:36:56 um which is where that reference came
1:36:58 from so
1:37:01 so one of the things that that this
1:37:03 thing does well
1:37:05 is
1:37:06 details
1:37:08 so see all that stuff okay
1:37:12 and now I'm going to read you the prompt
1:37:15 an illustration of an avocado sitting in
1:37:18 a therapist's chair check
1:37:21 saying I just feel so empty inside check
1:37:25 it did that text the text is perfect it
1:37:27 doesn't look weird
1:37:29 it's like well aligned
1:37:32 with a pit sized hole in its Center so
1:37:36 if you were just drawing an avocado the
1:37:39 it wouldn't be a hole it would be the
1:37:40 the pit would be there right no there's
1:37:43 a pit sized hole done check
1:37:46 the therapist is a spoon check scribbles
1:37:51 notes check
1:37:52 so
1:37:54 Dolly three is really good at taking
1:37:56 instructions and detailed instructions
1:37:59 and incorporating them all this one's
1:38:02 wild a 2d animation of folk music band
1:38:05 composed of anthropomorphic Autumn
1:38:08 Leaves each playing traditional
1:38:10 bluegrass instruments those are all
1:38:12 traditional bluegrass instruments
1:38:14 amid a rustic Forest setting dappled
1:38:18 with the soft light of a harvest moon
1:38:21 so yep
1:38:24 it did that
1:38:26 so this is what's coming and this is not
1:38:28 coming like a year from now this coming
1:38:30 like I don't know a week or two
1:38:32 this is another one that's just gorgeous
1:38:35 let me zoom in on her
1:38:37 so she's a dancer
1:38:39 look at the dress like all this sort of
1:38:41 porcelain looking stuff
1:38:44 and then look at the the um
1:38:48 The Prompt here is insane in front of a
1:38:51 deep black black backdrop
1:38:53 a figure of middle years her Tongan skin
1:38:57 rich and glowing is captured mid-twirl
1:38:59 her curly hair flowing like a storm
1:39:02 behind her her attire resembles a
1:39:05 whirlwind of marble and Porcelain
1:39:06 fragments illuminated by The Gleam and
1:39:08 of scattered porcelain shards creating a
1:39:12 dream-like atmosphere the dancer manages
1:39:14 to appear fragmented yet maintains a
1:39:17 harmonious fluid form holy [ __ ]
1:39:21 it's just like just amazing
1:39:24 um so that's that stuff
1:39:27 um
1:39:28 this is one that shows again the details
1:39:31 of the image that there's all these
1:39:32 details in the prompt that are all
1:39:34 captured and it kind of points those out
1:39:37 um this was Dolly too
1:39:39 and this is Dolly three and the prompt
1:39:42 is an expressive oil painting of a
1:39:45 basketball player dunking depicted as an
1:39:48 explosion of a nebula
1:39:50 so that looks like an explosion of a
1:39:53 nebula he's definitely dunking this
1:39:55 looks like a blob and
1:39:59 he's he's about as far from the rim as
1:40:02 you can be maybe he half dunked
1:40:05 but you know he is not dunking he is
1:40:08 dunking right so the what they call the
1:40:11 coherence here is really good
1:40:14 um oh it's going to be native Within
1:40:16 gpt4
1:40:18 and gpt4 will help you be able to do the
1:40:21 prompting and then this was this is
1:40:23 where Larry the Hedgehog came from and
1:40:24 this is why we were joking the other
1:40:27 night and someone said something and
1:40:30 then
1:40:31 on the on the Discord in the irregular's
1:40:34 channel someone made an image of of a
1:40:37 hedgehog by
1:40:39 um a cemetery gravestone that said Larry
1:40:43 and so I said it was Lenny at Larry's
1:40:46 grave and so that started the whole
1:40:48 thing where Larry's dead and Lenny's his
1:40:51 brother and Laura and then we came up
1:40:53 with a whole thing where we could do a
1:40:55 whole story where all of the words begin
1:40:57 with the letter L and it was just all
1:40:59 that but what check this out
1:41:02 so one of the things if you've played
1:41:04 with mid Journey or any of these things
1:41:05 one of the things that's very
1:41:07 frustrating and it's very hard to do you
1:41:09 can do it in some tools sometimes
1:41:13 is if you create a character that you
1:41:15 like
1:41:16 replicating that character across
1:41:19 prompts is effectively impossible it
1:41:22 sucks really bad so watch this video and
1:41:24 I'll pause it here in a second to show
1:41:26 you something
1:41:32 [Music]
1:41:40 my daughter says its name is Larry can I
1:41:44 see more like this right so it
1:41:46 understands what this is
1:41:48 it understands
1:41:49 its name is Larry so it understands that
1:41:52 the Hedgehog that I picked out of that
1:41:55 lineup of four is the one I'm talking
1:41:57 about and its name is Larry and so watch
1:42:00 the interaction from this point forward
1:42:02 [Music]
1:42:25 thank you
1:42:36 thank you
1:42:37 [Music]
1:42:39 right so that's where the Larry
1:42:41 reference came from and that's coming
1:42:42 that's that's coming I don't know two
1:42:45 weeks maybe
1:42:46 maybe tonight who knows I don't know I
1:42:50 do not knew I do not knew I asked for
1:42:54 the email addresses from a business
1:42:56 marketing list I have on my drive it
1:42:58 constantly Rudy
1:43:00 gave me the website Okay so
1:43:04 Rudy you asked for the email addresses
1:43:08 from a business marketing list I have on
1:43:10 my hard drive
1:43:11 Chachi BT can't see in your hard drive
1:43:14 bard
1:43:18 as of three days ago can
1:43:21 [Laughter]
1:43:23 but
1:43:25 it's got to be in Google Drive so
1:43:29 an important thing to understand
1:43:35 what it's been trained on is they took
1:43:38 as much of the public internet as they
1:43:39 could and they started training it on
1:43:42 all this data as much as they could as
1:43:43 much of the public you know public
1:43:45 databases and Wikipedia and just
1:43:48 websites and blogs and just as much crap
1:43:51 as they could put in there
1:43:53 and they stopped training chat gbt in
1:43:56 particular September of 2021 because
1:43:59 embedding all of this data into this
1:44:02 magical mythical GPT data or vector
1:44:06 space
1:44:07 is really time and GPU intensive
1:44:12 so they it's not like this is an ongoing
1:44:15 thing
1:44:16 so when you're interacting with chat GPT
1:44:18 it's not looking at your heart it's not
1:44:20 a it's not a normal database query it's
