aiLL 6/10/23 - The A.I. Tsunami - What's Coming & How You Can Ride The Wave

Description

Chapters

Transcript

0:01 foreign
0:19 with the AI learning Labs getting set up
0:21 here
0:24 welcome welcome welcome
0:25 happy Saturday is it Saturday I think
0:29 it's Saturday
0:30 yeah it's Saturday Saturday I can answer
0:33 your questions hey Emilio's wife what's
0:36 happening
0:37 um I can answer your questions about AI
0:40 I can't answer your ask questions about
0:42 what day it is apparently I can't talk
0:44 either I need to get some water uh
0:48 should I just walk away while people are
0:50 coming in I probably shouldn't but I do
0:53 need some water
0:53 [Laughter]
0:57 all right
0:59 um
1:00 let's see my name is Kyle Shannon this
1:02 is the AI learning lab
1:03 uh I'm here to
1:06 wax poetic about generative AI get water
1:09 yeah I'm gonna go get water
1:11 came for the hairdo come for the hair
1:14 stay for the jokes about how I look like
1:16 Rosie O'Donnell I'll be right back
2:17 all right I'm back I'm back I'm back
2:20 you've been waiting so long for me to
2:23 return
2:24 [Music]
2:26 saying oh the Rosie O'Donnell joke is so
2:30 my trolls are creative at least so so in
2:34 the past week I was told I looked like
2:37 uh Rosie O'Donnell
2:40 um oh there's a lot of people here let
2:41 me tell you who I am my name is Kyle
2:43 Shannon this is the AI learning lab I'm
2:45 a long time entrepreneur
2:47 um I talk about
2:49 what I'm learning in AI in generative AI
2:52 so I'm relatively new to it I've been
2:54 deep diving for about a year but I
2:56 started a um
2:58 AI Salon I have a newsletter called
3:01 everyday AI I've got this Tick Tock
3:03 Channel and I'm rolling AI into my
3:05 business which is uh it's called
3:07 storyvine it's an automated video
3:08 platform and I'm completely reimagining
3:11 the product because of generative AI if
3:14 you've got questions about AI pop them
3:16 in the
3:17 comments and then you can watch as I
3:20 comically try to keep up with them but
3:23 don't and then new people are like why
3:25 won't he answer my question I put a
3:27 question in 17 times and it didn't
3:30 answer it
3:31 and then as I'm scrolling through the
3:32 questions there's this question 17 times
3:34 because the guy that you know
3:36 it's usually a guy
3:39 um what else I was a
3:43 I I did some pioneering stuff back in
3:45 the early days of the web I started one
3:47 of the first online magazines and one of
3:49 the early digital agencies called
3:51 agency.com
3:52 and
3:54 um
3:55 why I mentioned that is the parallels
3:58 between what was happening in the early
4:01 days of the web and what's happening
4:02 with this when in the early days of
4:04 generative AI are like fascinating and
4:07 amazing and and it feels like a very
4:09 similar
4:11 um major shift that's happening that
4:14 we're at the very beginning of
4:16 um uh so so back in the early days of
4:18 the web the internet had been around for
4:20 a bunch of decades but it was mostly
4:21 used by scientists and researchers and
4:24 universities and then the World Wide Web
4:26 came along and all of a sudden us
4:28 regular people got to play with it and
4:30 was like oh that's kind of interesting
4:33 and similarly with AI it's been around
4:35 for a bunch of decades and um Google
4:39 invents this thing called the
4:40 transformer in 2017 and
4:44 you know it was a lot of years ago and
4:47 it's it's a scientific paper called
4:48 attention is all you need and that ends
4:50 up becoming the pivotal technology that
4:53 is that lays the foundation for
4:55 mid-journey and stable diffusion and
4:57 church EPT and all of this super sexy
4:59 stuff we get to play with
5:02 um and so it's a very similar thing and
5:04 and one of the differences is the speed
5:06 of adoption so
5:08 um uh the World Wide Web took about six
5:11 years to get from zero to 100 million
5:13 users Chachi BT six weeks so anyway all
5:18 right let me go answer some questions
5:21 stop babbling Kyle answer my question
5:23 I've got questions I've got questions
5:26 all right let's see what your questions
5:28 are best AI app the reason I put these
5:31 URLs up on screen
5:33 in my in my fancy little double camera
5:37 window
5:38 is that I think right now
5:41 and probably for a lot of years just
5:44 start learning chat GPT the the depth of
5:47 its capability is insane so
5:50 chat.openai.com is the official chatgpt
5:54 website you don't need an app most of
5:57 the apps out there are fakers there's
5:58 one official chat GPT app on iOS there's
6:01 not one for Android yet so don't
6:03 download any of that [ __ ] just point
6:05 your browser at that
6:06 I don't get paid for this like I should
6:09 hey chat EBT if you're listening hook me
6:12 up baby
6:14 you know what give me code interpreter
6:16 that would be payment enough okay and
6:18 I'll talk about that in a second so so
6:20 that's that bing.com is gpt4 but for
6:24 free it's Microsoft's chat bot that uses
6:28 open ai's gpt4 and it's connected to the
6:31 internet so if you don't want to pay 20
6:33 bucks a month to open AI
6:35 to get their version of chat jbt4 then
6:38 you use bing.com and then poe.com gives
6:41 you access to a bunch of different
6:42 models and I did something really cool
6:44 today
6:45 so one of the models you have access to
6:47 that uh chatbot just like chat EPT
6:50 except it's got all these different
6:52 models so Sage is pose model then you
6:56 got gpt4 Cloud plus Cloud instant 100
6:59 000 I'm going to show you that in a
7:00 second you got Claude instant which is
7:02 ripping fast and you got chat gbt these
7:05 three that say subscribe you have to pay
7:06 for but this one in particular well I
7:10 don't don't look at that yet okay we're
7:11 gonna clear this context
7:14 so here's what happens with
7:18 um
7:19 with
7:20 a hundred thousand okay here's what that
7:23 means the 100K in this case means a
7:26 hundred thousand tokens
7:28 what's that mean well with chat GPT you
7:30 have 4 000 tokens which for whatever
7:33 reason they they call them tokens and
7:36 instead of words so four thousand tokens
7:38 translates into about three thousand
7:40 words it's a geek thing the Geeks
7:42 decided to call them tokens because
7:43 technically that's what's going on but
7:45 realistically it's three thousand words
7:47 so a hundred thousand tokens means
7:50 seventy five thousand words so with chat
7:53 gbt if you have a really long document
7:55 and you like paste it in there chat gbt
7:57 goes like yeah I can't do it I don't
8:00 wanna it just can't deal with it but a
8:02 hundred thousand you could take say a
8:05 novel or a screenplay and pop it in
8:07 there so what I did earlier was I went
8:10 to
8:11 a site here's the script of The
8:13 Godfather
8:15 so I'm going to copy that
8:20 and then I can just go in here into into
8:23 this little chat window this is
8:24 mind-blowing to me
8:26 all right
8:28 um
8:29 based on this screenplay
8:34 and then I'm going to hit shift return
8:35 to make some line breaks I'm going to
8:37 paste
8:38 and what I pasted in was all of The
8:41 Godfather
8:43 right so this is Mario puzo's screenplay
8:46 so pages and pages and pages right
8:50 um chat EVT would absolutely choke on
8:52 this
8:54 wait okay now I'm at the end now based
8:56 on this okay
8:58 um please
9:00 create a table of the main
9:04 characters in markdown
9:09 and
9:11 provide
9:15 uh or n and tell me what happens
9:21 to them
9:22 and
9:24 on I don't know if this has got Pages
9:27 let's see
9:29 what happens to them and how far into
9:33 the story
9:37 uh let's let's see what that does
9:41 so it just read all of The Godfather and
9:44 now it's thinking those three little
9:46 dots waiting waiting wait for it
9:50 calm down people still waiting
9:53 still waiting
9:56 room
10:00 and it is building us a table
10:03 this is [ __ ] insane it's just insane
10:07 all right let's see Don Vito Corleone
10:10 was shot by celozo but survives Michael
10:13 eventually takes over the family Don
10:15 Vito passes away later in the story
10:16 Michael Corley Corleone initial State
10:19 initially stays out of the business but
10:21 he's dragged into it after Sunny is
10:23 killed Sonny he's killed in an
10:25 orchestrated hit right
10:27 so you can take incredibly long
10:30 documents and just start interacting
10:32 with it and the other cool thing about
10:34 this is
10:35 um what Claude does here is it kind of
10:38 senses that these are main characters so
10:41 oh you can't really see it these are
10:43 different colors
10:45 um like uh Michael Corleone is a
10:47 hyperlink and if I click on that
10:50 it says tell me more about Michael
10:52 Corleone so it writes a prompt for me
10:56 and now it's gonna go tell me about
10:58 Michael Corleone
11:00 wild huh all right
11:02 that's that so
11:05 the question was what are the Cool Tools
11:07 so so you could just start with chat gbt
11:10 or Poe 100K or uh what is it Claude 100k
11:14 from anthropic
11:16 you could probably spend the next decade
11:19 figuring out amazing [ __ ] to do with it
11:22 in terms of images
11:24 um
11:25 mid-journey is the best in my opinion if
11:28 you like geeking out then stable
11:30 diffusion lets you do like the developer
11:33 Community is just you know nuts unstable
11:35 diffusion but you got to geek out a
11:37 little bit to use it those are the two
11:38 image things
11:40 um I just did my first Runway ml Gen 2
11:44 if you haven't heard of that
11:47 so that was my first attempt
11:51 I said I described it as uh
11:56 what I don't know woman in a flowing
11:59 white dress walking through a field of
12:01 flowers with a gentle breeze and a
12:04 cinematic tracking shot
12:06 and that's what it generated in about I
12:09 don't know 30 seconds
12:10 so that's pretty cool to play with
12:14 um yeah I don't know there's lots
12:16 there's lots start with chat GPT start
12:19 there and and then uh
12:22 then all will be revealed to you all
12:24 right what's the best AI that's already
12:26 trained you can install on a server
12:28 honest to God there are so many there
12:31 are so many open source large language
12:34 models they're being released like every
12:37 three days and like each one's topping
12:39 the one before it
12:41 um there's a thing called GPT for all
12:43 and I think what it does is it it it
12:46 installs a bunch of [ __ ] on your server
12:48 and gives you access to multiple
12:50 different models so I think a project
12:52 like that where someone else is keeping
12:56 track of all those large language models
12:58 is probably the way to go so so look up
13:00 GPT for all
13:02 do you do Twitter spaces I have not yet
13:04 but I probably could and should I I
13:06 don't know I I like being able to see
13:09 [ __ ] and show [ __ ] because but that's
13:11 probably not a bad idea I don't know I
13:12 might also put together a um
13:15 what do you call it a workshop to
13:17 actually teach people how to do stuff
13:19 but I don't know
13:20 I got I gotta I got a company to run
13:23 people I can't just be sitting here on
13:25 Tick Tock being a goddamn influencer I
13:27 mean look at me I'm I'm like a handsome
13:30 Rosie O'Donnell
13:32 foreign
13:35 oh God
13:38 all right thanks Kyle I'm using gpt4 but
13:41 it's not connected to real-time data yes
13:43 it is what am I missing okay
13:46 let's go turn this [ __ ] on
13:49 um what opening I decided to do was
13:53 make it so convoluted and complex to get
13:57 to that you just can't figure it out
14:00 okay so you're at chat gbt if you've
14:04 paid for your 20 bucks a month and have
14:07 chat GPT you'll see jet gpt4 there but
14:11 what you probably haven't done you have
14:12 to go to your settings
14:14 so go to settings and then within your
14:17 settings the second choice down here is
14:19 beta features turn on the beta features
14:22 and you have to turn on you have to
14:23 explicitly turn on browse with Bing and
14:25 plugins once you do that now when you
14:28 roll over gpt4 you've got three
14:31 different options you've got the default
14:33 option which is not connected to the
14:35 internet and frustrating as [ __ ] because
14:37 all the data ends at September 2021 then
14:41 you have browse with Bing so this one is
14:45 who won the Nuggets game last night and
14:50 if you've been watching my lives you're
14:51 probably thinking come up with another
14:53 example Kyle that's dumb you keep saying
14:56 the nuggets
14:58 but it knows it doesn't know that
15:01 so it goes to browse the web
15:05 and it rewrote the prompt for me and it
15:07 says the Denver Nuggets one in in the
15:10 game 108 to 95 they held Miami to less
15:14 than a hundred
15:16 I'm in Denver by the way don't give two
15:19 shits about basketball but hey I'm I'm a
15:21 fair weather fan where you're in the
15:23 finals I'll root for you
15:26 and I don't care about Miami so yeah
15:37 come on calm down
15:40 calm down gonna be okay
15:43 lay down lay down all right
15:47 hey
15:52 all right so
15:55 thank you
15:57 25 years ago I had some really poorly
15:59 behaved dogs and we got a dog trainer
16:02 and what I learned when we got a dog
16:03 trainer was that they don't train the
16:05 dog
16:06 they train the humans
16:08 and you have to do [ __ ] like don't let
16:10 them jump on the furniture and you know
16:11 it's basically on you for for having a
16:14 dog that's not poorly behaved and so I
16:17 made a decision about 25 years ago oh
16:19 I'm just gonna have shitty behaved dogs
16:21 all my life because I don't have
16:22 boundaries certainly when it comes to
16:24 pets all right
16:26 oh no Tick Tock you just jumped me down
16:28 this is this is one of the uh when I
16:30 when I do my lives most of my time is
16:32 spent trying to find myself in the
16:34 comments
16:35 um oh it looks like there's some new
16:36 people if you're new here my name is
16:37 Kyle Shannon this is the AI learning lab
16:39 I'll answer your questions about
16:41 generative AI I'll make jokes I cuss
16:45 um
16:45 uh what else do I do
16:48 um I don't know
16:49 these URLs right here are just there
16:52 kind of as a public service announcement
16:54 the top one was chat gbt the next one
16:56 down is Bing which is gpt4 that you can
16:58 use for free and Poe gives you access to
17:01 multiple models that you can compare so
17:05 that's super cool and fun okay
