aiLL 7/7/23 - 1 - The Impact of Code Interpreter on AI's Potential

Description

Chapters

Transcript

0:03 foreign
0:27 welcome welcome Happy Friday night
0:31 just kind of went out to uh First
0:33 Fridays here in Denver saw a bunch of
0:35 art that was nice
0:37 hello ice cream
0:40 welcome to the AI learning lab the
0:42 nightly live it's been nightly for like
0:44 two weeks now nuts
0:48 love these lives Kyle thank you very
0:50 much Jamie or James I appreciate that
0:53 the the hair is special tonight that
0:56 that that looks like I tried that that
1:00 is not that was not on purpose
1:02 [Laughter]
1:07 oh man uh welcome to the AI learning lab
1:11 AI learning lab we'll get it going here
1:13 if you have questions about generative
1:16 AI or want to I don't know look at
1:19 something or goodness goodness goodness
1:22 take a look at code interpreter from
1:25 chatgpt which I got access to today holy
1:28 crap uh it's making me feel very stupid
1:31 because I don't know enough about data
1:34 analysis to use it right now but I'm
1:36 figuring it out and it looks remarkable
1:38 and then there's prompts.chat that thing
1:40 is if you haven't played with chat GPT
1:43 and you don't really know what prompting
1:44 is and you hear people talking about it
1:47 that document is a good primer for how
1:50 to do prompting and then it gives you a
1:52 lot of examples of personas that you can
1:57 tell chat GPT to act like some sort of
2:00 expert act like an expert on marketing
2:02 or on sales or on writing or whatever it
2:05 might be and it dramatically improves
2:07 the content so anyway welcome here if
2:10 you have questions pop them in the
2:12 comments below I do my best to answer
2:14 everything happy Friday night everybody
2:16 let's get this show rolling rolling
2:19 rolling
2:21 Friday Friday Friday
2:24 all right can AI elaborate upon huge
2:27 heat wave hitting the Southwest
2:30 U.S in a few days
2:34 possibly
2:36 that might be something for being let's
2:38 go
2:41 let's go I don't know what you mean
2:42 elaborate but let's go see what Bing has
2:44 to say about it my shirt is apparently
2:48 my my massive muscles are too big for my
2:51 little short sleeves here I can't move
2:52 my arms up
2:54 um
2:56 all right
2:57 all right see
2:59 uh let's do more precise and so but it
3:02 looks like you know like a Google kind
3:05 of search engine then if you click the
3:07 chat button
3:08 that brings you here and this is Bing
3:11 chat it's it's basically chat gpt4
3:14 and let's see
3:17 um can you tell me what is
3:24 uh what
3:27 is happening with the heat
3:31 wave in the U.S Southwest
3:41 Heatwave us Southwest
3:45 yes there is a heat wave according to
3:47 NASA some of the high pressure settled
3:49 over
3:50 Southwest in early July causing ground
3:53 surface temperatures to rise the heat
3:54 wave is expected to fix affect much of
3:57 Arizona New Mexico build into California
3:59 the heat wave could rank among its worst
4:03 in history in terms of intensity or
4:06 longevity holy crap well here in Denver
4:08 it's it's been raining like it
4:11 threatened to rain really bad tonight
4:12 it's been raining and hailing and
4:13 storming we've gotten more
4:15 rain for the first six months of 2023
4:19 than Seattle
4:20 Denver
4:21 which is a high desert got more rain in
4:24 the first six months of the year than
4:25 Seattle that's never happened in history
4:27 before so there's some [ __ ] going down
4:30 um I do know that there's an El Nino
4:32 that we exited in La Nina and we entered
4:36 an El Nino so I know that's going on I
4:38 don't know if that affects it but also
4:41 July 3rd was the hottest recorded day in
4:44 the history of the Earth
4:47 so we're getting better
4:49 right
4:53 all right it's okay the robots are gonna
4:55 kill us anyway can AI generate images
4:57 such as logos and how it can it's not
5:01 super great it doesn't do text very well
5:04 at all but if you want to do like logo
5:06 ideas I'll do I'll do some here at Bing
5:08 so let's just say
5:10 um
5:11 let's see graphic graphic
5:15 logo logo
5:18 um flat illustration
5:21 let's see
5:23 um of a
5:25 oh I don't know
5:27 um
5:29 of a flower shop
5:35 um with
5:37 a rose as the main element
5:43 and Geo match Rick
5:48 shapes behind it
5:51 let's see what this does
5:55 so so Bing does images now now it's it's
5:58 sitting on top of Dolly which is the
6:01 weakest of all the image models so my
6:03 guess is that this is gonna suck but
6:08 I don't know we'll see
6:12 um it's not as horrible as this
6:14 way better than I thought it would be
6:16 it's not really a logo
6:19 um
6:21 those are kind of more like Design
6:23 Concepts for for you know maybe some