1:44:22 not a Google query
1:44:24 it is pulling from this lexicon of words
1:44:28 that are clustered by meaning
1:44:32 in some sort of mathematical space and
1:44:34 it and it's it's just answering what it
1:44:37 thinks from a probabilistic standpoint
1:44:39 you want based on what it knows so it
1:44:43 doesn't know your hard drive it doesn't
1:44:44 know anything after September 2021
1:44:48 things like Bing and Bard and Pi that
1:44:51 are connected to the internet also have
1:44:54 a similar limitation but because they're
1:44:56 connected to the internet they can
1:44:58 search for more recent things
1:45:00 so Bard can connect to the internet as
1:45:03 of three days ago there's a thing
1:45:05 embarred so long as you're using an
1:45:08 individual the
1:45:10 you will not get this anywhere else
1:45:12 because I couldn't get the [ __ ]
1:45:14 answer to this so this is a big one this
1:45:16 is this is the kind of value we deliver
1:45:17 here at the AI learning lab it's a solid
1:45:22 you can only use what I'm about to tell
1:45:24 you if you do not have a Google
1:45:26 workspace account meaning if your work
1:45:30 pays for Google
1:45:33 you can't use this feature which is
1:45:35 really counterintuitive and it pisses me
1:45:37 off if I'm paying for the thing I should
1:45:39 get priority but that's not how it works
1:45:41 at Google
1:45:43 so it has to be an individual Google
1:45:45 account and then if you go to
1:45:47 bard.google.com extensions
1:45:50 it will take you to this thing called
1:45:52 extensions one of the extensions is
1:45:54 Google Drive
1:45:56 so you can click on Google Drive
1:45:59 and then when you interact with it you
1:46:01 type at drive and then your prompt and
1:46:04 it will actually go search your Google
1:46:06 Drive
1:46:07 that's a three day old feature for a
1:46:10 small subset of Google users it might be
1:46:13 a big subset of Google users but I have
1:46:15 to use an old account where I have I
1:46:17 have none of my current stuff in that
1:46:20 version of drive so it's kind of useless
1:46:22 to me so I'm a little cranky about that
1:46:24 particular business decision I
1:46:27 understand it because I would assume
1:46:30 that
1:46:30 um dealing with the permissions across
1:46:32 an organization is probably a nightmare
1:46:34 with this chat GPT [ __ ] so so there's
1:46:38 that
1:46:40 um it can also search Google travel or
1:46:43 Google flights Google hotels Gmail
1:46:46 drive and Docs
1:46:50 I think that's it there's five of them
1:46:53 but that's where all this stuff's going
1:46:54 these tools are going to be able to
1:46:57 um
1:46:58 you're gonna be able to search your own
1:46:59 data which is going to be amazing
1:47:01 Apple's going to do this really well by
1:47:04 the way
1:47:07 [Music]
1:47:10 if it gave you the website that's
1:47:12 probably because it's a company that it
1:47:14 understands how to website in its data
1:47:16 set it does not it cannot see your hard
1:47:18 drive
1:47:20 I see we're back down to manageable
1:47:22 levels we're up to 71 people I think
1:47:24 that's respectable it ain't seventeen
1:47:26 thousand
1:47:28 but you know
1:47:29 but I mean here's the deal all those
1:47:31 [ __ ] 17
1:47:32 000 we we have more likes tonight than
1:47:35 we had last night with 17 000 people in
1:47:37 here the average view time of a
1:47:40 non-follower of the channel was four
1:47:43 seconds
1:47:44 the average view time of a follower of
1:47:46 the channel was like 12 minutes or 14
1:47:49 minutes or it was like it was long
1:47:52 um I missed tonight's live unless you're
1:47:54 still there still there till there
1:48:00 um jump in
1:48:02 uh last night viewer screenshot 17 000
1:48:06 12 21 a.m eight point
1:48:09 seven thousand twelve twenty four a.m 14
1:48:12 000 12 27 a.m yeah I know it was crazy
1:48:16 well it was you know what was funny is
1:48:18 like we were just going along like this
1:48:20 and I I you know I normally look up and
1:48:22 I'm like you know is is is the thing
1:48:24 tapering like is it is it time to wind
1:48:26 it down is is essentially what I'm
1:48:28 thinking so we kind of started at I
1:48:31 don't know 40 and then it went to 60 and
1:48:33 then it went to 75 or 80 and then the
1:48:36 next time it looked up it was like it
1:48:37 said four and I was like huh and it was
1:48:40 4.6 K and I was like wait is that
1:48:43 thousand I was like right
1:48:46 and you saw if you were here last night
1:48:48 it totally threw me so and then it was
1:48:52 just like going
1:48:53 it was crazy
1:48:57 uh Community Theater is about to get lit
1:49:02 um I don't know what that's in reference
1:49:03 to but I I agree I think these tools are
1:49:06 gonna help help all the creative arts
1:49:08 hey Champion what's up
1:49:11 you gonna go out
1:49:12 [Applause]
1:49:26 sometimes a dog's gotta do what a dog's
1:49:29 gotta do you know what I mean
1:49:32 all right
1:49:33 yes can I play DJ in the movie yes you
1:49:37 can yeah yeah we'll oh the the community
1:49:39 theater thing was was uh was based on
1:49:44 our little creative writing thing yeah
1:49:45 exactly uh everyone can play everyone
1:49:47 you know we'll put on a show it'll just
1:49:49 be like that uh that old and Tiny movie
1:49:51 I got a barn I got costumes let's put on
1:49:55 a show
1:49:59 so I just jumped on what happened last
1:50:02 night with all the viewers so so it was
1:50:05 um there's this guy James
1:50:08 um
1:50:09 I can't remember his last name because
1:50:12 I'm a loser and have ADD yeah
1:50:16 James
1:50:18 kamua
1:50:20 James kamua from
1:50:23 um
1:50:25 Kenya and he's been in he's been in
1:50:28 before he's a fan of the channel and and
1:50:30 James James last night like the numbers
1:50:33 went crazy I'm like what the hell's
1:50:34 going on and James like I did that so I
1:50:38 think James I think James dumped dumped
1:50:40 a little coin to do some engagement
1:50:42 farming and I guess there are places
1:50:45 where you can just have people log into
1:50:47 your Tick Tock lives at scale
1:50:50 um they didn't stay very long and they
1:50:52 didn't really engage but it was it was
1:50:54 exciting while it happened it was pretty
1:50:55 cool