17:09 how do you think AI will change research
17:12 fundamentally
17:14 fundamentally it will change it
17:17 I think forever because it's kind of
17:18 instant like like the thing I just did I
17:20 don't know if you're here a minute ago
17:22 where I used
17:25 um po the Claude hundred thousand
17:28 um there and then it automatically
17:32 created a table of all the characters
17:34 and what happens to them in the
17:36 screenplay like like you can read and
17:39 summarize a screenplay so so think about
17:42 there are a lot of human beings on a
17:45 planet whose job it is to read
17:46 screenplays and and and create what's
17:49 called coverage for a screenplay so they
17:51 hand it to the execs you could
17:53 absolutely automate that so a screenplay
17:54 comes in you run it through this thing
17:56 it creates it creates automatic coverage
17:58 it will probably even score it
18:01 based on quality uniqueness what genre
18:04 it's in things like that
18:06 um instantly see you know same thing for
18:08 research and now that
18:10 um
18:11 excuse me now that
18:13 gpt4 is connected to the internet you
18:16 can basically just point it at big data
18:17 sets
18:18 and then you've got
18:21 um this new thing coming out called code
18:23 interpreter which I don't have access to
18:25 and I'm very bitter about that but it's
18:28 a new mode of chat GPT and you can
18:31 upload giant data sets to it and it does
18:35 full-on data analysis including creating
18:39 um you know graphs and charts and things
18:40 like that it's amazing so it's going to
18:43 fundamentally alter it so like here's
18:46 the thing
18:47 a lot of the work that we do in today's
18:49 world is doing
18:53 kind of low value processing right so so
18:57 if I read screenplays for a living like
18:59 I have to take a number of hours to just
19:01 basically get to understand what the
19:03 story is about who the characters are
19:04 what the main arcs are
19:06 well if I can do that in three minutes
19:09 or five minutes or if I can automate it
19:10 to just do it when a screenplay comes in
19:12 it generates that report
19:14 now I can use my brain for higher level
19:16 stuff so now I might rather than
19:19 considering you know 10 screen plays I
19:22 might consider a hundred screenplays and
19:24 I might really sort of mix and match
19:26 them and compare compare them in a
19:28 different way
19:29 so it's not like research goes away or
19:32 you know screenwriter coverage goes away
19:34 it's that the job completely changes and
19:36 I think that's what what all of us are
19:38 going to face in in all knowledge work
19:41 all creative professions
19:43 it is not that you know the AI is coming
19:46 to take our jobs it's that as human
19:48 beings learn to use these incredibly
19:51 powerful tools the jobs will change and
19:54 then people that so Paul rateser who
19:56 runs the marketing AI Institute says
19:59 um AI will not take your job but someone
20:01 using AI will right so so the purpose of
20:05 this channel is to encourage you to like
20:07 go there start playing get curious about
20:10 this stuff if you do a particular job if
20:12 you're a researcher start figuring out
20:14 how these tools can augment your work
20:18 and then that will serve you incredibly
20:20 well sit still no I have add this is how
20:23 it is
20:24 you're not babbling that's what
20:26 ADHD meds do I'm not on ADHD meds
20:30 mid-journey versus stable diffusion
20:32 mid-journey mid Journey mid Journey
20:34 however stable diffusion has a lot more
20:37 development going into it because it's
20:38 all open sourced so if you want to geek
20:41 out like if you're technical
20:43 and you don't mind going to GitHub and
20:46 installing things on Google collab and
20:49 you know virtual machines and [ __ ] like
20:51 that stable diffusion is amazing there's
20:53 a lot of tools there I still don't think
20:55 the quality of stable diffusion is as
20:57 good as mid Journey
20:59 both of them however have been trained
21:02 on copywritten images
21:06 so using mid-journey and stable
21:09 diffusion for commercial work especially
21:11 if like you're an agency doing it on
21:13 behalf of a client there's a risk that
21:16 at some point someone's going to get
21:17 sued for using that by one of the
21:19 artists whose data was trained so
21:22 Adobe their Firefly products while it's
21:25 not anywhere near as good from a just
21:28 raw quality standpoint as those other
21:30 ones it's only been trained on open
21:33 source public domain and Adobe stock
21:36 images that they own all the rights to
21:38 and they actually two days ago announced
21:41 Adobe for Adobe
21:45 Firefly for Enterprise so fireflies
21:47 their generative AI product so they're
21:50 now making it safe for companies to use
21:52 that so if you're in a commercial
21:53 environment I'd use Adobe Firefly just
21:55 because it's it's safe and that's going
21:58 to be their intent they're allowing
21:59 artists to opt out opt-in they're going
22:02 to be doing attribution and compensation
22:04 so
22:08 how different is three to four
22:11 um for for GPT the difference between
22:13 3.5 and 4 3.5 is much faster so if you
22:17 need speed
22:18 um 3.5 is like wicked fast
22:21 if you need Nuance in language
22:24 gpt4 is staggeringly good it's just it's
22:28 it's just way better apparently it codes
22:30 better as well if you're doing coding
22:32 apparently gp24 is much better at coding
22:35 but gpt3 is not bad like my um
22:39 my big Epiphany with chat GPT was riding
22:42 a python application in like 90 minutes
22:44 I've never written a line of python in
22:46 my life and I I created a functional
22:49 very sophisticated application in 90
22:52 minutes with GPT 3.5 so it's good to pay
22:57 the 20 bucks have you used the new GPT
22:59 plugins yeah I have
23:02 um they're good they're they're so let
23:05 me let me show you how to do the plugins
23:06 there so so basically
23:09 the way it works is you you have to pay
23:11 for GPT Plus
23:13 you have to enable the beta features in
23:16 your settings and then you have access
23:18 to browse with Bing and plug-in so if
23:20 you go to plugins you now have this drop
23:23 down menu and and here are the plugins
23:25 I've got installed so you can install as
23:27 many as you want but you can only
23:29 activate up to three per session
23:32 so the way I've kind of thought of it is
23:34 you you basically group together related
23:38 plug-ins it might be one or two or three
23:40 and what you're doing is creating kind
23:42 of a custom application so let me see
23:44 okay I did something last night
23:47 so I'm gonna I'm gonna turn on
23:49 experiences and then I'm gonna turn on
23:52 get your guide
23:55 and I'm gonna turn off ask your PDF and
23:59 I'm going to turn off link reader and
24:02 I'm going to turn on trip.com Okay so
24:04 I've got three wait no I don't now I
24:08 have three okay I have three travel
24:10 related plugins installed right
24:14 okay so now I can go down here and say
24:19 um
24:20 um I'm going to be in
24:24 um
24:25 Florida
24:29 in
24:31 the Tampa area
24:35 for five days in September
24:42 um make me a detailed
24:47 um itinerary of
24:51 day and evening
24:54 activities I like to fish listen to
25:00 music
25:02 and
25:03 eat like
25:05 a pig
25:09 all right
25:12 so now it's thinking and it should yeah
25:16 so it says using get your guide so it
25:18 fires off one of the plugins and then
25:20 each of these plugins you can open it
25:22 and see what it did so it
25:26 said Tampa Florida fishing tours so it
25:29 pulled fishing out of there so so it got
25:31 some results
25:32 and then it did
25:36 Tampa Florida music activities cool
25:40 and then it did
25:48 Tampa Florida food and Gourmet tours
25:52 and so it took all of those responses
25:55 and it said here's your detailed trip
25:58 you know start with a private in Source
26:01 inshore fishing in Tampa Bay
26:03 book here it says let's click on book
26:06 here sometimes it will give you
26:07 hyperlinks that are broken
26:09 no that's not broken so I can go book
26:11 that trip right so I didn't have to
26:14 Google finding some dude or dudette that
26:17 you know takes you out on on fishing
26:19 trips
26:21 um afternoon after a morning of fishing
26:23 enjoy a leisurely lunch at a local
26:25 seafood restaurant and that I can tell
26:27 it to book me that
26:29 um
26:29 day two day three day four day five so
26:32 and there's the you know little links it
26:34 has so that's plugins for you so so like
26:37 what I've just created in in whatever
26:39 that was a minute or two was a little
26:42 specialized travel application and and
26:45 this is I think this is
26:47 what we're gonna start seeing is that is
26:49 that really specialized and personalized
26:52 applications you'll be able to generate
26:54 them for yourselves kind of kind of in
26:56 real real time and I've even heard that
26:59 um
27:00 kind of the next iteration of plugins is
27:02 you won't have to manually turn them on
27:04 like I just did like it'll just figure
27:06 that [ __ ] out it'll it'll it'll
27:08 basically know the entirety of the
27:10 plug-in universe and when you ask it to
27:12 do something if it finds a plug-in that
27:14 can do it better than its large language
27:16 model it'll use it
27:18 what book do I recommend for a good
27:20 foundation on AI
27:22 if you want to learn about the
27:24 underpinnings the machine learning and
27:26 how all this stuff works
27:28 um I would probably go to MIT or
27:31 Stanford
27:32 um like online courses and take an
27:34 online course because they're going to
27:36 be the most up to date
27:38 um if you if you're just talking about
27:39 this generative AI stuff there's not
27:41 going to be a good book on it right now
27:42 it's too new it's too it's too early so
27:46 just get curious go there and start
27:48 playing start digging in start watching
27:51 YouTube videos start watching Tick Tock
27:54 videos of people that you
27:57 um trust Rachel Woods she's got a thing
28:00 called the AI exchange and she does tick
28:02 tock she's really good
28:04 um there's a guy named Mackenzie Bose
28:06 who's got an app called GPT boss which
28:08 is a kind of overlay on top of gpt4 it's
28:11 pretty slick he's got a good Tick Tock
28:13 Channel if you want to geek out there's
28:15 a guy on YouTube named David Shapiro
28:17 who's who's building autonomous agents
28:21 and he's got a whole open source project
28:23 doing that he's [ __ ] Wicked smart
28:25 when I want to feel humbled and stupid I
28:27 watch David Shapiro
28:29 I'm like oh I could do this and then I
28:31 watch him I'm like I can't do this
28:35 did you not have a degree in acting yeah
28:47 it's processor measurement okay I don't
28:49 know what that means what book would you
28:51 recommend holy cow I was looking for
28:53 something like this
28:54 like this hi nice to meet you Victor all
28:59 right I have paid version and wrote
29:01 basic code to help automate time
29:02 consuming part of my job yes
29:05 do that
29:07 here is my feeling
29:09 there is a very real possibility there
29:12 was a paper put out by Mark and Drayson
29:15 who created Netscape he was one of the
29:16 early pioneers of the internet of the
29:19 interwebs
29:20 he just wrote a really long paper on how
29:22 he thinks that um generative AI or AI is
29:26 going to be the the salvation of
29:27 humanity not the just the the death of
29:29 it
29:30 and
29:31 um
29:32 things like that things like using AI to
29:35 automate the shitty parts of your job
29:38 imagine a future where all of the jobs
29:41 that are the soul-crushing you're a cog
29:44 in a wheel
29:45 and you can't believe you [ __ ] do
29:47 this for a living and you live for
29:49 Friday and you live for the weekend and
29:50 God damn you hate your job what if all
29:53 those jobs were automated out
29:55 and you got to choose what you wanted to
29:58 do
29:59 that's a possibility of a future that
30:01 we're living into and so the sooner that
30:04 you learn what these tools can do and
30:07 start to do things like what's what's
30:08 your name
30:09 um Superfluous honey awesome Name by the
30:12 way
30:13 um
30:14 if you start doing what she's doing
30:17 um
30:18 okay
30:22 here's the downside
30:24 everything's gonna [ __ ] change
30:25 everything's gonna be disrupted if you
30:28 think you're immune from this you're not
30:31 the upside of it is every single one of
30:34 us has an opportunity to reinvent
30:36 ourselves in dramatically positive ways
30:38 if we choose to do it but you gotta step
30:41 in and do it you got to get curious you
30:43 gotta you gotta be willing to go to the
30:45 uncomfortable place of oh [ __ ] I gotta
30:47 learn something new really yep
30:50 because if you don't someone else will
30:53 and it's not necessarily just going to
30:54 be the young kids that know how to use
30:56 the Internet it's going to be old [ __ ]
30:57 like me gen xers that are like this is
31:00 [ __ ] cool man reinvent themselves and
31:02 all of a sudden have super powers again
31:04 that's coming
31:06 think we have synergies cool connect
31:09 with me my my contact info is in my bio
31:11 and you can DM me and I've got contact
31:13 info up there and you can join the AI
31:15 Salon that I do there are so many
31:17 opportunities
31:19 based on tokens you said on this
31:22 platform to download this thank you yes
31:24 I don't quite know what I said there and
31:26 I don't quite know what you mean but
31:28 awesome amazing thank you what book do
31:31 you recommend I don't for good
31:33 foundation on AI uh I I don't know of
31:36 any I'm I have ADD I can't read
31:38 [Laughter]
31:41 all right kind of concerning too yeah so
31:43 here's here's my here's my thoughts on
31:45 all this AI stuff I am more excited that
31:49 I've been in decades and I'm more
31:51 terrified than I've been in decades and
31:53 that feels like the appropriate balance
31:55 to me that if you're too excited about
31:58 it or if you're too scared about it both
32:00 of those are bad
32:02 um you should have a healthy balance
32:04 like life is about balance
32:06 total excitement and joy and shocking
32:10 horror
32:12 that's what AI is right now for me and I
32:15 think that's the appropriate thing all
32:17 right uh Duke dude stop asking questions