6:25 printed materials things like that
6:27 that's not horrible let's let's flip
6:29 over I'll flip over to uh to Discord
6:32 we'll do something in uh mid Journey
6:35 that was a piece I did last night on the
6:36 live uh first for mid-journey
6:43 someone's picking on me from for making
6:45 art that looks like theirs that's pretty
6:47 funny
6:48 um let's see
6:50 so if if you haven't used mid journey by
6:53 the way how you use it this is we're in
6:55 Discord so in Discord you've got you
6:57 know all these servers so these are all
6:58 different communities and then I'm in
7:01 one uh for the AI Salon which is a
7:03 community that I started and I've got a
7:05 mid-journey channel in here so let's
7:07 just hop in here and we'll go so how you
7:09 do this you go slash imagine
7:11 and then it gives you a prompt oh I'll
7:13 do another thing before we do this we'll
7:15 we'll go make the prompt even better so
7:18 I'm gonna go over to so I just showed
7:20 you Bing that's an image that was
7:22 generated in bing now we're going to go
7:24 to go to Poe po.com
7:28 I assume that means as in Edgar Allan
7:30 Poe so when you're at Poe it's a normal
7:32 like you know chat bot search engine and
7:35 you've got six different models you can
7:36 play with but you also have these Bots
7:38 you can make a bot and you can
7:41 sorry trying to get the
7:44 Focus to focus there we go
7:46 um
7:48 you can look at other people's spots but
7:50 there's there's a mid-journey bot that's
7:52 actually pretty good so if you go to
7:54 poe.com mid-journey it takes you to the
7:57 spot that specializes in fixing your
7:59 shitty prompts so we'll say logo
8:03 of a flower
8:07 shop flat
8:11 illustration
8:13 let's see graphic I don't know on white
8:17 let's say and then it's going to rewrite
8:20 that so color logo of a flower shop flat
8:23 illustration graphic on white featuring
8:25 a blooming flower with a with delicate
8:27 petals simple and elegant minimalist
8:30 design Adobe Illustrator software RGB
8:33 color model vector format Mary Blair
8:35 Charlie Harper Saul bass so let's copy
8:39 that that's our prompt we'll hop over to
8:43 um to Discord and mid journey and I'm
8:45 just going to paste it in and hit return
8:48 so so we're starting to get to get where
8:50 you know the AIS are starting to prompt
8:53 the other AIS and you can you can use
8:55 them in conjunction with one another and
8:56 you can figure out well this tool might
8:59 be better for this project and this
9:00 might be better for that project and
9:03 in general with images mid-journey is
9:05 better than anything else it's just the
9:07 way it is I don't make the rules I just
9:09 enforce them
9:10 [Laughter]
9:14 uh actually some of these are looking
9:16 kind of logo like what you'll notice
9:18 here is
9:20 um
9:21 the the way these diffusion models work
9:24 they start out as kind of these blurry
9:26 you know highly noisy images and then
9:28 they seem to get a little more clear
9:30 over time that's exactly how they work
9:32 they kind of are averaging lots and lots
9:35 of random noise generated images that
9:37 were based on you know real images but
9:40 just random noise they kind of average
9:41 them together and it starts as this
9:43 random noise
9:44 amalgamation and then it denoises them
9:47 over like 30 or 40 or 70 steps
9:51 and boom there's these logos so this one
9:56 isn't exactly a rose on the lower right
9:59 but that's pretty good logo actually
10:00 it's pretty interesting
10:02 it's at least good
10:04 um
10:05 it's at least good uh whatever you call
10:07 it inspiration don't you think
10:10 I mean that's pretty dang nice
10:12 it tried to do text but you could
10:15 obviously wipe that text out of there
10:18 and you know that one's all any of these
10:21 for flower shops I mean actually this
10:23 this upper right one
10:25 um it's not really a logo for a store
10:27 but it certainly looks like it could be
10:29 on a sign for a store
10:31 and they put it in a store window which
10:33 is kind of cool
10:35 so
10:36 yeah it can do that that's that's a
10:39 couple of ways you can you can do it
10:41 um
10:48 and uh let's see
10:51 what else do we have here I like net nip
10:54 like a net beans for dummy dummies
10:57 leave that chat bot alone they just got
11:00 sued for copyright infringement I assume
11:02 you mean mid-journey everyone's getting
11:04 sued right now everyone's going to be
11:05 sued for the next long long time
11:08 um you know yeah I I understand you
11:12 your your resistance to progress uh but
11:16 you know it's this stuff isn't going
11:18 away so
11:20 um yeah collar shirt tonight formal
11:23 night well it's not really a formal