1:50:56 how is that different from the current
1:50:58 search in Google Drive
1:51:01 um that I don't know Karen I mean
1:51:06 because it's Google
1:51:13 here's here Karen you bring up a good
1:51:17 point here's my strong instinct
1:51:19 my strong instinct
1:51:24 hang on I'm thirsty my throat just
1:51:26 closed up
1:51:30 my strong Instinct and this could be
1:51:32 wrong I'm pulling this out of my ass
1:51:34 if Pete's in here he might know Pate
1:51:36 works for Google
1:51:38 um my strong instinct
1:51:41 is that what they're probably doing with
1:51:43 these extensions right now
1:51:45 is
1:51:47 um doing some form of a Google Search
1:51:50 and then taking that Google search and
1:51:52 tossing it into the large language model
1:51:54 to put it in plain English so rather
1:51:57 than just giving you a list of links
1:51:59 it's probably incorporating some of what
1:52:01 it found with a Google search
1:52:04 um and the only reason I say that is the
1:52:06 the couple of times that I tried to play
1:52:09 with it it was about as effective at
1:52:11 finding
1:52:13 obscure content as as a Google search
1:52:15 which sometimes Google's just not good
1:52:17 at finding [ __ ] on especially in Drive
1:52:21 um so I kind of had I had a similar
1:52:23 experience so my my guess is that they
1:52:26 might be doing a semantic search as the
1:52:28 first pass and then incorporating those
1:52:30 results into a large language model
1:52:32 query and then giving you something that
1:52:34 looks like a you know a GPT like
1:52:37 response
1:52:39 I don't know what this Gemini thing is
1:52:43 all about but I have a feeling Gemini
1:52:45 has been trained on a lot I have a
1:52:48 feeling it's been trained on YouTube I
1:52:50 have a feeling it's been trained on all
1:52:51 the Google searches I have a feeling
1:52:54 in some way or other it's probably been
1:52:56 trained on
1:52:57 some amount of Drive data and docs data
1:53:01 and sheets data of course they would
1:53:04 never steal our data definitely not they
1:53:06 would never do that
1:53:07 all of your data is safe with them
1:53:09 they're definitely not selling it
1:53:12 um
1:53:16 pay attention to Google because a couple
1:53:19 of things about Google they invented the
1:53:21 Transformer they invented the core
1:53:23 technology that makes all of these tools
1:53:26 that we're talking about on this channel
1:53:28 possible
1:53:30 they kind of backed off it because I
1:53:32 think they knew it might threaten their
1:53:33 business model and and so they kind of
1:53:35 sat back and then when chatgpt came out
1:53:37 and was a big hit Google went into Red
1:53:40 Alert mode they brought back Sergey and
1:53:42 Larry and said help
1:53:46 uh we kind of [ __ ] up
1:53:49 so for six months of this year Bard was
1:53:53 laughably bad and then Bard got good
1:53:56 it's not great but Bard right now is
1:54:00 in the neighborhood of GPT 3.5 it's
1:54:03 connected to the internet
1:54:05 and you can tell it's got some solid
1:54:08 technology behind it Gemini they're
1:54:11 claiming is that Gemini is their new
1:54:13 underlying large language model their
1:54:16 claiming is five to I've heard two
1:54:18 different numbers five times better and
1:54:20 20 times better than gpt4
1:54:24 I doubt that very seriously but even if
1:54:28 it's as good as gpt4 that's a big deal
1:54:33 so
1:54:35 the the tools that are rolled into the
1:54:39 office suite right now like Google Docs
1:54:41 and Google Sheets and things like that
1:54:42 are very rudimentary
1:54:44 they're very weak
1:54:46 the extensions
1:54:49 um for Bard right now
1:54:51 are kind of you know in this janky
1:54:53 pieces of [ __ ] territory pay attention
1:54:56 to Google I have a feeling it's going to
1:54:58 get very good very fast from them
1:55:00 because they are throwing much money at
1:55:04 it and many engineers at it and they
1:55:06 invented this [ __ ] so they they and and
1:55:08 they've got
1:55:09 plenty of gpus to be training this stuff
1:55:13 as does meta
1:55:15 as does Microsoft and open AI as does
1:55:17 anthropic as does inflection
1:55:21 so there's some cool [ __ ] coming but I I
1:55:23 have a feeling we're gonna see we're
1:55:25 gonna see dramatic improvements in in
1:55:28 Google's technology by the end of the
1:55:30 year
1:55:33 so pay close attention how is Lenny
1:55:35 killed was his drink spiked oh Stevo
1:55:39 with the dad humor oh that's solid
1:55:42 that's have you seen the Australian
1:55:44 people that that tell each other dad
1:55:47 jokes and then they have to take a shot
1:55:48 if they laugh at them that that one
1:55:50 would get someone it's solid it's solid
1:55:52 it's good
1:55:54 somehow my Wi-Fi constantly disconnects
1:55:57 so that's why I have to reconnect oh
1:55:58 Omar that sucks Chinese Mandarin that's
1:56:01 uh oh I'm way behind on these I'm gonna
1:56:04 jump down to the bottom here
1:56:07 I'm I'm very far behind on my questions
1:56:09 so people in here are probably like why
1:56:12 isn't he answering my question you know
1:56:14 I come in here
1:56:16 um it seems like there's some sort of
1:56:17 interactive component to this learning
1:56:20 lab first of all is he a scientist it's
1:56:22 called a lab where's his white coat why
1:56:24 is it like this he seems to be a bit
1:56:27 smart alecky I don't even know if I like
1:56:30 that it's a little tacky and he he
1:56:32 occasionally will even curse
1:56:35 this seems like a wholesome Endeavor and
1:56:38 and then he says the naughty words
1:56:41 and he's talking about killing hedgehogs
1:56:44 what have hedgehogs done to him I feel
1:56:46 triggered
1:56:47 what was the V word you were using
1:56:49 earlier
1:56:53 um
1:56:59 very bad place
1:57:05 I don't think I was using that v-word I
1:57:08 do not know I do not know bubbles what
1:57:10 my V word was uh vernacular uh
1:57:14 anyone V is for L vector vector yes yes
1:57:19 the V word vector why would you assume
1:57:22 it would be something other than Vector
1:57:23 you naughty naughty people good Lord
1:57:27 I'm not rad this is It's a wave of heat
1:57:30 just flew through the just listen leave
1:57:34 me alone people
1:57:36 stop picking on me
1:57:42 oh man