32:22 um I love these lives thank you llms
32:25 need a lot of Graphics memory won't work
32:27 on my laptop oh that's changing real
32:29 quick
32:30 um just go just go to Twitter and look
32:32 up open source llm there are yeah it's
32:37 come they're coming out like every three
32:38 days
32:39 um increasingly optimized um large
32:42 language models you can install on like
32:44 an M1 Mac
32:47 um there's one that someone did that can
32:48 be installed on a phone
32:51 um so so the developer Community is
32:53 taking the Llama models that meta
32:55 accidentally released to the world they
32:58 might have been accidentally released
32:59 them but they did they released them to
33:00 the world so now the entire developer
33:03 Community has these core models that
33:05 cost millions of dollars to create and
33:07 they're fine-tuning them and optimizing
33:09 them and making installs for Macs and
33:11 PCs and things like that so
33:13 um just go look at the open source stuff
33:15 there's there's a lot there you may need
33:17 to geek out a little bit to get it
33:18 installed but it's getting easier every
33:20 day
33:23 handsome Rosie O'Donnell thank you all
33:25 right Duke what book would you recommend
33:27 Duke I've already answered and used I'm
33:30 asking the question
33:33 oh Duke
33:37 I use Bing for basic AI yeah Bing's
33:40 pretty good
33:40 thanks been pulling out my hair and I
33:43 don't have enough as it is
33:45 yeah exactly my I look much smarter now
33:48 because my forehead is so much larger
33:52 it's just the hair just keeps going
33:55 and then yeah
33:58 the glue's holding pretty good today
34:00 it's not it's not a total disaster
34:03 um
34:04 after the Apple launched Tim Apple went
34:06 on morning TV to talk AI Zuck went on
34:08 Lex podcast on VR yeah Zuck did one of
34:12 those yeah well whatever Apple
34:14 introduced wasn't very impressive our
34:16 howdy partner how you doing today
34:20 um
34:21 Zuck is in a little bit of a defensive
34:24 posture because Apple successfully they
34:26 didn't use the word metaverse once they
34:28 didn't use the word AI once and they
34:31 introduced some really revolutionary
34:33 technology and somehow mysteriously they
34:36 took a thing that is form factor wise
34:39 and from a technical perspective
34:40 essentially identical to every other
34:42 Goggle out there and they somehow
34:44 successfully redefined that as this new
34:47 thing called spatial computing
34:49 It's amazing And there's apparently 5
34:52 000 patents in in in those uh in those
34:54 classes
34:56 um so they're 3 500 bucks so they're
34:57 really expensive but but what they do
35:00 and the level of integration of AI and
35:02 machine learning and all of that stuff
35:04 within them is remarkable
35:06 the other thing that they do is they do
35:08 all their AI processing in the goggles
35:10 themselves they're not sending it off to
35:11 some other servers so they're going to
35:13 win the Privacy game I think they're
35:15 going to win the the uh well spatial
35:17 Computing that will be a new thing
35:19 that that they will have invented and
35:22 then
35:23 um
35:23 and then I think they're gonna they're
35:25 gonna win the um
35:27 like the the autonomous agents where
35:29 within the Apple ecosystem it will just
35:32 learn all about you and and you'll just
35:34 say hey oh you don't say Hey Siri
35:36 anymore you'll say Siri book me a trip
35:38 to Florida give me a five-day itinerary
35:40 and get me the cheapest tickets and make
35:42 sure that they're cancelable and it'll
35:44 just go off and do it for you I think
35:46 they're going to be brilliant at that
35:49 I need a notebook I've written too many
35:51 stick sticky notes with the info you've
35:53 shared go get a notebook they sell them
35:55 now that you can get them at like
35:57 drugstores you can get them you go to
35:59 Target or we call it Target because it's
36:01 fancier than Kmart used to be
36:05 um they sell them there there's uh
36:07 there's office supply stores they sell
36:09 them they sell pens
36:11 it's remarkable it's all AI generated
36:13 too
36:16 I am a smart ass by the way uh some
36:19 might call me an asshat
36:22 and give me another hat
36:23 I'll talk like a cowboy
36:26 all right I lost my place in the
36:28 questions again
36:34 trying to find where I was this is one
36:37 of my hobbies scrolling through Tick
36:39 Tock okay there's Emilio's wife talking
36:41 about the need for a notebook that's
36:43 really funny and awesome
36:44 I use creative
36:46 Tick Tock hat yeah
36:50 creative Fabrica spark for my AI artwork
36:53 it's free and it's easy cool
36:55 Built My Own autonomous agent on gpt4
36:57 but keep hitting token limits any tips
37:00 nope
37:02 uh if if anthropic so
37:06 if Claude 100K is available via API
37:09 which I assume it is because pose using
37:12 it I would use anthropics Claude 100K or
37:16 there's a
37:17 MPT 7B has a 65 000 token
37:22 a model called story something story
37:26 maker or story I don't know
37:30 look up MPT 7B 65k here I'll go do it
37:35 now for you I
37:37 m p t
37:40 Ash 7B 65k
37:46 or I did SK okay
37:49 mptb Beats gpt42
37:53 foreign
37:56 this is also commercially usable
38:00 so that's an option and then the Claude
38:05 Claude instant 100K is an option and
38:09 it's got much larger token limits the
38:12 other thing you can do if you're hitting
38:14 token limits is
38:16 um install Lang chain
38:18 and you'll have to do some programming
38:20 but basically you take large documents
38:22 and you break them up into into sections
38:24 and then you embed them and then you put
38:26 those vectors into a vector database and
38:30 it gets really geeky really quick but
38:33 you can either use larger token limit
38:36 models like the ones I just talked about
38:38 or you can install Lang chain and start
38:42 to learn about doing your old all your
38:45 data or you can wait
38:49 three months is my guess this is total
38:52 speculation
38:53 and Microsoft will roll all this GPT
38:57 stuff Within
38:59 um Office 365 and Google's Gonna Roll
39:02 barred across all of their office things
39:06 and I would be dramatically shocked if
39:08 you couldn't just point your those
39:10 things at your Cloud stuff and query
39:13 them so so it's coming
39:16 I currently work in QA but I want to get
39:19 in AI any advice on what routes I should
39:21 take
39:22 um
39:24 I
39:25 that's a sweet little heart hello thank
39:27 you
39:29 um
39:30 I think my advice for everyone in any
39:33 profession is
39:35 learn as much as you can about how AI
39:39 will be used to replace
39:42 your job as a QA person right so it's
39:46 going to be able to do things like
39:47 create testing regimens and in a lot of
39:50 cases if not all of the cases it'll
39:52 actually be able to go test them
39:54 um so what I would do is I would go to
39:56 chat and I would start writing your own
40:00 like like have it help you create
40:02 testing Frameworks just start start
40:05 using Chachi BT to augment your job and
40:08 amplify what you already do and figure
40:11 out how quickly can you build tools that
40:14 do what you do
40:16 because if you're able to do that with
40:19 that I guarantee you there's a couple of
40:21 punks sitting in their mom's basement
40:23 right now writing that tool so at some
40:25 point someone's going to release a tool
40:27 that's like the QA you know testing
40:30 button and I'll be able to upload my IOS
40:33 app on my Android app and go test this
40:35 and it will not only test it it will
40:37 analyze the code and it'll say okay here
40:39 are this three race conditions you have
40:41 here are the here are the 16 syntax
40:44 errors you have in The Code by the way I
40:46 fixed all the syntax errors
40:48 um I fixed two of the three race
40:50 conditions and you have a decision to
40:51 make on the third race condition do you
40:53 want to give up this feature give up
40:55 that feature or do you want to
40:56 completely refactor that section of the
40:58 app like that tool is coming it ain't
41:01 here yet but I I would strongly
41:04 encourage you to start to figure out how
41:07 to replace yourself in doing that you'll
41:10 get much much better at your job and
41:13 um
41:14 and you'll be able to identify when the
41:18 the tools are coming at you that um
41:22 are going to hit everyone else by
41:24 surprise
41:25 do I partner with others
41:27 um sure I I'm the CEO of a company
41:29 called storyvine so I'm happy to talk
41:32 about anything I'm in a
41:34 well I mean less of one for the first
41:36 three months of this year I was in a
41:37 just just say yes to everything mode now
41:40 now I have over committed pretty
41:41 significantly on a lot of stuff but
41:44 um I'm I'm very open to projects because
41:47 there's so much going on am I a robot uh
41:50 I am gpt5 thank you very much
41:53 you thought you thought gpt4 was cool
41:55 check this out
41:59 yeah yeah
42:02 all right
42:03 um hello from West Palm Beach Florida
42:05 hello hello Isabel favorite plugins I
42:09 haven't played with a ton of them I mean
42:12 the ones that are just sort of clear and
42:14 obvious to me like link reader and um
42:17 chat with PDF like PDF readers and and
42:19 Link readers are really powerful
42:22 that's just super cool but you can kind
42:24 of do that with just gpt4 connected to
42:27 the internet but I like I like those
42:29 plugins they're just very very specific
42:33 um
42:35 I mean I don't do a lot of trip planning
42:37 but if I did do a lot of trip planning
42:39 I'd like all of those
42:42 um I don't know right now I I haven't
42:43 used them enough I haven't used them
42:45 enough and there's a lot of them and
42:47 it's like like I said I mean
42:50 it is daunting enough getting your head
42:52 around what gpt4 makes possible much
42:55 less what gpt4 plus
42:59 260 micro applications that you can
43:02 wield within a context of some sort of
43:05 app you've conceived so
43:08 I've been watching all your videos and
43:10 reading your newsletter in the past two
43:13 days mind blowing thank you Jen that's
43:14 awesome oh yeah I've seen your comments
43:15 I appreciate the uh I appreciate the
43:17 engagement that's cool I'm glad it's
43:19 making a difference I think you said
43:20 last night didn't you say like you were
43:23 really hot you're like I'm really high
43:24 and this is blowing my mind
43:26 it blows your mind when you're that High
43:29 um I think there'll become black spots
43:31 in human knowledge when trading data is
43:33 selectively excluded
43:39 the training data being exclusive
43:43 selectively excluded kind of lives in
43:46 the conspiracy theory that two or three
43:48 Mega corporations are going to control
43:51 everything
43:53 um
43:55 because the the open source Genie has
43:57 been let out of the bottle because there
43:59 are so many open source models now and
44:02 understanding how to train these the the
44:05 the large data sets um is is much more
44:08 broadly understood now
44:10 it's possible
44:12 that that
44:14 um
44:16 that that can't happen now
44:19 could a country say it's illegal to use
44:22 open source large language models yeah
44:24 possibly
44:26 um will that stop people that really
44:27 want to use them no
44:29 so I thought
44:34 I I don't see that happening now but you
44:37 know listen every everything's up for
44:39 grabs because this is I mean here's the
44:41 deal about these Technologies these are
44:43 by far the most powerful Technologies
44:45 ever put in the hands of regular
44:47 everyday people like the the level of
44:50 power that you have in your hand hands
44:52 even with a safety guard rail gpt4
44:57 is significantly beyond what I think any
45:00 government is comfortable having their
45:02 people have access to and then if you
45:04 take the large language models or you
45:05 take off the guard rails
45:08 um there's a lot of people in power
45:10 there's a lot of power centers that are
45:12 about to be upended so I'll give you a
45:15 an example of of an area that I know
45:18 pretty well I I started an early agency
45:20 and I sold it to Omnicom so I know the
45:23 agency holding company model pretty well
45:28 um
45:30 I think that the billable hour business
45:33 model the time and materials model that
45:36 all holding companies are based on I
45:38 think that's about to blow up I think
45:41 clients are going to start demanding
45:43 fixed pricing which is going to destroy
45:47 the
45:48 Top Line and profits of holding
45:51 companies they're going to have to
45:53 completely reinvent their business model
45:56 um so when big Power centers get
45:59 disrupted like that you know powerful
46:02 people are going to do some [ __ ] to try
46:03 to control
46:04 to control us plebes so you know you're
46:09 not necessarily wrong that the big Power
46:11 centers are going to try to do that now
46:13 whether or not they can successfully do
46:15 that that's a different thing I I mean
46:17 the only way that you can really stop
46:20 this AI stuff in in my opinion if you
46:22 actually wanted to stop it
46:24 is throw a uh uh an EMP an
46:29 electromagnetic pulse throw some EMP
46:31 bombs that destroy all the computers but
46:34 they're not going to do that because
46:35 that destroys all the economies right it
46:37 it it
46:38 so I don't think it's going to go away
46:40 all right just like a marketing agency
46:42 will there be a need for an AI agency
46:45 teaching businesses how to use AI
46:47 certainly short term this is so so all
46:50 of these new technologies go through go
46:52 through the um
46:53 go through the same phases like I went
46:56 through this with the World Wide Web we
46:57 went through this with smartphones we
46:59 went through this with social media now
47:01 we're going to go through it with AI
47:03 so it initially starts out like it's a
47:05 fad it's a phase we're gonna get over it
47:07 oh the.com bubble burst the internet's
47:11 over we're never going to use the
47:12 internet now don't invest in the
47:13 internet it was all a waste of time
47:14 nothing to look at here and then it
47:16 [ __ ] transformed everything right
47:19 um same thing's happening here so so the
47:22 first phase is they dismiss it the
47:24 second phase is they try to destroy it
47:25 the third phase is they try to protect