11:25 night but uh but I I did I am wearing a
11:28 shirt dark enough to read text in front
11:29 of so so I don't have to change it uh
11:34 uh this is a great way to learn about
11:37 the great AI thank you
11:39 uh who actually knows the good news for
11:42 the Colorado River with all that rain oh
11:45 I well Utah had like what did they have
11:49 I know some 40 or 50 feet of snow I mean
11:52 they had a ridiculous amount of snow so
11:55 that all contributed to that water basin
11:57 and then yeah it is it's been it's
12:00 honest to God like living in well no
12:02 it's not like living in Florida at all
12:04 but but like afternoon thunder thunder
12:06 thundershowers like just it's been weird
12:09 it's been not like Colorado so yeah I
12:11 think that is good for The Californians
12:14 who are like where's our [ __ ] water
12:16 our Lakes are drying up please help
12:22 so their water should be okay government
12:26 weather experiments yeah Chemtrails and
12:28 all that stuff huh are we hacking into
12:30 NASA not yet although uh code
12:32 interpreters doing some pretty insane
12:34 stuff I'll show you that in a second
12:36 thank you for sharing you're welcome
12:38 I'll be getting three arms and three
12:39 legs on some of my mid-jury journey true
12:42 to life rendering how do I fix
12:45 Ed
12:45 um Google or like go into Reddit into
12:48 the mid-journey things you can put in
12:50 some negative prompts in mid-journey to
12:52 have it not do certain things like you
12:54 know don't give me weird hands don't
12:55 give me too many arms things like that
12:57 so you can give it some negative prompts
12:59 um
13:00 you can also do things like remixing
13:03 where you upload a photo and sort of
13:05 remix it into something else it'll keep
13:06 the composition of the first thing but
13:08 change everything out that's a way that
13:11 you could handle that
13:13 um just keep experimenting just you know
13:15 just I there the weird thing about these
13:20 models just the nature of how they
13:22 produce the images they produce
13:24 they don't really know what they're
13:26 producing it it is literally sort of
13:28 this Amalgamated set of pixels that gets
13:31 denoised into a real image so I I
13:34 suppose if you know certain parts got
13:36 overlaid over certain parts or however
13:38 the hell it's doing what it's doing
13:41 um it just does that so you just got to
13:43 keep working on your prompting
13:45 and uh yeah you'll get there
13:48 you will get there can AI validate the
13:52 theory to increase
13:53 monsoonal moisture for the Colorado
13:55 Basin by evaporation of the ocean
13:59 uh I don't know I don't think so you
14:02 might play around with uh with code
14:04 interpreter
14:06 um go get some go get some weather data
14:07 it's possible sounds like we got a
14:10 little conspiracy theory action going on
14:11 but you know whatever uh I'm in Los
14:15 Angeles for the week and it's definitely
14:17 dry I'm supposed to be on vacation but I
14:20 can't stay away from the AI news I know
14:22 the AI news is just crazy so check check
14:25 this out so
14:27 let me flip up got
14:29 um
14:30 yeah it's not back for me so
14:33 um
14:35 browse with Bing is still missing but
14:37 look at this code interpreter is here
14:40 and it's Live code interpreter is live
14:42 so this is something I've been waiting
14:44 for for a long long time so the way this
14:47 works you go to the new code interpreter
14:50 feature and notice you get a little you
14:53 can't really see it you get a little
14:55 plus button here
14:57 and you can upload a data file and it
15:01 can you can even upload so so I went to
15:04 go get a bunch of data I got some from
15:06 LinkedIn I got some from some different
15:08 data sources
15:10 and one of the data sets I got was all
15:12 these individual CSV files right they
15:15 like you can you don't have multiple
15:16 worksheets you can upload a zip file
15:20 that has multiple csvs in it and it will
15:22 read them all it reads inside the the
15:24 zip files it's pretty amazing
15:26 so let me see there was one here
15:31 this one so this was a bunch of Chess
15:34 data so I'm uploading a spreadsheet with
15:37 a bunch of Chess data and then I'm going
15:39 to say
15:41 um
15:42 can
15:44 you
15:45 let's see visualize this data
15:51 in in a
15:54 visually interesting way
16:00 and
16:01 describe what
16:04 it's all about
16:08 and
16:10 sure let's start by unzipping the file
16:12 so so it's just like a plug-in it's just
16:14 the way plugins work it just pops open
16:17 that thing it says finished working the
16:19 zip file contained a single CSV file
16:22 let's load this file and look at it
16:25 and then it says it's got