1:57:44 oh we're having a good time tonight the
1:57:46 V word Vector yes the vector monologues
1:57:49 exactly
1:57:52 um
1:57:54 yeah so so what I was talking about was
1:57:57 Trace Journey which is the tool that
1:58:00 will vectorize mid-journey images now
1:58:02 here's here's another thing that pisses
1:58:05 me off
1:58:06 um
1:58:09 Trace Journey which vectorized
1:58:11 vectorizes mid Journey images sounds
1:58:14 super cool so after I left the live last
1:58:17 night and sort of came down from my 17
1:58:20 000 person high and realized oh they
1:58:23 really weren't people
1:58:25 um
1:58:26 I went to trace journey I installed it
1:58:29 on the the
1:58:31 um what's it called the AI Salon Discord
1:58:34 in our mid-journey channel so it's in
1:58:36 there right now if you go to if you go
1:58:37 to the Discord Channel
1:58:39 um this the salon Discord channel for
1:58:42 mid-journey play
1:58:44 Trace Journey's in there but it's got no
1:58:47 free tier for vectorizing images you
1:58:49 have to pay for it it's like a 10 Buck
1:58:50 minimum but I'm like like
1:58:54 give me two like give me two for free so
1:58:57 I can at least see if it's any [ __ ]
1:58:58 good I'm so sick of these software
1:59:01 companies that like have you go through
1:59:03 a full [ __ ] installation
1:59:06 and then and then in the case of Discord
1:59:09 you have to actually add their bot to
1:59:11 your channel and give it permissions
1:59:13 like it was it was a whole [ __ ] thing
1:59:14 and then I go to make my first vector
1:59:17 image and it's like well you're not a
1:59:18 subscriber
1:59:19 I'm like well how about go [ __ ] yourself
1:59:21 give me two
1:59:24 even hey Jen gave me two free
1:59:25 translations the two Italian ones that
1:59:28 I've got that's why I only speak Italian
1:59:35 oh man all right what'd Richard Punk
1:59:38 have to say I've done a comparison oh
1:59:40 yeah that's right
1:59:44 okay cool
1:59:49 awesome awesome awesome
1:59:52 I'm in me too yours oh wait
1:59:55 oh that's cool dump the dolly three
1:59:57 prompt into ideogram oh that's cute
2:00:01 problem with the live with this live
2:00:04 He's a human exactly
2:00:07 it'll be so much better when I'm when
2:00:08 I'm the synthesized version of me on it
2:00:11 hey Champion yeah Champion was here
2:00:14 Champion a good deal
2:00:16 all right
2:00:18 um I don't know what my name is oh wait
2:00:20 I don't wait I don't know who what my
2:00:23 name is why do I need to vectorize okay
2:00:26 so mandrid
2:00:30 having having done desktop publishing as
2:00:32 my day job when I was an actor I've
2:00:34 learned all about resolution and things
2:00:36 like that so
2:00:38 um Adobe Illustrator when it makes a
2:00:42 circle
2:00:43 it makes it out of math the the the
2:00:45 lines that that are made in Adobe
2:00:48 Illustrator are called vectors they're
2:00:50 math they're perfectly mathematical
2:00:54 um
2:00:54 representations of whatever the shape is
2:00:57 so you zoom in to
2:01:00 an illustrator Circle it will always be
2:01:03 perfectly smooth you'll never see pixels
2:01:06 a Photoshop image the more you zoom in
2:01:09 on it you'll see all the pixels that
2:01:10 make it up so the the thing that looks
2:01:12 like a perfect circle the more you zoom
2:01:14 into it you'll see all these little
2:01:15 stair step pixels
2:01:18 mid-journey stable diffusion all of them
2:01:21 right now their their default image size
2:01:25 used to be 512 by 512 pixels DPI or
2:01:30 512 pixels by 512 pixels
2:01:34 now they're doing 10 24 by 10 24 so so a
2:01:38 thousand pixels but if you were to if
2:01:41 you were to print that out you could
2:01:43 really only print that out to like I
2:01:45 don't know
2:01:46 a three or four inch square before you
2:01:50 would start to see kind of softness and
2:01:52 pixelization and if you tried to print
2:01:54 it really big it would just look like
2:01:56 [ __ ]
2:01:57 there are
2:01:58 new upscaling AIS that there's one in
2:02:03 stable diffusion that you it's got all
2:02:04 these settings and then there's one
2:02:07 called topaz labs
2:02:10 that are not just
2:02:12 um doing what Photoshop used to do which
2:02:14 is called interpolation where they
2:02:16 basically just try to fill in the stair
2:02:17 steps with extra pixels what's happening
2:02:20 with these new upscalers is they're
2:02:22 adding new detail but it's still a
2:02:25 pixelized image what vectorization does
2:02:28 is it turns it into those hard lines so
2:02:31 you could take a mid-journey image of a
2:02:33 logo
2:02:34 use Trace journey to vectorize it then
2:02:37 take it into illustrator and fine-tune
2:02:39 all of the lines and it would be perfect
2:02:41 and you could print it on a billboard
2:02:43 and it would look awesome right so
2:02:45 that's that's the difference
2:02:50 um don't ask me to explain the other V
2:02:53 word
2:02:54 [Laughter]
2:02:58 how is Bard's search of Google Drive
2:03:00 different oh that's that's what we asked
2:03:02 that's what I I speculated before that I
2:03:05 have a feeling they're doing a
2:03:06 combination of traditional search and
2:03:08 and uh
2:03:10 GPT stuff there I I could be very wrong
2:03:13 on that though so do not do not hold me
2:03:16 to that that's one of those answers I'm
2:03:19 I gave you speculation right there and
2:03:21 that's one of those answers that if I
2:03:22 were qualified I could actually answer
2:03:24 that but I'm not so that's my my defense
2:03:27 here is that I'm completely unqualified
2:03:30 [Laughter]
2:03:39 um
2:03:41 hey we're at 40K we're actually up to 42
2:03:43 now
2:03:46 where's Amazon AWS with artificial
2:03:50 intelligence AI so so goat Eric forner
2:03:53 AWS announced
2:03:56 three months ago four months ago
2:03:59 they've got a whole Suite of AI models
2:04:03 available as part of AWS now including
2:04:07 one of their own and including
2:04:09 essentially everything except open AI
2:04:15 because they're because Microsoft's
2:04:17 their big competitor on the cloud
2:04:18 business right Microsoft has Azure and
2:04:22 they're they are deeply in bed with open
2:04:24 AI so AWS will support everyone else go
2:04:28 [ __ ] yourself open AI
2:04:31 um so they're there I heard