47:27 from it right so universities using chat
47:30 GPT detectors what if the students use
47:34 it the right papers oh my God
47:37 well it's going to be in Microsoft Word
47:39 it's going to be in Google Docs
47:41 everyone's going to [ __ ] use this for
47:42 everything
47:43 so so so that's that the next phase is
47:47 we need a czar of AI oh you know how to
47:50 spell AI you're the VP of AI they put it
47:53 in a corner and they make someone go
47:54 figure this out for us and it's off to
47:56 the side and like social media like
47:59 we were talking about integrating
48:01 digital in 1996 and
48:05 a lot of big agencies and big companies
48:08 are just now starting to integrate
48:10 digital fully into business right it's
48:12 it's fully a part of business but but a
48:15 lot of them still have these digital
48:16 centers of excellence where it's off to
48:18 the side so you're going to absolutely
48:20 see that
48:21 but the tools are going to very quickly
48:23 I think much more quickly than
48:26 um than just digital did they're going
48:29 to inundate all of all the processes of
48:32 businesses because it's so much more
48:33 efficient it's so much more powerful so
48:36 I think short term probably for the next
48:39 three to five years you could make you
48:42 could have a very good business
48:44 um teaching businesses how to
48:46 incorporate Ai and not just how to
48:48 incorporate it how to deal with the
48:51 implications of it like I I think that
48:54 might be
48:55 an area there's sort of tactical teach
48:58 them how to use it but there's how do
49:01 you deal with if you're a Holden company
49:03 and you rely on the time and materials
49:05 business model
49:07 and you can con you can scenario plan
49:09 out it's not going to be that many years
49:11 until we can't do that anymore well what
49:13 do you [ __ ] do
49:15 do you go first do you say [ __ ] it we're
49:17 gonna we're gonna go to fixed pricing
49:19 and we're gonna beat all the other
49:21 holding companies to the punch that's
49:23 that's a possibility
49:25 so AI will be used as a tool not
49:27 something that will take over jobs well
49:29 so AI absolutely AI is a tool
49:32 Microsoft is branded their AI copilot
49:35 which feels like the right branding to
49:36 me like that's the right positioning it
49:38 feels like the right definition humans
49:40 are the pilot AI tools are the co-pilot
49:43 they happen to be incredibly [ __ ]
49:45 capable co-pilots like I guarantee you
49:48 that AI can can land a plane in wind
49:50 shear at some point maybe not right now
49:52 but they're gonna it's gonna be able to
49:54 land a plane in wind shear better than a
49:56 human pilot but
49:58 as long as human lives are at stake
50:00 human Pilots will still be in charge for
50:02 many years to come right it's going to
50:04 take many many years for that to happen
50:07 um
50:09 but it's a tool
50:11 AI will not take jobs
50:13 people using AI will take jobs and jobs
50:17 will transform
50:19 think anywhere where you've got
50:21 incredibly complicated processes and you
50:25 have job people's whose job it is to
50:27 deal with the complexity
50:29 of the thing so think Insurance
50:33 underwriting think
50:36 um legal research think
50:39 um
50:40 I don't know anywhere where you've got
50:41 incredibly complicated systems oh you
50:44 know like medical paperwork you know
50:47 patient records right now incredibly
50:49 complicated because the insurance
50:51 companies tell you you have to put in
50:53 all these codes right and that's mostly
50:56 manual and getting more digital but but
50:58 that's going to quickly move to fully AI
51:01 automated right
51:04 um so that's where you're going to see
51:06 the biggest disruption is jobs that are
51:09 ultimately
51:11 low value to the company but have to get
51:13 done right
51:15 so what's going to happen some companies
51:18 will just fire all those people right
51:19 and that's what will make the Nightly
51:21 News
51:22 the smarter companies will say oh
51:26 I don't have to pay these 10 000 people
51:29 to to to
51:31 manually read faxes anymore right I can
51:35 just have ai do that
51:37 what if I train them up to be super high
51:39 level customer service people
51:42 we get rid of our
51:45 offshore customer service call centers
51:47 because they're they're clueless anyway
51:49 like no matter how much we train them up
51:50 it's a horrible experience and they
51:52 never have the answers anyway we replace
51:54 them with AI and then we take the people
51:56 that were doing this low level
51:59 um you know document processing and we
52:02 train them to be high level customer
52:04 service people so the AI will take the
52:06 initial call you'll never be put on hold
52:08 you'll always be talking to a robot
52:10 that's way more capable than the current
52:12 customer service people are and it will
52:14 know when it needs to escalate your
52:16 issue to someone up here if it can't
52:18 solve it it'll escalate it so you'll
52:21 call someone will pick up the phone
52:23 you'll talk to them it'll be useful if
52:25 you have a truly unique problem you'll
52:27 get escalated immediately and you can
52:29 talk immediately so like
52:31 we don't know right now what the new
52:33 jobs are going to be but there's going
52:35 to be all sorts of new jobs created but
52:37 the short term two to three year window
52:40 where all of this automation happens is
52:42 going to be like radically disruptive
52:45 it's going to be really painful it's
52:46 going to suck
52:49 um how will this affect electronic
52:51 Discovery can it be used
52:54 for early case assessments yeah I think
52:57 so I I mean like I just did if you
52:59 weren't here earlier
53:01 I did
53:05 so let's go let's go get a
53:09 let's go get a legal brief I don't know
53:11 anything about the law legal
53:14 brief
53:16 um examples okay
53:19 whoa lots of them okay
53:22 um
53:23 I don't know are any of these long I
53:25 assume some of these are long
53:30 I don't know what this is like how
53:33 wait legal brief
53:35 um let's see legal brief
53:38 um
53:40 um let's do Supreme Court
53:46 um PDF let's see what we get there
53:53 is this a
53:57 this is bad interface Bing visit site
54:04 oh sorry
54:05 just looking at my nose okay is this
54:08 long
54:09 uh yeah it's pretty long
54:12 okay so I'm gonna do select all there
54:14 I'm going to copy now what I'm going to
54:16 do is I'm going to go to PO
54:18 so Poe's got this thing called Claude
54:20 instant 100K so this is a hundred
54:23 thousand token
54:25 um
54:27 100 000 token large language model so
54:30 it's like chat GPT except you can paste
54:33 in up to 75 000 words so I'm going to go
54:35 over here
54:37 and I'm gonna and I have to I have to
54:39 this is for the subscription version of
54:41 poe.com to get access to this but I'm
54:45 going to clear my my context history
54:47 then I'm going to say based on this
54:50 legal brief
54:52 colon
54:54 paste
54:55 so I just paste it in all of the brief
55:00 and then I'm gonna go please
55:03 summarize
55:05 um this
55:07 I guess case
55:09 let's say this document because I don't
55:11 know anything about law as if I were a
55:17 seventh grader writing a rip writing a
55:22 report
55:24 give me
55:26 mark down formatted
55:31 um
55:32 bullets
55:36 if appropriate
55:39 all right so now it just read that whole
55:41 document
55:44 and
55:47 didn't jump down to the bottom so we're
55:49 going to scroll all the way to the
55:51 bottom
55:51 we'll get there oh it's going to be so
55:54 good Jesus this was a long document
55:56 okay
55:57 so here's a summarized version of the
56:01 legal brief in bullet points ice IFC is
56:04 an international organization owned by
56:06 104 member countries in 1945 Congress
56:09 passed a thing I IFC was designated
56:12 designated as an international thing
56:15 Indian Nationals who live near an IFC
56:18 Finance power plant are claiming that
56:20 the plant caused them harm right so so
56:22 this wrote a summary of of that thing so
56:26 I don't know enough about the world of
56:28 Law and legal but the fact that I just
56:30 took a supreme court legal briefing and
56:33 summarized it that quickly yeah I think
56:36 that changes things
56:38 I could be wrong but but if if little me
56:43 could understand what the hell was just
56:45 talk going on there
56:47 um it looks like the the uh the the
56:49 number of people here are kind of
56:51 dwindling so I'll either go but like tap
56:53 the screen a bunch and um if you can
56:56 share this uh share this live
56:59 I don't if you know how to do that I
57:00 don't know how to do that
57:02 um just so we can get some more people
57:03 in here uh otherwise I'll just I'll just
57:04 take off
57:07 um openai has gotten about 75 of my
57:09 coding questions wrong I would say
57:11 um
57:12 go to GitHub copilot which is gpt4
57:15 that's that's dedicated to coding that's
57:17 probably going to be better and if
57:19 you're not using gpt4 I understand that
57:21 gpt4 is way better than
57:24 um GPT 3.5 you might also try Bard so
57:28 bard.google.com is really [ __ ]
57:32 horrible for it for writing for language
57:34 stuff but apparently it's strong at code
57:36 so I would also say give Bart a shot
57:41 um work with time 100 true not sure what
57:43 I said I know open AI but what are the
57:46 other sites oh so bing.com is
57:49 Microsoft's search engine but it also
57:52 has button you find a little button and
57:54 you push the little button it's so
57:55 exciting all right
57:58 so you go to bing.com and it has this
58:00 new little button and the new little
58:02 button is called chat
58:04 and chat is that is gpt4 that's like
58:09 chat GPT but it's gpt4 and it's free and
58:12 it's connected to the internet and then
58:14 it's got like a couple of modes it's got
58:15 more creative more precise so if you're
58:17 doing research go to more precise if
58:20 you're doing you know writing a
58:21 screenplay or the Great American novel
58:23 go to more creative so that's Bing
58:26 and then Poe
58:28 is the thing I just showed with that
58:30 legal document Poe gives you access to a
58:32 bunch of different models Sage is pose
58:35 model and and the company behind Poe is
58:37 quora right so the question and answer
58:39 site that that's been you know doing Q a
58:42 for decades they took all that data and
58:45 so Sage is is pose large language model
58:49 you got gpt4 you got Claude plus which
58:52 is anthropic's version of gpt4
58:56 you've got Claude instant 100K which is
58:59 the thing where you can put in giant
59:00 documents
59:02 um Claude instant is like GPT 3.5 and
59:05 chat gpt's GPT 3.5 the other thing you
59:07 can do on Poe is you can make your own
59:10 bots so like I made a Shakespeare bot
59:12 and a Yoda bot and a sling blade bot and
59:15 a Emily Dickinson bot so you can make
59:17 your own Bots which is kind of fun it's
59:19 it's a good way to practice your prompt
59:21 crafting and then the other two URLs
59:25 prompts.chat is just a document that
59:28 will teach you about prompting and then
59:29 gives you like hundreds if not thousands
59:32 of prompts that you can try like act as
59:34 a legal assistant or a paralegal act as
59:37 a novelist act as a mathematician which
59:42 surprisingly actually makes uh chat GPT
59:46 work better and then futurepedia is just
59:49 a directory of massive massive amounts
59:51 of tools AI tools so if you want to go
59:55 see all the video tools and the music
59:56 tools and the whatever it's there's just
1:00:00 there is right now
1:00:03 an infinite amount of things to explore
1:00:05 and so what I've personally done is I've
1:00:07 just kind of locked myself down to this
1:00:11 [ __ ] plus the image generation tools
1:00:13 that I like which right now is
1:00:14 mid-journey
1:00:16 um all the other stuff I'm just I don't
1:00:17 have the time to play with but you can
1:00:19 if you want to run down a rabbit hole
1:00:20 like I I do strongly suggest like if
1:00:23 you're a musician go find all of the
1:00:26 music tools out there figure which ones
1:00:27 are good which ones are [ __ ] and
1:00:29 then like just start learning that
1:00:31 corner of the AI industry
1:00:34 it will serve you well
1:00:36 are you a gamer on Steam at all no well
1:00:40 I'm a Sim race I like Sim racing it's
1:00:42 it's good for my ADD
1:00:44 um it's like left right you know Stomp
1:00:47 the gas
1:00:48 um but but any any sort of games that
1:00:51 require attention and learning and not
1:00:55 so good uh but I've I've not been on
1:00:57 Steam
1:00:59 um how do you see AI affecting education
1:01:00 schools in particular if they don't
1:01:03 um
1:01:04 if they don't quickly get their [ __ ]
1:01:06 together and stop resisting this and
1:01:08 start embracing it they're gonna quickly
1:01:10 find themselves out of
1:01:12 um out of students because well not
1:01:15 quickly these are Big institutions
1:01:17 they're going to take for [ __ ] ever
1:01:18 to change but
1:01:19 um
1:01:21 the the whole resisting that this is
1:01:24 coming is is a feudal resistance
1:01:28 um and and there's a real opportunity
1:01:30 for teachers
1:01:32 to reinvent how they engage with
1:01:34 students
1:01:35 drive a lot more critical thinking
1:01:38 um so rather than saying you can't use
1:01:40 chat GPT to to write a paper why don't
1:01:43 we bring chat EPT into the classroom
1:01:45 have it write a paper and then have the
1:01:48 students analyze what chatgpt wrote and
1:01:50 figure out what did it do good what did
1:01:52 it do wrong where did it hallucinate how
1:01:54 would you do it better really
1:01:56 demonstrate your knowledge of The Thing
1:01:57 by analyzing a thing that the tool made
1:01:59 and in doing that you'll get better at
1:02:01 using the tool right because these tools
1:02:03 are going to be everywhere
1:02:05 so so I think the effect is going to be
1:02:08 dramatic if you want to see
1:02:10 um a really uh cool thing so so this was
1:02:14 cool before AI so Khan Academy
1:02:17 um salcon's Khan Academy where you know
1:02:19 he's got this idea of flipping the
1:02:21 classroom where like you do the learning
1:02:22 part out of the classroom you do the
1:02:24 engaging part in the classroom like I
1:02:26 think we're moving there
1:02:28 they got access to gpt4 back in August
1:02:31 of 22.