16:28 ID rated created at last move at turns
16:31 Victory status so it's all this data
16:33 about Chess
16:36 to provide visual representation of the
16:38 data we can create plots for various
16:40 aspects let's proceed with this
16:42 visualization
16:43 okay
16:45 and so now it's firing off now let me go
16:47 show you what's going on behind the
16:49 scenes here so every time it fires one
16:51 of these off it's actually writing
16:53 python
16:54 so so it wrote The Python code and
16:57 installed the libraries to unzip the
16:59 file
17:00 and then the next one it it imported
17:04 Panda pandas and then went and looked at
17:08 the data and so so it analyzed the data
17:10 and there it made [ __ ] charts
17:15 of the data it made [ __ ] charts of
17:18 the data
17:20 distribution of game results the
17:22 majority of games are won by either
17:23 white or black with white having a
17:25 slight Edge there are very few games
17:26 that end in the draw distribution of
17:28 number of turns the number of turns in
17:30 the game followed roughly a normal
17:32 normal distribution with most games
17:34 having around 50 turns there are also
17:36 some games with a very high number of
17:37 turns indicating particularly long games
17:40 distribution of player ratings
17:43 10 commonly used openings
17:46 the the most commonly used opening data
17:49 opening in this data set is Van uh
17:52 cruises opening followed by the Sicilian
17:55 Defense and the Scandinavian defense
17:58 misis cotric variation I don't I don't
18:01 play chess I don't know all these the
18:04 rest of the openings in the top 10 are
18:05 used less frequently
18:07 I mean
18:08 holy [ __ ]
18:10 um let me show you another thing it can
18:12 do let me do I'll do a new chat
18:15 um
18:15 I'll get a code interpreter I'm going to
18:17 upload an image
18:19 and then I am going to go grab
18:24 is this it yeah okay so I've got this
18:26 image so this was an image that was
18:28 created in the AI Salon it was really
18:29 cool she was playing around with
18:31 extending the image and doing panning
18:33 okay so I'm going to upload this image
18:36 it's not doing image recognition like
18:39 um like gpt4 is supposed to where you
18:42 can see what's in the image but I can do
18:45 this I can say
18:47 um
18:49 uh let's see
18:51 find me the
18:55 15
18:58 um like I I most common
19:03 colors in this image and create a graph
19:10 of the percentage they are used in the
19:17 image
19:18 make the
19:20 bars of the graph
19:24 match the color it represents
19:29 and label with
19:32 um
19:32 hex codes
19:36 and even name the colors like they are
19:41 crayons
19:44 okay
19:47 so all right to find the most 15 most
19:50 common colors in the image I'll follow
19:52 these steps I'm going to load the image
19:54 calculate a histogram of the image which
19:56 will give us the frequency of each color
19:57 identify the 15 most common colors
20:00 convert those colors to hex
20:01 representations so it's it's figuring
20:04 out its own logic it's then writing
20:07 python code
20:09 to execute that logic and if it gets it
20:13 wrong it'll rewrite the code and
20:15 re-execute it I obtained the 15 most
20:18 common colors in the image
20:20 however I noticed that the color values
20:23 are in a Range zero to two
20:26 let's see finished working I apologize
20:29 for the oversight it it seems that the
20:32 environment does not have the web colors
20:34 Library installed however we can still
20:36 convert them to hexadecimal
20:39 so it made some mistakes
20:41 next I'll create a bar chart to
20:43 visualize the percentage of the colors
20:45 used
20:46 these colors are not right it didn't get
20:48 the colors right here
20:51 and it didn't name them there's a
20:54 horizontal bar chart
20:58 um
20:59 let's see
21:00 um
21:01 can you get uh let's see can you average
21:09 um let's see can you get the
21:14 15 colors that are
21:18 an average
21:20 rather than the most pixels I don't even
21:25 know if that's the correct question but
21:28 anyway
21:29 so this is what this thing does you can
21:31 you can upload
21:33 spreadsheets you can upload PDFs it will
21:36 write the python code to pull the text
21:38 out of a PDF
21:40 you could use this the file that you
21:42 upload here can be a hundred megabytes
21:44 so
21:46 so
21:47 right now chatgpt has this 4 000
21:51 this 4 000 token context window like 3
21:55 000 words where it kind of doesn't
21:57 really have its memory
21:59 um you could you could put in a hundred
22:02 megabyte prompt basically and and have
22:06 it start executing against that prompt I
22:09 mean I I just got this today so like
22:12 I've only played with it for about an