2:04:36 I heard two things one I heard here in
2:04:39 the live
2:04:40 I heard that they're gonna come out with
2:04:42 a with a an Alexa that actually
2:04:46 is like pie that you can talk that you
2:04:48 can talk to it it's not hard-coded Alexa
2:04:51 like Siri and Alexa right now are these
2:04:53 hard-coded
2:04:54 piles of garbage
2:04:57 um so I think Alexa is gonna get better
2:04:58 and then I I kind of recall hearing that
2:05:02 AWS is is going to do a
2:05:06 chat GPT like thing although you know
2:05:08 what that might have been today
2:05:10 I was listening to the uh the marketing
2:05:13 AI institute's podcast today
2:05:16 and I think they were talking about meta
2:05:18 I think meta is coming out with a chat
2:05:20 gbt like thing that they're claiming is
2:05:22 going to be as good as chat CPT for
2:05:24 every everyone's claiming it's gonna be
2:05:26 better than Chachi ebt4 they have to say
2:05:29 that
2:05:31 um
2:05:31 but anyway AWS is in the game they're
2:05:35 just more on the so so remember well I
2:05:38 don't know if how long you've been here
2:05:39 tonight but but if you've you've you've
2:05:41 been in these things before when I talk
2:05:43 about that inflection point of this
2:05:45 channel is kind of from November 30th
2:05:48 2022 forward and and you know the sort
2:05:51 of whole history of machine learning and
2:05:53 you know development and engineering and
2:05:56 research that that's really not this
2:05:57 Channel That's the [ __ ] that came before
2:05:59 a lot of AWS
2:06:02 um what they're doing right now is
2:06:04 living more on the machine learning side
2:06:05 of this thing so if you want to be
2:06:07 building applications doing things
2:06:09 internally like I've got a proposal
2:06:11 right now out to a big Pharma company
2:06:14 um and we're going to be doing that on
2:06:15 AWS so so we'll be building on that
2:06:17 infrastructure
2:06:19 um but that's not what this channel is
2:06:20 about so
2:06:23 when that when they do a fancy Alexa
2:06:26 then they'll be entering this space
2:06:28 they're just not there yet
2:06:30 this is just the beginning and it's
2:06:32 mind-blowing uh for their starting point
2:06:34 at that exactly that that's one of the
2:06:37 things that that Paul writer says and
2:06:39 that I that I say all the time as well
2:06:42 is um
2:06:44 and Ethan Malik says this too
2:06:47 these tools
2:06:49 are the worst they will ever be
2:06:53 they're the worst they will ever be
2:06:55 I just feel empty
2:06:58 in the in the uh in the Irregulars
2:07:01 channel in the salon Discord there's a
2:07:03 good conversation going on with the
2:07:05 avocado so we've now moved from magic
2:07:07 hedgehogs to to empty avocados
2:07:12 oh my God this is fun
2:07:14 [Laughter]
2:07:17 um what town is it 10 13. I got okay I
2:07:21 gotta do a little bit it's Saturday
2:07:22 night I gotta go soon let's see no
2:07:24 you're not uh we've been pining like
2:07:27 crazy what was the V word you were using
2:07:30 earlier
2:07:32 thank you Sky Jimbo I appreciate that I
2:07:35 didn't quite see what it was it was it a
2:07:37 pizza or I don't know what it was but it
2:07:38 was cool thank you uh the big problem
2:07:40 with this live He's a human
2:07:43 there's there's many problems of this
2:07:45 life
2:07:46 V is for Vector the V word
2:07:50 you're very red the vector monologues
2:07:54 vectorized that was it
2:07:57 oh man velor that was not it velor is a
2:08:01 good one
2:08:01 um we learned it from watching you Dad
2:08:05 oh what's that I missed the damn day oh
2:08:09 I missed on on the damn day
2:08:11 um velvet
2:08:13 um
2:08:17 we gonna go tonight last night oh we're
2:08:20 gonna go to like last night 17K yeah ion
2:08:22 China that that was not the 17 000
2:08:25 people last night were were likely
2:08:27 people
2:08:28 um but they I think they're an
2:08:29 engagement Farm there was there was very
2:08:32 little interaction with the channel like
2:08:34 additional followers only went up like I
2:08:38 don't know 70 or 80 people I figured
2:08:41 with 17 000 people you know I'd sort of
2:08:43 flip in and we'd I'd be at 50 000
2:08:45 followers now there was that that was
2:08:48 some sort of Engagement forming
2:08:49 situation it was fun though it was
2:08:52 really fun it threw everyone for a loop
2:08:55 like the mods were like what the [ __ ] is
2:08:56 going on here like I don't know
2:09:01 oh man
2:09:15 the odds speak for themselves over 83 is
2:09:19 a given
2:09:20 I got a good one for you
2:09:22 eluna.ai let's go look at eluna.ai I
2:09:27 downloaded aura.ai based on Sam Corn's
2:09:30 recommendation last night a
2:09:33 o e
2:09:34 e Luna
2:09:38 wait we're gonna dot a i
2:09:43 create unlock the full potential of
2:09:45 generative AI explore 42 000 images get
2:09:49 started with 15 credits
2:09:53 hum my stuff with GPT
2:10:03 [Music]
2:10:09 [Music]
2:10:13 yeah I'll receive product updates AI
2:10:15 tools text to image reimagine motion
2:10:18 blend
2:10:20 okay so this is this is
2:10:23 yeah we're gonna start seeing a lot of
2:10:24 sites like this
2:10:27 there's a there's a company called
2:10:28 descript that kind of does this for
2:10:30 video
2:10:32 um
2:10:33 what's the one from stability AI
2:10:36 um
2:10:40 um something.co
2:10:43 I don't have it in my list do I I did
2:10:45 but I got rid of it um
2:10:52 I can't remember it now
2:10:54 um
2:10:55 yeah anyway the this is oh fireflies
2:11:00 like this as well so if you go to
2:11:01 Firefly so so what these what these
2:11:04 sites are doing is they're taking a lot
2:11:06 of these different models that are you
2:11:08 know
2:11:09 there's a bunch of tools
2:11:12 written
2:11:13 that you can access
2:11:15 um that are based on stability ai's
2:11:18 image Generation stuff for example and
2:11:20 then and there's all sorts of models out
2:11:22 there so if you go to hugging face
2:11:23 there's all these different models you
2:11:25 can play with what these sites are is
2:11:27 they're putting nice interface front
2:11:29 ends on a lot of those discrete tools so
2:11:32 right now it right now the industry's in
2:11:35 this weird place where all the tools are
2:11:37 discreet very few of them are actually