1:02:32 so they spent six or seven months
1:02:35 refining the experience and they created
1:02:38 this thing on Khan Academy called
1:02:39 conmigo which is a clever name but it's
1:02:41 basically a teaching assistant that will
1:02:44 use Socratic reasoning like if you just
1:02:47 ask it to answer a question for you it
1:02:49 won't it will query you and go back and
1:02:51 forth with you and teach you the
1:02:53 concepts and it won't move on until it
1:02:55 has done you have demonstrated to it
1:02:57 that you understand the concepts sitting
1:03:00 underneath the thing so so I think the
1:03:03 um I think the impact on education is
1:03:04 going to be quite profound and I think
1:03:07 Educators that are resisting it need to
1:03:09 get their heads out of their asses
1:03:11 um have I run into any tools for
1:03:13 instructional design such as creating
1:03:15 how to's
1:03:18 gpt4 let's let's go to the video the
1:03:21 masterful at uh instructional design
1:03:25 I've done a lot of it but I don't know
1:03:27 [ __ ] about it
1:03:29 um
1:03:30 I I don't know [ __ ] about it as as a as
1:03:33 a profession but that's one of the
1:03:36 things that that chat ept's [ __ ]
1:03:38 masterful at so here's what we're gonna
1:03:40 do I
1:03:42 want you to act as
1:03:46 an instruction
1:03:48 instructional designer
1:03:52 and help me design
1:03:56 some course work
1:04:00 the first
1:04:02 thing I need to know are
1:04:06 um
1:04:07 no
1:04:09 is
1:04:10 what are the main
1:04:14 Frameworks
1:04:16 for creating instructional
1:04:21 materials let's just call it that
1:04:27 creative creating them requires a
1:04:30 systematic approach
1:04:32 so there's the addy model there's the
1:04:34 Sam model successive approximation model
1:04:37 well of course there's that one the Sam
1:04:39 model everybody knows that
1:04:41 well you know dick and Carrie right they
1:04:44 came up with a model
1:04:45 Bloom's taxonomy
1:04:47 you know I haven't played with Bloom's
1:04:49 taxonomy in like a week
1:04:53 there's Kirkpatrick's four Level
1:04:55 Training evaluation okay it apparent
1:04:58 there's a lot more of these than I had a
1:05:00 [ __ ] clue about so there's nine or
1:05:02 there's seven of them here Ganges nine
1:05:05 events of instructions I'm gonna say
1:05:07 okay so I'm gonna say okay let's go with
1:05:10 gagny
1:05:13 period
1:05:14 um I want to teach a one hour
1:05:21 generative AI
1:05:24 overview
1:05:26 for people in the marketing department
1:05:33 that will
1:05:37 that will move them from
1:05:41 fear to curiosity
1:05:46 please give me
1:05:50 that outline
1:05:54 sounds like an interesting session
1:05:55 that's one of the things about chat GPT
1:05:57 is it it responds in a very human kind
1:06:00 of way I find myself like
1:06:03 um acknowledging it like hey you did a
1:06:05 good job there
1:06:07 um I I find myself using please and
1:06:09 thank you as if I'm talking to just like
1:06:10 a really capable intern
1:06:12 but okay so what's it done gain
1:06:15 attention five minutes start the session
1:06:17 with an intriguing question or a
1:06:19 statement such as imagine a world where
1:06:21 AI can generate creative content for
1:06:23 your marketing campaign show examples
1:06:24 okay that's cool
1:06:26 inform Learners of the objectives we're
1:06:29 going to State the objectives of what
1:06:31 you're about to learn
1:06:32 stimulate recall of a prior learning
1:06:35 um what do you already know about AI in
1:06:37 our marketing marketing department that
1:06:39 that the boss said we're not allowed to
1:06:40 use chat GPT right whatever present the
1:06:43 content and then the content
1:06:45 introduction case studies benefits
1:06:47 challenges provide learner guidance
1:06:49 explain some possible first steps
1:06:52 illicit performance
1:06:54 provide feedback assess performance
1:06:56 enhance retention and transfer right so
1:06:58 there's the nine steps of the gagny
1:07:00 approach which I knew nothing about they
1:07:02 sound reasonable to me they've blocked
1:07:05 out the time for the hour
1:07:07 and I'll just say
1:07:09 um What If This Were a three-day
1:07:14 session what might the agenda look like
1:07:19 boom
1:07:23 expanding a one-hour session into a
1:07:25 three-day session allows for more
1:07:26 in-depth exploration of generative AI
1:07:29 day one
1:07:31 so so it's doing the same nine hour or
1:07:35 the nine segment gagne approach but it's
1:07:38 just splitting it into three days and
1:07:39 expanding it
1:07:44 case studies and Hands-On session
1:07:47 so
1:07:50 now are there other tools there might be
1:07:54 other tools for that are specialized to
1:07:58 um
1:08:00 to instructional design and if you want
1:08:03 to find those go to futurepedia.io and
1:08:05 just look type in instructional design
1:08:07 there might be some created for it what
1:08:10 you'll probably find is that you
1:08:12 probably know more like the tools that
1:08:14 are going to be dedicated to this might
1:08:16 not be all that exciting I I would go
1:08:19 teach yourself how to do it with gpt4
1:08:22 okay enough of mentioning creative
1:08:25 fabric of spark this isn't don't spam in
1:08:28 here please leonardo.ai I think is
1:08:30 really a web front end to stable
1:08:33 diffusion
1:08:35 um I think it's still in preview though
1:08:39 leonardo.ai but I would I would
1:08:41 encourage you if you want to do image
1:08:43 stuff I would encourage you to
1:08:46 um get on the wait list for this
1:08:49 because it's um
1:08:56 cold please
1:09:03 so so it gives you lots of different
1:09:06 models so you can create images in these
1:09:08 different models across the top
1:09:11 like here's a vintage photography style
1:09:13 so I want to generate with this model
1:09:16 I don't want to tell you what role I am
1:09:19 I'm a professional
1:09:22 what ah pop-ups make me
1:09:28 right so I'm going to say a um
1:09:32 70s muscle
1:09:36 car oops not a 702 muscle car 7ds muscle
1:09:40 car
1:09:41 on a winding Mountain Road
1:09:47 you know going fast and so I so I picked
1:09:50 a model
1:09:52 that does vintage photography so this
1:09:55 look should look like some vintage
1:09:57 photograph of a 70s muscle card doing
1:10:00 something
1:10:03 um
1:10:04 I think it's decently fast you're
1:10:06 currently on the free plan yeah there
1:10:08 you go so there's there's a bunch of
1:10:12 you know
1:10:14 cars on Windy mountain roads that one
1:10:17 looks pretty good
1:10:19 but I don't like that color so I want to
1:10:22 do
1:10:24 um 70s
1:10:26 lime
1:10:29 green
1:10:31 muscle car
1:10:33 on a on a Mountain Road going fast I'm
1:10:37 going to say
1:10:39 um cinematic motion blur
1:10:44 so sometimes adding things like that
1:10:45 will help but anyway there's lots and
1:10:47 lots of other things that Leonardo does
1:10:49 you can easily train if you've got
1:10:51 photographs like I'm a photographer so I
1:10:53 uploaded a bunch of my sky photographs
1:10:54 to it it learned them and now I can just
1:10:57 call that that style
1:11:00 so let's let's get our green car so now
1:11:03 here's our green car right so Leonardo
1:11:05 is really good I think it's I think it's
1:11:07 awesome that's a little weird looking
1:11:10 that sort of looks like a GTO Judge
1:11:14 that looks like an old Mustang it looks
1:11:16 like a 7D Mustang 69 Maybe 70.
1:11:20 that sort of looks like a cross between
1:11:22 a Camaro a judge and a Mustang so anyway
1:11:26 um yeah I think it's good good call good
1:11:29 call on the tool
1:11:32 all right how I feel about faceless
1:11:34 videos using AI
1:11:36 how do I feel about faceless videos
1:11:38 using AI you mean like text videos I
1:11:40 don't quite know what you mean I think
1:11:42 all videos are going to use AI I have a
1:11:44 video uh technology company that has
1:11:47 human beings tell their stories and and
1:11:50 um
1:11:50 I am I am deeply exploring all of the
1:11:53 Avenues of AI including
1:11:56 um what if we didn't need people which
1:11:57 my initial
1:12:00 response to that was go [ __ ] yourself Ai
1:12:02 and then the more I thought about it the
1:12:04 more I thought
1:12:05 we're gonna have a competitor that comes
1:12:07 up against us like where we go into a
1:12:09 client we can say Hey you could have
1:12:11 like 20 of your employees talk about how
1:12:13 awesome it is to work here someone's
1:12:15 going to come along with a product that
1:12:16 says I can take a pic picture of your
1:12:19 employees their their employment record
1:12:20 and I can generate
1:12:23 a video per person
1:12:26 um talking about how awesome it is to
1:12:27 work here and then they just all they
1:12:29 have to do is sign off on it yes I'm
1:12:31 okay with you having my likeness say
1:12:34 that it's great to work here
1:12:36 that's going to be one of my competitors
1:12:38 so
1:12:39 I should probably figure out how to do
1:12:41 that [ __ ] so
1:12:43 it here's the thing it's almost
1:12:45 irrelevant how I feel about it this
1:12:48 stuff is coming whether you want it or
1:12:50 not I think that's the thing that that
1:12:51 maybe why this channel exists is that
1:12:54 the disruption that this generative AI
1:12:57 stuff is going to cause is coming
1:12:58 whether you want it or not what I'm
1:13:00 choosing to do is stand on the surfboard
1:13:01 on the tsunami rather than be on the
1:13:05 Cozy beach cottage which is where where
1:13:07 we all are in our current jobs right now
1:13:10 which is terrifies terrifying how can I
1:13:12 beat AI detection programs there's lots
1:13:14 of ways to do that there basically it's
1:13:17 basically you you take something that
1:13:18 was generated with chat EBT you throw it
1:13:20 into something that rewrites it and
1:13:22 rewrites it in a way that's not
1:13:23 detectable but that's a war that's going
1:13:25 to go back and forth here's the deal
1:13:27 I think that's a very short-lived thing
1:13:29 if you're dealing with it you're going
1:13:31 to need to find ways I don't know what
1:13:33 the tools are because I don't give a
1:13:34 [ __ ] about them because when Microsoft
1:13:36 includes generative AI in Office 365 and
1:13:40 then uh you know in Google does it in
1:13:43 Google Docs
1:13:44 you're it's gonna it's just gonna be in
1:13:47 everything like those those writing
1:13:48 detection tools are going to be
1:13:51 um just very short-lived in my opinion
1:13:54 um who said that quote hey I want to
1:13:55 take your job but someone using it will
1:13:57 Paul rateser
1:13:59 r-o-e-t-z-e-r from the marketing AI
1:14:02 Institute you should follow him on
1:14:04 LinkedIn
1:14:05 um he posts good posts and then you
1:14:07 should listen to his podcast they've got
1:14:08 a weekly podcast where they sort of
1:14:10 catch up on the week of AI what's cool
1:14:12 about Paul
1:14:14 and the marketing AI Institute it's like
1:14:17 five or six years old so he's been on
1:14:20 this train since like the early days of
1:14:22 like gpt2 gpt3 like he's seen this
1:14:24 coming all of a sudden you know since
1:14:27 chat EBT you know he's he's way ahead of
1:14:30 the curve the thing about Paul and what
1:14:33 they talk about there is I just like
1:14:36 every time I think huh I kind of think
1:14:38 it's going to go in this direction I'll
1:14:39 get on the podcast and he'll say I think
1:14:42 it's going to go in this direction like
1:14:43 I just I feel like I'm personally very
1:14:45 aligned
1:14:47 um just in terms of kind of analyzing
1:14:50 what's happening with what they're doing
1:14:51 there so I I think they're he's a really
1:14:54 good guy to follow
1:14:56 I'm looking for an AI to create gauge
1:14:59 graph like fuel gauge
1:15:02 um
1:15:02 I assume we can just go back to Leonardo
1:15:05 well let's go to well yeah we'll do
1:15:08 we'll do Leonardo
1:15:09 um
1:15:10 um let's see classic
1:15:19 um fuel gauge
1:15:25 face
1:15:26 with
1:15:28 black
1:15:32 backup and yellow
1:15:36 tick marks
1:15:40 um modern
1:15:42 well I call it classic so let's see
1:15:45 um updated design I don't know whatever
1:15:49 that's called design
1:15:51 we'll see what that generates
1:15:54 like any of the any of the image
1:15:56 generation tools you know so so what I
1:15:59 would start teaching yourself is
1:16:00 prompting because the the like where
1:16:04 we're heading is what the what you
1:16:06 effectively have to do with these tools
1:16:08 is you have to be good at critical
1:16:09 thinking and good at asking for stuff so
1:16:11 there's some fuel gauges right
1:16:15 um if that's not what you want then you
1:16:17 basically just describe what you do want
1:16:19 now you'll notice that the numbers and
1:16:21 the words are messed up the these tools
1:16:24 are not good with with words and numbers
1:16:26 yet
1:16:27 like they're out of order but but you
1:16:29 know you you get the the basic idea here
1:16:34 right
1:16:35 so
1:16:36 what did that take me 30 seconds so go
1:16:40 just go start experimenting that the
1:16:42 thing about these tools is they're so
1:16:44 much more capable than than
1:16:47 I think they are like every time I try
1:16:50 something new where I'm like well it
1:16:51 probably doesn't oh [ __ ] it does that
1:16:54 good too
1:16:56 best used for the link reader plugin
1:16:59 um
1:17:01 have it go read the pages of your
1:17:05 website and give you SEO well there's
1:17:07 SEO analysis tools out there
1:17:10 um what I've been using it for is I'll
1:17:13 take scientific white papers or I'll
1:17:15 take like a really complicated article
1:17:17 and I'll turn that into a Twitter thread
1:17:19 or I'll turn that into a LinkedIn post
1:17:22 um where it's not writing something new
1:17:24 from scratch it's like read this long
1:17:26 document and then summarize it for me in
1:17:28 some interesting way and as as I get the
1:17:30 prompt styled in like the I have a I
1:17:32 have a Twitter thread maker right now
1:17:34 that's [ __ ] brilliant like it's just
1:17:36 it's not what I did isn't brilliant but
1:17:39 the output of it is consistently like
1:17:43 95 percent complete like I don't need to
1:17:45 edit much in it
1:17:47 it's and you know and it takes me
1:17:50 30 seconds rather than having to read it
1:17:52 for
1:17:54 20 minutes to four hours because of my
1:17:57 80d and then go in and highlight the
1:17:59 little pieces of it
1:18:01 what were the actual Salient points here
1:18:03 like like I did a um
1:18:06 I did a 15 segment Twitter thread on
1:18:08 Mark and drayson's you know AI might
1:18:11 save us thing it was just it was just
1:18:13 brilliant at it
1:18:16 all right won't this only give you
1:18:19 results through
1:18:20 a previous year well so the the core
1:18:24 model for chat GPT stops at September 21
1:18:26 but gpt4 connected to the internet or
1:18:29 bing.com connected to the internet
1:18:32 um do both so there they are it does not
1:18:36 stop there
1:18:39 how do I see AI affecting education we
1:18:41 covered that one
1:18:46 instead of organizing of course I'd
1:18:48 prefer Zoom class where folks
1:18:51 can ask questions and try ideas that's
1:18:53 not a bad idea I might do that Amelia's
1:18:55 wife I gotta figure out how to set that
1:18:57 up but I can go ask chat GPT to help me
1:19:00 set that up
1:19:01 I did a I had a
1:19:03 um a podcast for a year where I used
1:19:06 stream yard
1:19:08 I just read that paper today yeah it's
1:19:10 good paper salvation like cleaning the
1:19:13 planet from cancer
1:19:15 um
1:19:15 that kind of stuff's coming
1:19:17 genetic diseases cancer it's going to
1:19:20 start solving it's going to start so Sam
1:19:23 Altman was being interviewed once and he
1:19:25 said he thinks it's plausible that in a
1:19:26 relatively short amount of time
1:19:29 we're going to have a single year where
1:19:32 we accomplish as much scientifically as
1:19:35 we have in the past 500.