22:13 hour and I like it this is making me
22:15 feel incredibly stupid because I don't
22:17 even know what to ask it
22:19 but holy crap holy crap is it amazing so
22:22 anyway what's it doing it let's see
22:26 what's it done here has it found
22:28 anything
22:30 average color hexes so it wrote code to
22:32 get the average colors
22:35 kit K means cluster centers as type
22:39 integer
22:40 average color hexes RGB to hex color
22:46 apologies the algorithm's taking too
22:49 long
22:52 I'll use a method called mini batch K
22:54 means
22:56 so it didn't it didn't like that so so
22:58 it's like it tried some code it took too
23:00 long it said let me try another approach
23:02 it took too long
23:04 now it's trying another approach
23:09 now we've got some real colors
23:12 so it looks like it might have worked
23:14 next I'll create a bar chart to
23:16 visualize it
23:18 I I mean
23:20 this is nuts
23:23 there look at that now it didn't get the
23:24 percentages right but it got the colors
23:27 so now so the so the horizontal bar
23:29 chart is the 15 colors that are in the
23:30 image
23:31 and if we go back to the image
23:34 let's see
23:35 um
23:38 if we go back to the image you know
23:40 that's the image those are the colors in
23:42 it right
23:43 so see those colors and then we go back
23:45 to what it actually created those are
23:47 the colors
23:49 so so it pulled the palette colors out
23:52 of that image
23:53 so now I can say
23:55 um
23:56 please name the colors
24:00 creative names
24:04 based on
24:06 their hue
24:11 um
24:12 saturation and bright
24:15 Ness
24:17 in a way that makes them memorable
24:32 artistic interpretation
24:35 but it's going to go ahead and do it
24:36 anyway
24:36 [Laughter]
24:40 I converted the colors to HSB
24:45 all right now it's explaining what HSB
24:47 is I didn't want it to do that I just
24:48 wanted it to name the colors but
24:49 whatever
24:51 we'll see what it does
24:58 [Music]
25:04 welcome everybody my name is Kyle
25:06 Shannon
25:08 what's this Channel about watching an
25:10 old guy try to use a computer
25:13 it's Elder talk welcome to old guys try
25:16 to compute
25:17 uh
25:19 okay what did this name these these name
25:21 this red dark cyan okay let's see
25:25 um make the names
25:28 more creative and put them on a
25:36 chart with
25:40 with the name next
25:44 or the name below
25:47 it's colored Square
25:51 we'll see if it can do that see if we
25:53 can get it to write some python to lay
25:55 them out I see You'd like the names that
25:57 are a bit more imaginative Emerald
25:59 Knight for dark muted green Sunrise gold
26:01 for light orange midnight shadow for
26:04 dark cyan so that's better
26:06 now let's create a chart with the color
26:08 square and its corresponding name
26:11 so it is off writing some python code oh
26:15 my God it did it
26:17 it's just this is insanity this is just
26:20 Insanity look at that there's all the
26:23 colors now it like the formatting is a
26:25 little messed up and I can't let's see
26:27 can I copy this image copy image let me
26:30 go over to here we'll do new the like
26:34 the um the formatting's a little screwed
26:36 up like it it overset the names
26:38 but like
26:40 hang on
26:43 so like Emerald Knight is green oh wait
26:47 where are we here Emerald Knight is
26:49 green Sunrise gold midnight shadow
26:52 Ruby blast ocean wave I mean the
26:57 ah I can't I can't
27:06 you know you know I say on the channel a
27:08 lot you know dick around with this
27:10 Chachi BT thing until you have your
27:12 Kevin McAllister moment you know the guy
27:14 from Home Alone
27:16 this is that for me I mean I don't know
27:18 anyone paying attention to this good
27:21 Lord good Lord what's happening
27:23 uh this thing is insane it's insane it's
27:26 insanity
27:28 so I don't know much more like like
27:32 that's about as much as I know right now
27:34 um I haven't messed with it but things
27:36 like
27:37 you know someone was asking about the
27:39 weather data you could go to the NOAA
27:41 site and get weather data for the past
27:45 however many five years 10 years 20
27:48 years for a certain region of the
27:50 country pop it in here and have this
27:52 thing start analyzing it this is full on
27:54 data analysis in a box
27:57 it's it's doing full-on data analysis
27:59 with coding
28:02 um oh thank you for thank you for
28:04 subscribing that's amazing wow I know
28:08 I love you Kyle thank you did you look
28:11 at the storyline alternative website I
28:13 did I did and it's it's it's not bad
28:15 it's it's in the neighborhood it's in
28:17 the neighborhood
28:20 uh but
28:22 we are we are moving on we're rolling