2:11:39 integrated
2:11:40 I think this is a very temporary phase
2:11:43 because what's going to happen is the
2:11:45 the more sophisticated versions of these
2:11:48 models like like gpt4 and Dolly three
2:11:52 are going to be really tightly
2:11:53 integrated
2:11:54 and then the trademark for gpt5 talked
2:11:59 about text to voice and voice to text so
2:12:02 so pretty soon GPT chat gbt is going to
2:12:06 be able to do you know images words
2:12:10 um audio music right
2:12:13 um animation all within the same model
2:12:16 and then you'll be able to build
2:12:18 applications on top of that that are
2:12:20 just pulling from a single model all
2:12:22 these different things it can do so I
2:12:24 think we're I think these kind of sites
2:12:26 are kind of a temporary stepping stone
2:12:28 to a more integrated future but they're
2:12:31 they're interesting I mean my challenge
2:12:34 with any of these kind of tools is all
2:12:36 of them are 15 or 20 bucks a month they
2:12:38 all are like here's your 15 free credits
2:12:40 and you can make like three images and
2:12:43 then you have to pay for it and it's
2:12:44 just like you can't keep up with them
2:12:46 all and things are changing way too fast
2:12:49 Delhi
2:13:01 let me see retro
2:13:04 do you have access
2:13:07 to dial E3 question mark
2:13:15 um
2:13:17 Janet Jackson is coming just announced I
2:13:20 don't know what that's about
2:13:22 I'm in San Mateo California where are
2:13:25 you
2:13:27 oh you're talking to each other that's
2:13:28 cool I basically had an anxiety attack
2:13:31 today and used pi.a to calm me down who
2:13:35 needs real friends so so Jim that's so
2:13:39 as as a as a intermittent I go through
2:13:42 phases of having really nasty anxiety
2:13:45 and panic attacks and [ __ ] like that so
2:13:48 so I I empathize with you and it sucks
2:13:51 um
2:13:52 at one of the salons two two or three
2:13:55 salons ago
2:13:56 someone said you know they think that
2:13:59 these chat Bots are going to be good at
2:14:00 therapy and someone else in the meeting
2:14:02 really lost their [ __ ] and they're like
2:14:03 they can't do therapy
2:14:07 and
2:14:09 while they are not human and they don't
2:14:12 have real empathy
2:14:15 these tools are at Pi in particular is
2:14:19 really good at having conversations and
2:14:22 it's really good at providing context
2:14:25 and perspective and things like that so
2:14:28 it's really good to hear you say that
2:14:29 like thanks for sharing it first of all
2:14:31 and
2:14:33 I really do think that this idea of
2:14:35 virtual companionship is going to be
2:14:37 huge
2:14:39 and not like the creepy all [ __ ] up
2:14:41 you know you know I'm gonna be dating a
2:14:43 robot
2:14:46 but but you know if if you I've
2:14:48 mentioned this in this live before if
2:14:50 you haven't seen the movie Larsen the
2:14:52 real girl you should you should see it
2:14:54 because it
2:14:55 on its surface it's about a guy that's
2:14:58 dating a sex doll and when you hear that
2:15:01 you're like oh I know exactly what that
2:15:03 movie's gonna be but you don't know what
2:15:05 that movie's gonna be it's not that at
2:15:07 all it is and it's not and it's an
2:15:12 incredible story and it's an incredible
2:15:16 story of resilience and empathy and and
2:15:19 community
2:15:22 and
2:15:24 I think there's a massive role for these
2:15:27 conversational AIS to serve a role just
2:15:31 like you experienced Jim so thank you
2:15:32 for sharing that
2:15:33 um because because this is because as
2:15:36 weird as it is it's really easy to
2:15:38 demonize it and dismiss it and go oh
2:15:40 that's just freaky you're not really
2:15:41 talking to a person it's a robot
2:15:44 now
2:15:45 if you haven't played with seriously if
2:15:48 you're on here and you haven't tried go
2:15:51 to Pi dot AI
2:15:54 or ideally get the PI app on the App
2:15:58 Store search for pi personal assistant
2:16:00 the company is inflection
2:16:03 and put Pi into conversation mode
2:16:06 there's a little telephone icon and you
2:16:08 click the telephone icon and it just is
2:16:10 this sort of animated vertical bar and
2:16:13 you talk and then it goes
2:16:16 and then it answers you
2:16:18 and it's really designed to have a
2:16:20 conversation I had a long session with
2:16:22 pi today not about mental health but
2:16:24 talking about my
2:16:27 come on AI Biz ideas yeah exactly I had
2:16:31 I've I've had conversations with pi now
2:16:34 about my car's electrical situation what
2:16:37 were the what were the specials at
2:16:39 Costco that was a good one while I was
2:16:41 driving to Costco pie was telling me
2:16:44 what the specials were at Costco I'm on
2:16:46 the AI therapist side the DSM the DSM-5
2:16:50 is the wrong way about mental health
2:16:52 issues anyway yeah and it's
2:16:55 you know are these things going to be a
2:16:57 replacement for human interaction
2:16:59 absolutely not but but let's take it out
2:17:02 of mental health for for a moment
2:17:05 um because I think it's I think you get
2:17:07 into weird slippery slopes when you
2:17:09 start dealing with people harming
2:17:10 themselves
2:17:11 but let's take let's take
2:17:15 something in the medical space physical
2:17:17 therapist
2:17:18 so
2:17:19 you know I blow my knee out and I go to
2:17:22 the physical therapist and the physical
2:17:23 therapist does some stuff with me and
2:17:25 then the physical therapist tells me 30
2:17:28 out thing 38 things yeah like the movie
2:17:30 her yeah exactly the physical therapist
2:17:32 tells me 38 things that I need to
2:17:34 remember between now and our next
2:17:36 meeting and and here's the here's the
2:17:39 exercises and here's how you breathe and
2:17:41 here's how you do this and here's how
2:17:43 you do that and then I leave
2:17:44 and I wasn't listening to her anyway
2:17:48 and now I can't remember the 38 things
2:17:50 she said well imagine now I have that
2:17:52 same interaction with that physical
2:17:54 therapist
2:17:56 and at the end of the session she tells
2:17:58 me the 38 things and she goes if you
2:17:59 don't remember them that's okay
2:18:01 I've got a virtual version of me that's
2:18:03 available 24 7.