1:19:38 and and if you want to do an interesting
1:19:40 exercise go to gpt4 and have it make you
1:19:42 a table of the top 30 or 40 scientific
1:19:46 achievements in the past 500 years and
1:19:49 just look at all the stuff on that list
1:19:51 and you'll be like oh yeah oh that was a
1:19:53 big one oh that was a big one oh the
1:19:54 polio vaccine right you just go down
1:19:56 through them
1:19:57 flight
1:19:59 um and then imagine that all happening
1:20:01 within a one-year period
1:20:04 like crazy
1:20:06 how can you vectorize large chunks of
1:20:08 text and pass pass it as context to gpt4
1:20:12 so you have to use something like Lang
1:20:14 chain the good news about Lang chain and
1:20:17 these tools is that the developer
1:20:18 Community is is making it easy to do
1:20:21 that like I think you can install Lang
1:20:23 chain now in like I don't know 20 lines
1:20:26 of code or like it's some ridiculously
1:20:28 stupid small amount of code but you'll
1:20:30 probably need to set up a virtual
1:20:31 machine and and then you know you'll
1:20:34 you'll have a module that will chunk up
1:20:36 your large pieces of data it will embed
1:20:39 it it will vectorize it then it'll put
1:20:40 those vectors in a vector database and
1:20:42 then you'll be able to just interact it
1:20:44 with it like chat GPT
1:20:46 um I saw a thing on Twitter like two
1:20:49 days ago that someone's created a
1:20:52 a company that is basically putting a
1:20:54 front end on top of Lang chain
1:20:57 um
1:20:58 so Lang chain is the thing I would go
1:21:01 um explore there it's but it's going to
1:21:03 be geeky is there any AI that can grade
1:21:06 my algebra students worksheets probably
1:21:08 just chat GPT so
1:21:12 um yeah it it knows math and and chat
1:21:15 gbt with the Wolfram Alpha plug-in like
1:21:18 if it's not doing accurate grading of
1:21:21 the algebra worksheets
1:21:23 um oh I guess one question I have is if
1:21:26 it if you need it to like
1:21:28 scan the visual and then grade it there
1:21:31 you're gonna have to wait for it so
1:21:34 gpt4 is multimodal it can it can use
1:21:38 pictures but the current implementation
1:21:40 of it it can't so you're just gonna have
1:21:43 to wait a little bit if you if you
1:21:45 wanted to like look at like handwritten
1:21:47 algebra
1:21:49 um but if it's if it's the kind of thing
1:21:50 where it's in PDF form where you can
1:21:52 just pop it in
1:21:54 um you can probably do that right now
1:21:55 and and there may be a tool like I would
1:21:57 say go to futurepedia.io and look up
1:22:00 document analysis or or like
1:22:03 um computer vision or something like
1:22:05 that there's probably apps right now
1:22:07 that can do that
1:22:09 S3 photographer
1:22:11 and already trying to learn AI
1:22:13 applications loving Firefly yeah Firefly
1:22:15 is amazing and so is generative fill in
1:22:18 Photoshop if you have not seen
1:22:19 generative fill
1:22:22 let's see oh no that's not a fun one
1:22:25 hang on
1:22:26 let's go get this one this is a fun one
1:22:30 so here we have some dude
1:22:33 some some sailor dude in Times Square
1:22:36 and so I can just select this bottom
1:22:39 white area
1:22:41 oh sorry
1:22:43 so there's the dude in Times Square this
1:22:45 is something I made in mid-journey
1:22:47 and so in this so to get to this if
1:22:51 you've got Photoshop
1:22:54 um Creative Cloud you go to Beta apps
1:22:57 and you download the Photoshop beta and
1:23:00 and it's got this in it so I just
1:23:02 selected the bottom half of this image
1:23:04 and it pops up this contextual menu here
1:23:06 and on the left this button says
1:23:08 generative fill and so I can put in here
1:23:12 um Lakers
1:23:14 basketball shorts
1:23:18 and I'm going to say generate
1:23:20 and it should put shorts on this dude
1:23:25 that are Lakers shorts
1:23:31 now it didn't do it but like it gives us
1:23:34 three options let's see
1:23:36 they're not Lakers okay so it doesn't
1:23:38 know what the Lakers are so we'll say
1:23:39 generative Phil
1:23:41 will do
1:23:42 um yellow
1:23:45 basketball shorts
1:23:48 oops oh dang it yellow basketball
1:23:52 shorts with
1:23:55 purple
1:23:57 stripes
1:24:01 and so it's going off it's thinking it's
1:24:03 thinking
1:24:05 we're thinking here paintball
1:24:08 give him some yellow shorts oh those are
1:24:11 funny where's my purple stripes there's
1:24:13 no purple in there
1:24:17 yeah we don't we don't have I could keep
1:24:19 messing with that but that's funny and
1:24:21 then
1:24:22 you know up in the sky
1:24:25 we could do um
1:24:28 we could do generative fill
1:24:31 um large
1:24:34 um Godzilla
1:24:37 attacking City
1:24:42 and it's thinking it's thinking
1:24:47 genius takes time like at least 30
1:24:50 seconds
1:24:52 and it might not be genius it might be
1:24:54 funny all right there's no Godzilla
1:24:56 there
1:24:57 let's see if it gave us a Godzilla it
1:24:59 did not oh there's one there's a monster
1:25:02 so it's on top of the sign
1:25:05 so anyway that's generative Phil if you
1:25:06 haven't played with generative Phil it's
1:25:08 just amazing and then
1:25:09 um who was it that came out with it is
1:25:11 it stability
1:25:13 I think stability came out with uncrop
1:25:17 so let's go play with that
1:25:21 um stability uncropped
1:25:25 so stability has a a site called clip
1:25:28 drop
1:25:30 so it's clip
1:25:35 oh dang it clip drop
1:25:39 dot Co
1:25:41 and then one of the and then you go
1:25:43 slash it's clipdrop.co slash uncrop
1:25:47 which is a good brand name
1:25:50 and then what you do it's it's it's
1:25:52 exactly like what we just looked at but
1:25:54 this is it's a more simple
1:25:55 implementation it's pretty smart
1:25:58 um
1:26:00 let me go to finder
1:26:05 what's this picture oh that's the video
1:26:07 I made okay here's one so I'm gonna just
1:26:10 grab a picture here and I'm gonna drag
1:26:13 it yeah I'm gonna drag it into the
1:26:16 little window
1:26:18 and so it gives me that and so now I can
1:26:21 grab these blue handles
1:26:23 and I can just
1:26:25 you know expand them to like if I want
1:26:29 this little version of me to be
1:26:33 in a city I can just sit hit next
1:26:37 and it will now
1:26:39 uncrop that cropped image
1:26:48 oh that's cool that's really cool so
1:26:52 there's one
1:26:53 like I didn't have to type anything that
1:26:55 put a weird hand there
1:26:57 but there's another one and also with a
1:26:59 weird hand
1:27:00 that's not a weird hand like like it
1:27:03 seems to like it does a really good job
1:27:06 of inferring that this is like a
1:27:08 futuristic sci-fi world like that one's
1:27:13 really good
1:27:15 right so
1:27:19 I don't know people
1:27:21 everything's changing everything is
1:27:25 changing it's all changing you should be
1:27:29 excited and you should be terrified you
1:27:32 should be both if you're not pee in your
1:27:35 pants then you should be pooping them
1:27:38 all right Jarvis from Iron Man is coming
1:27:40 I agree yes balance balance between
1:27:42 excitement and fear
1:27:45 um is there any AI that can grade my
1:27:46 algebra I think we talked about that
1:27:48 where is he reading comments from way
1:27:51 way way earlier I'm very behind because
1:27:54 I sit on these things for like two hours
1:27:56 I tend to sit on these for two hours
1:27:58 until my butt gets numb and I dehydrate
1:28:01 myself because it's just how my
1:28:03 personality works and I also try to
1:28:05 answer everyone's questions so
1:28:08 you know that's just the that's just the
1:28:10 shakes I'm an old guy on Tick Tock if
1:28:13 you're expecting me to behave like a
1:28:15 twerking 20 something it ain't gonna
1:28:18 happen I'm a bitter gen xer so [ __ ] it
1:28:21 this is how it works this is how we roll
1:28:24 here at the AI learning lab
1:28:27 mm-hmm
1:28:30 all right
1:28:33 oh it just jumped me to the bottom L.A
1:28:35 Lakers maybe yeah maybe it'll do that
1:28:37 but I don't know ain't worth it you can
1:28:39 go figure out the uh how to get the
1:28:41 actual LA Lakers it's yeah that would
1:28:44 probably do it but I don't know maybe
1:28:45 not who knows
1:28:49 I'm here to show the concepts you guys
1:28:52 can go do the Brilliance my level of
1:28:55 Brilliance is
1:28:56 is thin I'm trying to find where I was
1:28:59 in the questions which this is one of my
1:29:01 hobbies
1:29:02 when Tick Tock jumps me to the bottom oh
1:29:05 I'm too far I went too far I went too
1:29:07 far
1:29:08 open AI he's gotten about 75 percent of
1:29:11 my coding questions wrong boy I'm way
1:29:13 far back hold please
1:29:28 what was that article you showed the
1:29:30 other day about the biggest Week in Tech
1:29:33 AI history
1:29:37 it might have been
1:29:39 at my newsletter every day AI
1:29:46 because we had a
1:29:49 we had an issue
1:29:53 [Music]
1:30:05 AGI timelines so issue 10 was about
1:30:10 a category five week of AI announcements
1:30:12 and then the following week was also a
1:30:14 really big one
1:30:17 um
1:30:19 I I mean what's been what's been
1:30:21 happening is that like I I just I don't
1:30:24 think there's been a week where there
1:30:26 hasn't been a significant amount of of
1:30:28 uh
1:30:30 advancement so I don't I don't remember
1:30:32 what article it was sorry about that yo
1:30:33 man I have a project that I've been
1:30:35 working on and most of it is still hand
1:30:37 written how can AI fix that just go this
1:30:41 is why I have these here
1:30:43 so I I did this I'll tell you a cool a
1:30:46 cool thing this was another one of my
1:30:47 epiphanies
1:30:48 so go to just go start using the tool
1:30:52 um and start taking your handwritten
1:30:53 stuff and and
1:30:55 start collaborating with chat GPT so I I
1:30:58 had created a card game about aliens
1:31:02 um
1:31:03 like four or five years ago and because
1:31:05 I have 80d like I went through the sort
1:31:08 of the burst of of the idea and and
1:31:10 basically I got it as complete as the uh
1:31:13 for for as long as the adrenaline lasted
1:31:16 right so I think I had like 45 cards
1:31:19 identified like there were aliens and
1:31:20 there were spaceships and there were
1:31:22 areas like Area 51 and S3 and all you
1:31:24 know all that sort of [ __ ]
1:31:26 and then there were like government
1:31:27 cover-ups there was all sorts of stuff
1:31:29 right so I had I had like 45 cards
1:31:31 written just the title of them I had
1:31:33 about a third of them descriptions
1:31:35 written
1:31:36 and then
1:31:38 it all it hit me in putting that
1:31:41 together that doing the research to like
1:31:44 write all of the cards I would need was
1:31:46 a daunting task and I was like oh [ __ ]
1:31:48 it
1:31:49 and so when I started playing with chat
1:31:51 EBT I'm like I wonder if it can do
1:31:54 if it can help me with this so I went
1:31:57 into my Excel thing with my you know
1:32:00 radically incomplete list of cards and I
1:32:02 copied and pasted it and I described
1:32:04 what the card game was and I said here's
1:32:07 the beginnings of a list of cards and I
1:32:09 pasted those in there and I said now
1:32:12 give me
1:32:13 15 categories of cards that could be
1:32:16 part of this game and Boom Out came the
1:32:18 15 categories and then I said now now
1:32:22 um like give me
1:32:24 a bunch of you know government locations
1:32:27 right as as cards with the description
1:32:29 and what happened there boom out they
1:32:31 came within like 60 minutes I had 300
1:32:37 well-written fully researched cards for
1:32:40 this card game
1:32:42 like and and then and then I even put
1:32:45 one of the cells and I you know I pasted
1:32:47 it all back into uh Google Sheets and
1:32:50 and I did an open Ai call in one of the
1:32:52 cells so I used the description of the
1:32:54 card to go to to Dolly and generate the
1:32:57 image for the card now Dolly sucks so
1:32:59 the images sucked but like I got a proof
1:33:02 of concept of taking like a poorly
1:33:04 written poorly researched project and
1:33:07 within 60 Minutes I totally blown the
1:33:09 doors off it
1:33:14 I'm a gpt6 model very nice to meet you
1:33:17 gpd5 here
1:33:19 all right hitting token limits build
1:33:22 your own
1:33:23 exactly there someone just I just saw
1:33:27 that this past week
1:33:29 um some group of researchers
1:33:32 um has created a model that's got a 5
1:33:34 million token limit so so basically you
1:33:37 could probably put in all your company's
1:33:39 data or you know some large chunk of it
1:33:41 um they said it's not very good right
1:33:44 now but it can take five million tokens
1:33:46 so
1:33:47 do you have a like list of shortcuts or