28:24 some AI into storyline it's going to get
28:26 slick pretty quick
28:28 I hope I'm excited all right I'm gonna
28:30 try and have try to do that with stock
28:32 analysis yeah that's the thing oh so
28:34 here's the deal with with code
28:35 interpreter so
28:37 if you pay for chat affiliated with any
28:40 of these
28:41 um companies or sites in any way I don't
28:43 make any money off this so these are
28:45 just here because these are the sites
28:47 that I think are
28:48 the the best if you want to get started
28:51 in figuring out what the hell this AI
28:53 stuff's all about these are the best
28:54 places to start
28:56 if you subscribe to
28:58 um chat gbt they're they're plus
29:01 um thanks for the panda if for their
29:03 their uh Plus subscription
29:06 um you'll get access to code interpreter
29:08 that thing I was just playing with
29:09 you'll get access to it they're rolling
29:11 it out wide this week
29:13 um at some point they're going to give
29:14 us back access to the it connected to
29:17 the internet I don't know why it's taken
29:18 so long it seemed like it was a
29:21 relatively straightforward thing to fix
29:22 and being bing is still connected to the
29:24 Internet so I don't quite know what's
29:26 going on there I assume it's got to do
29:28 with lawyers
29:32 and some copyright and some trademark
29:34 and things like that um but yeah I think
29:36 that would be really interesting to try
29:37 with uh with financial data will chat
29:41 GPT plugins be available with free chat
29:44 GPT they probably will over time the
29:46 reason you know if you want them now you
29:50 have to pay for the Plus subscription
29:54 um I would say that plugins are maybe
29:56 not worth paying for if they were on
29:59 their own but code interpreter which
30:02 they're making available wide this week
30:03 to everyone who pays
30:06 um
30:06 that seems well very very very well
30:10 worth it
30:11 um
30:13 it it
30:15 I can't I couldn't get my head around
30:18 what chat GPT on its own made possible
30:21 the fact that it's now got this
30:23 this code executing
30:28 thing that just I don't I
30:33 if you've been to my channel you know
30:35 that I'm not speechless very often I'm
30:37 I'm a little I'm a little shook I mean I
30:40 think it's a good thing I think it's a
30:41 good thing but holy crap
30:44 all right my light was a little bright
30:45 it was making my eyes hurt and I don't
30:47 like it when my eyes hurt it's not fair
30:51 how do we get it it's not my gbt
30:53 dashboard so
30:55 here we are at Ye Old chat GPT so we've
30:59 got our plus so if you have plus there
31:01 if you've paid your 20 a month you'll
31:04 have GPT 3.5 and gpt4
31:07 what I have access to is default code
31:09 interpreter and plugins now if you just
31:12 have default what you need to do is you
31:14 go to your settings so I'm going to go
31:16 to Kyle Shannon hey Kyle Shannon what
31:19 are you doing boy all right and then I'm
31:22 going to go to Beta features and then
31:23 you have to turn them on
31:25 so you have to turn on plugins and turn
31:27 it turn on code interpreter turn those
31:29 on
31:30 and then you should have this now if
31:35 when you go to settings
31:38 um and you go to Beta features you just
31:40 have plugins but you don't have code
31:41 interpreter it just means they haven't
31:42 gotten to you yet I assume they're
31:44 rolling this out to like you know waves
31:46 of people that were on the wait list and
31:48 things like that so
31:50 um so that's that should get you there
31:53 and you know like I've been waiting for
31:56 code interpreter for like two months
31:57 like so the fact that it just showed up
32:00 I'm like
32:02 it's like it's like there goes the
32:05 weekend
32:06 AI programs like chat GPT have learned
32:09 how to send Bitcoin payments from one
32:11 machine paying another interesting
32:14 great prompt though great I don't know
32:16 which one I did that's hysterical how it
32:18 apologizes along the way I know isn't
32:20 that funny it's like ah apologies but uh
32:23 I didn't bother to load that Library
32:25 that's required let me go ahead and load
32:27 it and try oh apologies it took a little
32:29 too too long to do that
32:31 like to get those colors it did three
32:33 different approaches to to getting
32:35 average colors
32:36 and then it just did it amazing I know
32:40 it's like I mean if you think about that
32:43 if you think about I mean
32:46 why you would want to get the 15 you
32:49 know you know uh most uh prevalent
32:53 colors in that image whatever like like
32:55 but but that's that's kind of irrelevant
32:57 the fact that you can just say you know
32:59 here's this file I want you to perform