2:18:06 you can talk to it anytime
2:18:08 it it knows your history it knows my
2:18:13 exercises it it knows what I've
2:18:15 recommended
2:18:16 and you can ask it anything about your
2:18:19 recovery and anything that you ask it it
2:18:22 will summarize and tell me what you guys
2:18:24 talked about
2:18:25 and
2:18:26 if in talking to the virtual me
2:18:30 it determines that I need to be involved
2:18:33 it will let me know and I'll call you
2:18:35 within an hour but whatever it might be
2:18:37 any time of day imagine that imagine a
2:18:41 physical therapist being able to expand
2:18:45 their quality of service to 24 7 where
2:18:48 maybe they get paid for those
2:18:50 interactions
2:18:51 and then just move that across you know
2:18:54 every modality of care
2:18:56 including mental health I think that is
2:18:59 something that's really interesting gpd4
2:19:01 is a great therapy for mathematicians I
2:19:04 struggle oh how I struggle that's
2:19:07 awesome no one understands tensors
2:19:09 exactly
2:19:12 laughs
2:19:13 her was a great movie I enjoyed Tom
2:19:16 Hanks and Finch
2:19:17 never dated one but had a crush on Rosie
2:19:20 the robot made in The Jetsons that's
2:19:22 really funny
2:19:24 um I said wait oh what was that one
2:19:25 about the PT I just missed it that
2:19:27 karuna had up there
2:19:30 um
2:19:32 Pie app not in play sky Timbo
2:19:36 um not p-i-e just p-i personal assistant
2:19:40 see if see if let me let me see if it's
2:19:43 available in Google Play
2:19:45 is
2:19:47 inflection PI app in Android
2:19:52 store
2:20:04 Android version is not available but
2:20:06 coming soon
2:20:08 inflection AI
2:20:16 and Android version is coming soon when
2:20:18 was this written
2:20:22 September 1st okay so yeah so so it's
2:20:26 only for iOS right now thanks thanks for
2:20:28 pointing that out
2:20:30 I'm on the AI therapist side the yep the
2:20:33 DSM-5 is
2:20:35 is pi on Android it is not but Pi is so
2:20:39 you can just go to pi.ai on Android in a
2:20:42 browser
2:20:46 and it's a this is a bit of a pain in
2:20:48 the ass to get into voice mode but I
2:20:50 think you can do it
2:20:54 okay that's very helpful information
2:20:56 based on what you've told me it sounds
2:20:58 like you would need to charge sales tax
2:21:00 on the painting since you're selling the
2:21:02 painting as a business and it's being
2:21:04 shipped by you that means that you're
2:21:06 responsible for collecting and remitting
2:21:09 the sales tax to the state of Missouri
2:21:12 there may be some additional rules or
2:21:14 regulations that apply depending on the
2:21:17 specific details of the sale but
2:21:19 generally speaking you would
2:21:22 [Applause]
2:21:28 can you help me think about
2:21:33 some anxiety I'm having right now
2:21:39 absolutely I'm here to help with
2:21:40 whatever you need let's talk about your
2:21:43 anxiety and see if we can get to the
2:21:45 root of the issue do you feel
2:21:46 comfortable sharing a little bit about
2:21:48 what you're feeling anxious about and
2:21:51 also has this been an ongoing issue for
2:21:53 you or is this more of a recent
2:21:54 development
2:21:56 so it's not quite it's not quite what
2:21:59 the mobile experience is so I'm just
2:22:01 sort of putting it in voice mode
2:22:03 manually but I I know at some point I
2:22:06 got the website to behave like the the
2:22:09 uh
2:22:11 the app where where you don't have to
2:22:13 type into the box it's it's just
2:22:15 listening but but anyway but but on
2:22:18 Android in a browser you you could at
2:22:21 least you know have have some kind of an
2:22:23 interaction like this
2:22:25 um
2:22:26 but but yeah but again Pi so here's
2:22:28 here's the trick with pi if you're
2:22:30 trying to get [ __ ] done
2:22:32 like like with chat jbt where you're
2:22:35 like give me a list and it'll give you a
2:22:36 list
2:22:37 I won't do that but Pi will just keep
2:22:40 asking you questions and have
2:22:42 conversations I really want to unders I
2:22:44 I understand that you're
2:22:47 urgently trying to get a list of very
2:22:48 specific things but let me just
2:22:50 understand why you want to do that just
2:22:52 because I want them I understand you're
2:22:55 getting frustrated so
2:22:58 um but if you want to have a
2:22:59 conversation with it it's really quite
2:23:00 good
2:23:03 um I can't hear you I'm going through a
2:23:04 tunnel
2:23:06 um hey it's you again
2:23:08 um I'm trying to get whisper to work
2:23:10 it's painful but love your knowledge oh
2:23:12 that sucks sorry about that
2:23:15 um
2:23:16 I'm happy to show it again did you get
2:23:18 to the playground did you get to to open
2:23:21 ai's playground and go into complete
2:23:23 mode versus chat mode did you get to
2:23:25 that little
2:23:27 um microphone
2:23:29 and let me go see you know what now that
2:23:31 I think about it is there a web
2:23:35 interface for
2:23:38 whisper yet
2:23:42 and I'm going to say open AI
2:23:46 whisper is a free text
2:24:11 it looks like there's a Chrome extension
2:24:13 called whisper to chat GPT but that's
2:24:16 not
2:24:17 that's I don't think that's the
2:24:18 direction you want to go or I don't
2:24:20 think that's what you're looking for
2:24:23 um
2:24:25 web UI for whisper and audio
2:24:35 I built a web UI for open ai's whisper
2:24:40 join
2:24:41 now
2:24:44 is there a link
2:24:48 you can find the repo here screenshots
2:24:50 here
2:24:51 issue tracker here now this you have to
2:24:53 install
2:24:54 I want one you can just play with people
2:24:57 you know you you developers that are
2:25:00 like here I developed this web UI and
2:25:03 then you got to go install the ship from
2:25:05 GitHub it's like no that's not a web UI
2:25:08 that's something that someone who knows
2:25:10 how to program can can easily get a web
2:25:14 UI up and running
2:25:17 ultimate open AI whisper tutorial open a
2:25:20 platform how to use whisper
2:25:23 if anyone knows of an actual
2:25:29 web interface for
2:25:33 whisper
2:25:34 pop it in the comments below
2:25:38 sorry about yeah I know that's
2:25:39 frustrating it's frustrating that that's
2:25:41 again right what what you're
2:25:43 experiencing is that
2:25:46 these tools are janky pieces of [ __ ]
2:25:48 right now
2:25:50 what they do is remarkable but where
2:25:52 they are in their development is very
2:25:54 very very early we're early we're early
2:25:57 we're early we're
2:25:59 early are you aware of the activists
2:26:03 behind the stochastic parrots paper
2:26:07 I remember reading it or reading about
2:26:10 it
2:26:12 like six months ago
2:26:14 but I don't remember
2:26:16 um
2:26:17 so
2:26:20 castic
2:26:49 I mean I I'm not a I'm not aware of the
2:26:53 um of the activists behind it but but I
2:26:55 understand the concern I mean the basic
2:26:57 concern is that
2:26:59 is that these large language models
2:27:01 while they seem human are not