1:33:51 blog anything yeah everyday AI so it's
1:33:54 if you go to
1:33:56 um
1:33:57 everyday AI news on Twitter
1:34:00 There's the link to the subscription
1:34:02 there it's free
1:34:05 where's my Twitter where's my Twitter I
1:34:08 lost my Twitter
1:34:12 so profile
1:34:15 so if you go here wait
1:34:19 see the uh
1:34:23 right there at everyday AI news is the
1:34:26 Twitter handle go there and that's we do
1:34:29 we do a weekly
1:34:31 um
1:34:33 wrap up of the the big important stuff
1:34:35 going on in AI
1:34:37 um also Rachel Woods here on on she's
1:34:40 got a newsletter called the AI exchange
1:34:42 that's pretty good
1:34:44 um there's a bunch of them just start
1:34:45 following people on Twitter that that
1:34:46 are into this AI stuff you'll get up to
1:34:48 speed pretty quick is there a business
1:34:50 opportunity to teach small businesses
1:34:52 how to use it yeah I think there is how
1:34:54 would you go about it
1:34:55 um start giving it away for free teach
1:34:57 yourself teach yourself enough
1:35:00 about how to use these tools to do if if
1:35:03 you're good at marketing then you can
1:35:06 Target like you know
1:35:08 companies doing marketing stuff if
1:35:10 you're good at process you can have it
1:35:12 do that so I'd say figure out what
1:35:13 you're good at learn how to use the
1:35:15 tools and then give away
1:35:18 three engagements you know reach out to
1:35:20 people that you know that have small
1:35:21 businesses and say hey can I come in and
1:35:23 teach you how to do this for free so I
1:35:25 can just get practicing and then once
1:35:27 you do it for free then you just come up
1:35:28 with a price
1:35:29 500 bucks 10 000 bucks whatever it is
1:35:32 what's a good business to start using AI
1:35:34 the one I just described is one
1:35:37 um
1:35:40 like right now things are changing so
1:35:42 fast that I wouldn't get too married to
1:35:45 any solution or any
1:35:47 um tool right now because things are
1:35:49 moving so fast so I think I think the
1:35:51 the best
1:35:52 thing to do in business right now is
1:35:54 just
1:35:55 figure out how to solve an existing
1:35:58 problem faster using generative AI
1:36:01 because that's that's going to be the
1:36:04 like every business is going to have to
1:36:06 level up like everything's going to get
1:36:08 more efficient it's going to do this so
1:36:10 just take something that you know and
1:36:11 figure out how to do it better faster or
1:36:14 cheaper using Ai and then start putting
1:36:16 that out in the world and just you know
1:36:18 like keep iterating because
1:36:21 something that seems remarkable today is
1:36:24 going to see it's going to be a big yawn
1:36:25 six six months from now so don't get too
1:36:27 married to anything but start putting
1:36:29 stuff out in the world and start talking
1:36:31 about it on LinkedIn like as you learn
1:36:33 stuff like
1:36:35 seriously I would start like join I
1:36:38 created a thing called the AI Salon the
1:36:40 salon.ai go join that come hang out with
1:36:44 us we we meet every other week we've got
1:36:45 a Discord channel one of the women that
1:36:48 that started with
1:36:50 um the AI Salon she put that she was an
1:36:53 organizer of the AI salon on her
1:36:55 LinkedIn profile and got a job a week
1:36:56 later doing AI
1:36:59 um
1:37:00 and she didn't know much about AI before
1:37:03 that like it was a new thing so like
1:37:04 start talking about hey I'm teaching
1:37:06 myself chat gbt you can make money
1:37:11 um I'm starting to teach myself
1:37:13 um chat TBT and here's what I learned
1:37:15 and here's you know I tried this and it
1:37:17 didn't work but I tried this and it
1:37:18 worked great all that sort of stuff
1:37:21 how would I generate passive income I
1:37:23 don't know I I mean like like all the
1:37:26 passive income stuff most of those
1:37:28 people are making money teaching you how
1:37:30 to make passive income that's how
1:37:32 they're making their money not on
1:37:33 passive income they're making their
1:37:35 money on teaching you how to right it's
1:37:37 all [ __ ] it's you know you got my
1:37:39 money today make money while you're
1:37:41 asleep
1:37:43 it's all [ __ ]
1:37:45 um
1:37:47 if you can find a passive income
1:37:50 business
1:37:52 like selling T-shirts or [ __ ] like that
1:37:55 you can use generative AI to make the
1:37:58 images and write the copy and do all
1:38:00 that but use and and help you do the
1:38:02 marketing and the marketing plan and all
1:38:03 that but you still have to do the
1:38:05 business and and
1:38:07 the quality of the stuff you create has
1:38:09 to be good enough to rise above the
1:38:11 noise right so there's no magic formula
1:38:14 it's all just you know are you solving a
1:38:17 problem that someone else is willing to
1:38:19 pay for and if you are doing that then
1:38:22 you have a business and if you're not
1:38:24 there's there's there's not a Magic
1:38:26 Bullet when it comes to that stuff
1:38:28 unfortunately and and the problem is if
1:38:31 there is one
1:38:32 and it's really that easy then everyone
1:38:34 will start using it and it'll stop
1:38:36 working right so there are blips in the
1:38:38 radar where someone will crack the code
1:38:40 on something like you know I made fifty
1:38:42 thousand dollars in three minutes you
1:38:43 know with this thing and then they put
1:38:45 out the course on that and then
1:38:46 everybody does that and all of a sudden
1:38:48 that thing stops working
1:38:49 and at that point they're just making
1:38:51 money on selling the course
1:38:53 I heard a good argument about AI
1:38:55 learning that human Innovation comes
1:38:57 from mistakes oh that's interesting yeah
1:38:59 and it's like what AI does is it gets
1:39:02 you to 80 really quickly and and like if
1:39:05 it gives you shitty stuff and you keep
1:39:07 working with it it'll give you
1:39:08 increasingly better stuff that's it
1:39:10 that's that's really interesting
1:39:12 Delta gummies and GPT can be crazy oh
1:39:15 that's funny
1:39:17 um
1:39:17 uh that sounds like crypto it's a fad
1:39:20 now now destroy then Mass adoption
1:39:24 uh yeah although crypto kind of screwed
1:39:27 the pooch because they had too many
1:39:28 crypto Bros with not actual businesses
1:39:32 that were just scamming people out of
1:39:34 money so crypto and blockchain have some
1:39:37 catch not some catching up they have
1:39:38 some reputation some well-earned
1:39:41 reputation repair to do before people
1:39:44 are going to trust it again I think AI
1:39:46 is actually going to drive adoption of
1:39:48 blockchain and nfts
1:39:50 um in in very specific use cases like
1:39:52 authenticity
1:39:53 um so so we'll see flight computers
1:39:56 already land yeah I know they do but you
1:39:58 know you know there needs to be the
1:39:59 override and all that sort of stuff like
1:40:01 we're not going to have
1:40:03 um
1:40:03 pilotless airplanes anytime soon as long
1:40:07 as insurance companies are involved they
1:40:09 won't have it
1:40:10 we'll always have elevator operators
1:40:11 exactly that's true judges have biases
1:40:14 let's use AI to process the law yep and
1:40:17 if you're interested in that Ampersand
1:40:20 um check out the company anthropic
1:40:21 they've got what's called a
1:40:22 constitutional learning model where
1:40:24 their AI they write basically a set of
1:40:26 rules like don't hurt humans increase
1:40:29 Prosperity uh whatever whatever their
1:40:31 rules are and then when it processes an
1:40:33 answer it runs it through the
1:40:35 Constitutional model and iterates it
1:40:37 until it passes so so that's a company
1:40:40 that's based on you know some form of
1:40:42 automating the the
1:40:45 um analysis piece so as not to introduce
1:40:48 bias and and I I assume what you'll be
1:40:50 able to do is is opt into the kind of
1:40:53 Constitution that that matches your
1:40:55 values
1:40:56 I had a help desk interview last year
1:40:58 told them I was working on python for AI
1:41:00 in the future cool because AI would have
1:41:03 our help desk jobs no offer well
1:41:07 it's gonna destroy the help desk uh
1:41:11 world as we know it it's going to make
1:41:14 it much better
1:41:16 what operate chat gbt I don't know what
1:41:18 that means why no one explains how it
1:41:20 works you want to see how
1:41:20 [Laughter]
1:41:25 hold
1:41:29 how it works when you understand how
1:41:32 this [ __ ] works it is
1:41:35 more mind-blowing than before you knew
1:41:39 it
1:41:40 so
1:41:42 where do I have it
1:41:44 here it is okay
1:41:47 so we're going to talk about how Chachi
1:41:50 BT works
1:41:52 and if you are a machine learning person
1:41:55 and a scientist and if I [ __ ] this up
1:41:58 and butcher it please let me know
1:42:00 but I've done this enough now that I
1:42:02 think we're we're generally in the
1:42:03 neighborhood so if oh I can do this I've
1:42:06 got it presenting in a window let me
1:42:08 make this smaller hold for a second here
1:42:15 wait is that in a window
1:42:23 it is isn't it or is it not
1:42:28 hang on hang on I'll be right with you
1:42:35 [Music]
1:42:39 okay
1:42:40 all right so I got this thing so now it
1:42:42 fits in there okay
1:42:44 so
1:42:46 we'll talk about a couple of things
1:42:47 generative GPT stands for generative
1:42:50 pre-trained Transformer so generative
1:42:52 means that it's generating stuff
1:42:54 let me get rid of this background that's
1:42:56 distracting
1:42:59 that's good
1:43:02 okay
1:43:04 generative means that it's generating
1:43:05 stuff pre-trained means that it's
1:43:08 pre-trained you don't have to bring your
1:43:10 own data machine learning
1:43:12 um models you used to have to bring your
1:43:14 own data and you can only use relatively
1:43:16 small data sets the Transformer the T in
1:43:19 GPT stands for Transformer that's the
1:43:22 technology that allows
1:43:24 these models to be trained on really big
1:43:26 training data like all of the internet
1:43:28 so so GPT Bard you know Claude all of
1:43:32 these things the base model is trained
1:43:34 on massive amounts of data
1:43:37 um so generative it makes [ __ ]
1:43:39 pre-trained you don't have to bring your
1:43:40 own data Transformer is the model that
1:43:42 it's trained on that's what GPT stands
1:43:44 for okay the other thing you hear a lot
1:43:47 is llm large language model that's just
1:43:50 basically this is the the this
1:43:55 particular tool that looks at all this
1:43:58 language and you can interact with right
1:44:00 it's large it uses language and it's a
1:44:02 model and there's lots of different
1:44:04 versions of that
1:44:06 then there's this thing called latent
1:44:08 space and this is this is how this [ __ ]
1:44:11 works and it's this is a dramatically
1:44:13 oversimplified explanation of it but the
1:44:17 latent space is basically a mathematical
1:44:18 rep representation of words clustered by
1:44:22 meaning meaning so you've got sort of
1:44:24 think of it like these semantic little
1:44:26 canyons and sub Canyons so all those
1:44:28 different colored dots represent like on
1:44:31 an x y a two-dimensional x y axis
1:44:36 Bunches of different words so
1:44:39 um you might have like the blue cluster
1:44:41 might be animals and the black cluster
1:44:43 might be plants and the
1:44:45 orangey cluster might be I don't know
1:44:48 different types of wood and what
1:44:49 whatever now this is just a 2d
1:44:53 um representation but basically what
1:44:54 they do is they do what's called
1:44:55 embedding and when they embed
1:44:58 they take these tokens they take these
1:45:01 words and they turn them into a
1:45:02 mathematical string like like just a
1:45:05 series of numbers that plots that word
1:45:08 in a particular Vector space right in a
1:45:11 cluster of meaning semantic clusters
1:45:14 so so an example might be you might have
1:45:17 the word dog that there it might be
1:45:20 clustered in a group of words that means
1:45:22 animal but it might also be clustered in
1:45:24 a group of words that mean you know he's
1:45:26 a dog right so so you can have the same
1:45:29 word can have different meanings and
1:45:31 then what happens is you you create your
1:45:35 prompt and it basically takes your
1:45:37 prompt and it turns it into mathematical
1:45:39 stuff as well and and then
1:45:42 um
1:45:43 the large language model predicts what
1:45:47 should be the next word in the response
1:45:49 based on your prompt and then it just
1:45:51 does that over and over and over again
1:45:54 um and and
1:45:56 um and so it kind of works like this so