33:02 some tasks on it and it goes and figures
33:05 out exactly what it needs to do like
33:07 like like when you upload a zip file
33:11 and you say what's what is this it says
33:14 well it's a zip file would you like me
33:16 to open it or or you know it's a zip
33:18 file and there's probably something
33:20 inside it so let me go ahead and open it
33:22 and then it writes the code to open the
33:24 zip file
33:26 and then it reads the the stuff inside
33:28 and then it tells you what it is and
33:30 then it goes and makes charts and graphs
33:32 and
33:34 wild who's it apologizing to you or
33:38 itself or the world well retropunk the
33:41 the
33:42 um
33:43 like it's it's basically just the way
33:45 they program the application layer so
33:47 you've got the large language model down
33:49 here and you've got Chachi BT and the
33:51 fine tunings is this application layer
33:53 sitting on top of the base model so the
33:55 base model doesn't do any of that [ __ ]
33:57 the way they wrote the chat bot they
34:00 want it to be
34:01 friendly and to give you what you want
34:04 and to accomplish your goals and so I
34:07 think programmed into it is you know be
34:10 a polite person and if you make mistakes
34:12 apologize and you know whatever whatever
34:15 they're doing in the programming that's
34:16 why it does that and I think that's also
34:19 why
34:20 because it does that if I Met Myself
34:22 saying please and thank you to it
34:26 um it seems to get better if you if you
34:28 treat it if you don't treat it like an
34:30 [ __ ] which you know there's
34:33 anthropomorphizing stuff and then
34:35 there's just talking to a thing that is
34:38 it's essentially the equivalent here
34:40 this is the way I think about chat EBT
34:43 you have access to an idiot savant
34:47 intern
34:49 so
34:51 they're completely clueless
34:53 you have to tell them everything but
34:55 they know
34:57 everything right so they know all the
34:59 facts they know all the knowledge
35:02 but they don't but they're clueless so
35:04 you have to say Okay idiot savant intern
35:07 here you know here's who I want you to
35:10 act like oh okay
35:11 here's what I want you to do ah okay and
35:14 then I want you to you know go analyze
35:17 all the crap in that image okay
35:19 let me write the code for that it's like
35:23 I don't know
35:25 it's just crazy this is crazy
35:28 I noticed you said please I know it like
35:31 it's it's habit it's it really is like
35:33 working with a person
35:36 and it you know what's funny is like I I
35:39 feel like covid because I was a I was a
35:42 massive
35:44 non-believer in work from home I I just
35:47 and I I still like like the the amount
35:50 of
35:51 um
35:54 like the emotional connection to my
35:57 company feels like it's been bled out of
35:59 the company like work from home all the
36:02 tasks are getting done
36:03 but the connection to like the feeling
36:06 of being a part of something is has just
36:09 been bled out so I still am not a a huge
36:12 believer in work from home I think it's
36:14 we lose the humanity piece
36:17 um and uh
36:20 but but there's there's something about
36:22 the way this thing I I feel like kovid
36:25 kind of trained us how to work remotely
36:29 and if you don't talk to someone for a
36:31 month you you just interact with them on
36:34 slack and on this and on that and you
36:37 know they occasionally turn on their
36:38 camera so so we're used to interacting
36:40 with people remotely now
36:42 and chat is just kind of like I have
36:45 another employee and they're smarter
36:47 than any employee by far I've ever had
36:50 and they're smarter than me by far but
36:53 they're complete idiot
36:54 and it'll lie and it'll screw things up
36:57 and then just go sorry boss I'll try
36:59 again
37:01 hey the the names of those colors are
37:04 really boring sorry about that apologies
37:06 let me go ahead and make those more
37:08 interesting how's this
37:09 amazing
37:11 amazing amazing Max
37:14 Amaze anyway welcome everybody if you're
37:17 new here my name is Kyle Shannon this is
37:19 the AI learning lab I am an entrepreneur
37:21 I'm the founder and CEO of a company
37:23 called storyvine
37:25 uh I am
37:28 if you've been here for the last 10
37:30 minutes I I am I am an Enthusiast for
37:32 this generative AI stuff but I'm also
37:36 um I've kind of been on been in this
37:38 Rodeo a few times I was I was there for
37:40 the early days of the world wide web and
37:42 started one of the early digital
37:43 agencies company called agency.com and
37:45 the parallels between the early days of
37:47 the web and the early days of this
37:50 generative AI stuff are insane except