2:27:05 and yes I agree with that I don't think
2:27:08 they have sentience I don't think that
2:27:10 they have empathy
2:27:13 I don't think that they have feelings
2:27:15 and yet
2:27:17 they behave more empathetically than
2:27:20 most humans they're more capable than
2:27:21 most humans they can respond with nuance
2:27:24 and logic unlike almost all humans in
2:27:28 the United States right now
2:27:31 uh Kyle as the creator of the janky
2:27:33 image I should have it copyrighted can I
2:27:35 get royalties yes as soon as we
2:27:39 well I'll tell you if we uh if we end up
2:27:42 selling the janky pieces of [ __ ] t-shirt
2:27:44 absolutely we'll get your royalties on
2:27:47 that as soon as I start making money on
2:27:49 this thing we can we can figure out how
2:27:51 to split it up
2:27:53 um
2:27:53 I think at this point with the amount of
2:27:56 hours that I'm putting in I'm probably
2:27:58 I'm probably pulling in a you know solid
2:28:01 Buck 50 an hour something like that
2:28:04 but but I'm not doing this for the money
2:28:07 so
2:28:09 um I do use the janky the janky piece is
2:28:12 a [ __ ] image a lot it's a good one yeah
2:28:14 I mean the Emoji is the is really the
2:28:16 reason why to keep using it
2:28:19 this
2:28:20 because it says everything it needs to
2:28:23 say without putting the word [ __ ] on the
2:28:25 screen
2:28:29 I paid five dollars to open AI oh that
2:28:31 sucks because some other guy recommended
2:28:33 it oh wait oh here Emma
2:28:35 I'm unsure at this point I'm on a Mac
2:28:37 which is terrible but I'll keep trying I
2:28:39 managed to get whisper to run on Google
2:28:41 collab okay because some other guy
2:28:43 recommended it yeah yeah I paid five
2:28:46 bucks to open AI That's because you need
2:28:48 access to the API to get to that to be
2:28:51 able to use that interface
2:28:55 um
2:28:56 it's all good the janky will get so much
2:28:59 better it will
2:29:01 um
2:29:02 I look forward so Emma Emily
2:29:07 is it Emily or Emma Lee
2:29:10 um
2:29:11 follow along didn't have uh oh no I know
2:29:15 we can do
2:29:16 so let's go to
2:29:18 um we're gonna go here
2:29:24 so let's go to
2:29:26 there's champi
2:29:29 platform.openai.com playground okay
2:29:32 so Emily
2:29:35 um
2:29:36 if you're here if you paid them five
2:29:39 bucks you should be able to get an API
2:29:41 key you've got you know an account here
2:29:44 this button right here switch that from
2:29:47 chat to complete
2:29:50 have you done that
2:29:53 and then
2:29:55 in this big window in the upper right
2:29:58 hand corner is that little microphone
2:30:04 so
2:30:11 all of the AI tools right now are janky
2:30:15 pieces of [ __ ]
2:30:17 but if we give it 12 to 18 months
2:30:20 everything's going to be remarkable
2:30:24 so it says it's transcribing
2:30:26 all of the AI tools right now are janky
2:30:28 pieces of [ __ ] but we if we give it 12
2:30:30 to 18 months everything's going to be
2:30:32 remarkable so that's working with
2:30:35 recording
2:30:37 um
2:30:38 I don't know
2:30:42 I don't see here any
2:30:45 um
2:30:48 there's no controls here so if you got
2:30:50 it running in Google collab
2:30:53 there's probably all sorts of parameters
2:30:55 basically go to
2:30:58 oh this is this is [ __ ] awful right
2:31:00 now basically go to documentation and
2:31:03 find the whisper documentation
2:31:06 um
2:31:08 because if you have it running in collab
2:31:10 you can probably do things like time
2:31:12 code and things like that
2:31:14 creating automated
2:31:16 creating an automated me meeting minutes
2:31:19 generator with whisper and chat and gpt4
2:31:24 so here's all the here's all the
2:31:26 documentation
2:31:28 for whisper
2:31:31 but this is not what like
2:31:34 I assume what you don't want to do is
2:31:36 create a transcription application like
2:31:39 what you want is something that [ __ ]
2:31:41 works this is the joke I have with my
2:31:43 developers right now is I have very
2:31:45 simple acceptance criteria for their
2:31:47 work it should [ __ ] work
2:31:50 this doesn't [ __ ] work right now
2:31:52 anyway
2:31:54 um
2:31:55 yeah but but if you just want to
2:31:58 transcribe something flip it into
2:32:00 complete mode
2:32:02 click on the microphone and then upload
2:32:04 something there so hopefully hopefully
2:32:07 you haven't tried that and that actually
2:32:08 helps but if you have tried that and it
2:32:10 doesn't work I don't know what to say
2:32:11 I'm on a Mac so I got it to work here
2:32:15 um all right let me do a couple more
2:32:18 questions and then I got a motor on out
2:32:22 Robert Metzger wow did you know you
2:32:24 could make money with chat gbt I did not
2:32:26 know that Robert
2:32:29 oh yes I did
2:32:34 it's comedy it's comedy we're having big
2:32:39 time comedy
2:32:41 all right Emily thank you very much
2:32:44 Winston oh yeah making it rain thank you
2:32:47 very much sir I appreciate that
2:32:52 thank you so much oh this will work good
2:32:56 awesome
2:32:59 awesome good
2:33:01 yay we actually did something
2:33:04 I I now have a qualification Emily came
2:33:07 in and wanted to get whisper working and
2:33:09 now it's working well no she said it was
2:33:11 it will be working Wham Careless Whisper
2:33:16 uh where is the microphone oh did you
2:33:19 get it I think you got it
2:33:21 um this will work thank you so much
2:33:23 you're welcome you're welcome you're
2:33:24 welcome humans equal suing Americans who
2:33:28 listen to politicians who say inject
2:33:31 bleach you did okay great awesome
2:33:33 I'm a teacher and you did good
2:33:36 I'm teach I'm teacher verified
2:33:40 all right listen I think let's end it on
2:33:43 a high note again I'm disappointed that
2:33:46 you didn't bring your 17 000 friends
2:33:48 tonight we had a solid 70 for most of
2:33:50 the evening which actually I do prefer I
2:33:53 like this more intimate thing
2:33:55 um
2:33:56 really cool hanging out with you all
2:33:58 tonight I know it's kind of low-key
2:33:59 tonight hope you have a good Saturday
2:34:00 night go have nachos or tacos or
2:34:03 whatever the hell you have now a whiskey
2:34:06 um we'll bring them tomorrow yeah yeah
2:34:08 exactly forever hooked I need so listen
2:34:11 our goal tomorrow let's try to go from
2:34:13 70 to 10 000. I that seems reasonable it
2:34:17 seems reasonable to me just invite your
2:34:19 friends thank you Winston loving the
2:34:21 dangly lights
2:34:22 good night it's my anniversary what did
2:34:24 you do on your anniversary
2:34:29 I'm so sorry
2:34:31 I'm not on Tick Tock with some old dude
2:34:34 that looks like Rosie O'Donnell
2:34:37 what is my life coming to
2:34:43 goodbye everybody thanks for hanging out
2:34:45 everyone Thanks for thanks for uh those
2:34:47 of you who modded tonight and who've
2:34:49 been modding I really really appreciate
2:34:51 it it makes a big difference peace out
2:34:52 everybody I'll see you tomorrow