1:45:59 I I typed in the prompt the the quick
1:46:01 the quick brown fox jumps over the lazy
1:46:04 blank and the answer is dog we all know
1:46:06 the answer is dog because our brains
1:46:08 work like that right well here's how the
1:46:09 large language yeah how the large
1:46:11 language model does it it basically
1:46:13 deselects all the clusters of words that
1:46:16 are not the right sort of semantic
1:46:17 neighborhood based on what you wrote and
1:46:20 then you can't really see it but there's
1:46:21 a bunch of light blue dots there maybe I
1:46:23 can make this bigger so you can just see
1:46:24 the the
1:46:26 this a bit better
1:46:30 um
1:46:31 so all those light blue dots are numbers
1:46:35 right these strings of numbers that are
1:46:37 kind of semantically related and then
1:46:40 one of those strings of numbers will
1:46:41 have a higher probability that says 96
1:46:44 so one of the world's words will have
1:46:47 the highest probability that it's the
1:46:49 right next word to go in that sentence
1:46:51 and then that number that that Vector
1:46:54 number is associated with the word dog
1:46:56 and then the machine inserts the word
1:46:59 dog into that sentence right and it's
1:47:01 you know there's other words that could
1:47:03 have been there but it puts the word dog
1:47:04 in there and it does that over and over
1:47:06 and over again so when you see it right
1:47:08 like a big long
1:47:11 um you know a big long response
1:47:14 um it's doing it word by word in in this
1:47:17 kind of mathematical way so so this is
1:47:19 the reason so I hope that helps for one
1:47:21 thing
1:47:22 um that's how they work and this is also
1:47:25 the reason when they say well it doesn't
1:47:27 like it says it hallucinates why does it
1:47:29 hallucinate
1:47:31 because it's not actually analyzing the
1:47:34 thing that it wrote It's just saying I
1:47:37 think the next word should be this and
1:47:38 then I think the word after that should
1:47:40 be this and basically it just does that
1:47:41 and it writes all this [ __ ] it's not
1:47:43 going back and looking at it
1:47:45 there are models being developed right
1:47:47 now that do go back and look at it right
1:47:49 and say it created a citation here was
1:47:52 that citation actually accurate but
1:47:53 that's a completely separate process
1:47:55 when it writes that big long thing it's
1:47:57 completely unaware of what it wrote so
1:47:59 sometimes these models will act
1:48:02 empathetic right they will present an
1:48:05 empathetic response they're not feeling
1:48:07 empathy they're just saying based on
1:48:09 what you said there you could probably
1:48:11 use a hug so so the the next series of
1:48:15 words that should be there are some kind
1:48:16 of empathetic response so so that's why
1:48:19 it gets [ __ ] wrong and but but also when
1:48:22 you watch it Write a response like the
1:48:24 first chapter of a novel and it does it
1:48:26 in like 10 seconds like what's happening
1:48:28 mathematically like these people are
1:48:30 [ __ ] smart that invented this [ __ ]
1:48:32 it's crazy
1:48:34 all right
1:48:36 all right I gotta run in a few I do two
1:48:38 thanks for all the great info you're
1:48:39 welcome yeah I gotta go too
1:48:41 I gotta go mow or do I don't know
1:48:43 whatever I gotta go live my my Saturday
1:48:45 life and I'm going out to dinner with
1:48:48 one of my employees is up from Durango
1:48:50 so that's kind of cool
1:48:52 what website is this I don't know which
1:48:54 one I don't know which one I was on
1:48:55 sorry about that
1:48:57 e
1:48:59 here we come to save the day
1:49:09 what's going on uh move that up too high
1:49:11 there we go okay all right
1:49:13 hi I hope as well did you try mid
1:49:15 Journey yes I heard it's a good one as
1:49:17 well I think mid-journey is the best one
1:49:21 mid journey is really good resistance to
1:49:22 AI is like when calculators came out
1:49:25 they took
1:49:26 um and now look at math Baseline
1:49:28 capabilities that's exactly right
1:49:29 they're I think there were even I've
1:49:32 seen pictures of this now they could
1:49:33 could have been mid-journey created but
1:49:35 I've seen pictures that professors were
1:49:37 actually like protesting the use of
1:49:39 calculators right so same exact thing
1:49:43 Paul rates are marketing AI Institute
1:49:45 thank you for putting up there hey Kyle
1:49:47 how are you today good Liz thank you
1:49:50 I'm quite new here but enjoying it cool
1:49:53 I'm glad that you're helping one another
1:49:55 because I don't get to these questions
1:49:56 very fast
1:49:58 um a problem in AI inventing inventing
1:50:01 sources for papers a problem in AI
1:50:04 inventing sources for papers and the
1:50:06 sources actually do not exist that is
1:50:08 absolutely a problem and and you know
1:50:11 that's really on the humans this will
1:50:13 get better over time so part of the
1:50:15 challenge with
1:50:17 all this stuff is that it's very very
1:50:19 new and it's very very raw
1:50:22 um
1:50:22 there was a lawyer I think in the past
1:50:24 week that created a
1:50:27 had chat GPT write a legal brief that he
1:50:31 handed to a judge
1:50:33 and he didn't check the sources to see
1:50:35 if they actually were right or existed
1:50:39 at all they didn't and the judge caught
1:50:43 him and said what the [ __ ] is this so
1:50:47 if you're a lawyer and use chat GPT for
1:50:50 a little leg up which you should be
1:50:52 doing because why not it's the most
1:50:54 powerful writing tool in the history of
1:50:56 mankind check your work
1:51:00 that's on the human but but you should
1:51:02 just know that these things
1:51:03 um that these things hallucinate a lot
1:51:07 and and you need to you need to know
1:51:09 that and and kind of understand what
1:51:11 that actually means and how it does it
1:51:13 and why it does it all that sort of
1:51:14 stuff and I I mean I just explained it
1:51:17 like when I explained how these tools
1:51:18 actually work that's why
1:51:21 why they hallucinate because it's not
1:51:23 aware of what it's writing canva bulk at
1:51:26 it with chat EBT tables is insane it's
1:51:29 it's absolutely insane if you don't know
1:51:31 what that is so you could you can have
1:51:34 you could have let's go do it we'll go
1:51:36 to chat GPT chat GPT you can make money
1:51:41 all right we're gonna do a new thing
1:51:42 we're going to go to gpt4 because it's
1:51:44 just better we're going to say create a
1:51:47 table of
1:51:50 um T-shirt
1:51:52 sayings
1:51:54 uh um we did one the other set let's say
1:51:58 four
1:51:59 um Gen X
1:52:01 dad's
1:52:05 um
1:52:06 who
1:52:08 let's see
1:52:10 um
1:52:11 just want to be left alone
1:52:16 um they should be funny and memorable
1:52:20 and
1:52:22 um
1:52:23 short
1:52:26 enough for
1:52:29 t-shirts
1:52:33 um
1:52:34 in the table
1:52:37 have
1:52:39 um a suggested image wait suggested
1:52:45 image
1:52:47 the saying and a target audience
1:52:54 as columns okay
1:52:58 so this should just create us a
1:53:02 relatively brilliant table of t-shirts
1:53:04 that we could go make
1:53:06 so the suggested image cartoon of a man
1:53:08 on a desert island I'm not anti-social
1:53:10 I'm pro-solitude okay the first one is
1:53:13 [ __ ] good
1:53:15 I'm not anti-social I'm pro-solitude
1:53:17 with a dude on a deserted island a
1:53:20 cartoon of a dude on a deserted island
1:53:22 really that's awesome okay so so so this
1:53:25 has now just generated this table you
1:53:28 can copy and paste this into a CSV file
1:53:31 into a Google Sheets file create a CSV
1:53:33 then you go into canva and you could
1:53:35 pick a bunch of designs and you
1:53:36 basically just say swap out these text
1:53:39 elements with what's in the CSV and it
1:53:41 will spit out
1:53:43 20 or 30 t-shirt designs I mean or
1:53:47 Twenty or thirty thousand there there
1:53:49 was some jerk on tick tock he was he was
1:53:53 making I think he was making a hundred
1:53:56 thousand t-shirt designs a day and
1:53:59 listing them like he he had figured out
1:54:01 generative AI it wrote a hundred
1:54:03 thousand you know t-shirts generated
1:54:06 designs for them and uploaded them to
1:54:08 some T-shirt shop or I think it was just
1:54:10 designs I don't think it was sayings
1:54:12 like this but yeah that stuff's
1:54:13 absolutely amazing
1:54:17 um
1:54:20 all right I've gotta I gotta run so let
1:54:22 me do a couple more questions I'm kind
1:54:24 of down at the bottom I skipped a bunch
1:54:25 of questions
1:54:27 so Tick Tock told me I did a great job I
1:54:30 went live for 90 minutes I I tend to get
1:54:32 that we're down to 19 people here so
1:54:34 this is just the hardcore the the
1:54:36 hardcore AI learning lab fans so thanks
1:54:39 for hanging out with me what's the name
1:54:41 of the weekly group it's called it's
1:54:42 called
1:54:43 um the the AI Salon it's the the there's
1:54:47 a link to it in my bio and it's the
1:54:49 salon dot AI is a link tree that's got
1:54:52 the Discord it's got the Meetup info for
1:54:55 the actual meetings it's got a LinkedIn
1:54:57 group
1:54:58 so yeah it is on Discord and you can do
1:55:00 mid journey within our Discord it's
1:55:02 pretty cool
1:55:03 saw someone call themselves an AI expert
1:55:06 you better catch up Kylie yeah I know
1:55:08 experts everywhere here's the deal
1:55:10 anyone that tells you well there are AI
1:55:13 experts that are like machine learning
1:55:14 scientists they're experts but anyone
1:55:16 talking about this generative AI [ __ ] if
1:55:19 any of them tell you they're an expert
1:55:20 at this right now
1:55:22 they're wrong no one's an expert right
1:55:24 now everyone's trying to figure this
1:55:25 [ __ ] out
1:55:26 I sell a legit literal step-by-step
1:55:28 course on how to digitize any business
1:55:30 and has mrr that's awesome
1:55:33 it has over 90 modules crypto a fad no
1:55:36 digital currency is the future yeah I
1:55:38 agree with that but they they have some
1:55:41 uh some well-earned PR [ __ ] they got to
1:55:44 dig out from
1:55:46 um Jen what is the significance of the
1:55:48 Wired Magazine behind you
1:55:51 um
1:55:54 I'm in it
1:55:56 so this is from
1:55:59 July 1995
1:56:02 and
1:56:05 I've talked about I started a
1:56:08 one of the first online magazines it was
1:56:10 called Urban desires
1:56:17 I probably should have bookmarked this
1:56:21 Nate did I was it just there no I'm in
1:56:25 the right section I think
1:56:31 that was me
1:56:34 30 years ago
1:56:37 um and
1:56:39 that is urban desires and that's the
1:56:42 little write-up so as in Wired Magazine
1:56:45 and it's Richard Dawkins which is cool
1:56:47 too so that's that and then
1:56:51 if you know the uh if you know the dude
1:56:54 uh Jason kalkanis the the investor dude
1:56:57 he did this uh magazine in New York
1:57:00 called silicon alley reporter and that's
1:57:03 my uh partner Chan Sue and I from
1:57:06 agency.com was the agency we started and
1:57:09 we were profitable so it was called men
1:57:11 in the black get it
1:57:13 it's a profit joke
1:57:18 so yeah that's that's the
1:57:22 19 devotees we're down to 18 now can I
1:57:25 can can I suggest hardware specs such as
1:57:28 CPU side SSD memory for best results
1:57:30 using AI it really depends on what you
1:57:32 you want to do the the open source
1:57:34 models are getting really efficient if
1:57:36 you're on the Mac I would say like with
1:57:39 Max in general always get the latest one
1:57:41 and get the highest highest one you can
1:57:43 the new the new M2 Ultra which I think
1:57:45 that's like 40 Grand so you don't need
1:57:47 to get that
1:57:49 um but but like anything Apple silicon
1:57:51 there's a lot of developers developing
1:57:53 stuff for Apple silicon if you're on PC
1:57:55 you basically need a good um GPU a good
1:57:59 graphics card like a 30 or 30 90 or
1:58:01 higher but that's if you want to
1:58:03 actually do the training if you just
1:58:05 want to install large language models
1:58:06 and use them
1:58:08 you can probably get away with something
1:58:09 lesser as these open source models get
1:58:12 smaller and smaller so I'd say give it
1:58:14 six months and you won't you won't need
1:58:15 a really special machine you'll even be
1:58:17 able to run it on phones
1:58:19 thanks for hanging with it's been every
1:58:20 day lately cool all right
1:58:23 um I'm out of here I gotta go do some
1:58:25 mowing I got to get ready for dinner
1:58:27 tonight it's been fun hanging with you I
1:58:29 appreciate the questions I appreciate I
1:58:31 appreciate the hardcore 20 that are left
1:58:33 all right we'll talk soon bye