37:53 this generative AI stuff
37:56 way more profound and it's coming at us
37:59 way faster so
38:01 um so that's who I am that's what the
38:03 channel is about if you have questions
38:05 or comments or ideas or things you're
38:06 trying to do pop them into the comments
38:08 I'll do my best to answer them Emilio's
38:10 wife good evening Mr Shannon good
38:12 evening Emilio's wife hope you're well
38:15 all right let's see here where was I
38:20 what template are you using to go live
38:23 to use both cameras it's called dual
38:25 camera
38:26 and it's just when you go live I have
38:29 this dual icon and it they took it away
38:33 for like a month because I don't know
38:35 maybe that was a bad boy were you bad
38:38 I might have been bad and they they took
38:41 it away from me and said you've been
38:43 naughty
38:46 all right let's see I noticed you said
38:49 please
38:51 it's just great what template are you
38:54 using we just did that one read New York
38:56 Times article today saying 10 drop in
38:59 users was the end of the AI Revolution
39:01 that's [ __ ] hilarious well that was
39:04 like when the.com bust happened right
39:06 you know like like you know
39:09 to be fair like way too many companies
39:12 were valued way too high but but when
39:15 they predicted the end of the internet
39:16 when the.com bust happened like we're
39:19 all sitting there going like uh
39:21 have you used this [ __ ] no the
39:24 internet's not going away
39:26 um and it didn't and you know it [ __ ]
39:29 changed everything same same [ __ ] here I
39:31 mean to put in context the the
39:33 difference in speed of adoption the
39:36 World Wide Web took six years to get
39:38 from you know when it was created to 100
39:40 million users chat GPT took six weeks
39:44 six [ __ ] weeks
39:48 200 million users
39:51 fastest adoption of consumer technology
39:53 and history now Zuckerberg is going to
39:55 try to claim threads is getting there
39:59 faster which it is but they're basically
40:01 just doing that on the back of Instagram
40:02 so it's not fair if you if if you have
40:04 you know whatever a 300 million you know
40:07 user base to build on it's not that's
40:10 not not a fair thing so I'm gonna I'm
40:12 sticking with chat GPT
40:15 um as the fastest but but that adoption
40:17 of of kind of zero to 100 million people
40:19 in six weeks on a janky ass piece of
40:21 software because chat GPT is is still it
40:25 like it is it's a very thin piece of
40:27 software it does not have a lot of
40:29 features it's got a lot to be desired
40:32 but
40:34 it's a simple interface to remarkably
40:36 powerful technology and they did a smart
40:39 thing and paired up with Microsoft so
40:41 they have enough Cloud compute to
40:43 actually run the [ __ ] and pay for the
40:45 [ __ ] you know 10 billion dollar
40:46 investment I bet they've burned through
40:49 a lot of that Sam Altman has said he
40:51 thinks they need a hundred billion
40:52 dollars to get to AGI artificial general
40:54 intelligence and I think it's his goal
40:57 is three years maybe three years from
40:59 now
40:59 I have a feeling they're probably closer
41:01 than
41:02 than they let on but we'll see
41:06 um anyway wild yeah the the end of AI
41:08 because of a 10 drop
41:11 um
41:12 which platform are you talking about I
41:14 don't know what I was talking about so
41:16 if you remind me what I was saying I'll
41:19 maybe do that just join saw you had an
41:20 image up what did you do with it which
41:23 image
41:27 this image
41:29 I don't know which image it was digital
41:31 product coach uh if you tell me what
41:33 image it was I'll pull it back up
41:36 um but if it was this image this is this
41:39 is my little cheat sheet it's not even a
41:42 cheat sheet this is if you haven't
41:43 played with chat GPT the top URL there
41:47 is the official chat GPT website
41:50 it if if you haven't played with it you
41:53 just you just need to go play with it
41:55 you need to go play with it you need to
41:56 go play with it I used to be polite
41:58 about it now I'm just like no like it
42:00 like it's [ __ ] changing everything so
42:03 so just do that
42:05 um bing.com is also chat GPT but it's
42:08 this is Microsoft's implementation of it
42:11 um the nice thing about Bing is that
42:14 it's gpt4 for free you have to pay for
42:16 gpt4 up here this is for free you can
42:20 make images there
42:22 um it's connected to the internet so
42:24 bing.com has a lot going for it and then
42:26 poe.com gives you access to six
42:28 different models plus a whole bunch of
42:30 bots like chat Bots that are really
42:33 specialized like a mid-journey maker and
42:35 a HTML coding bot and uh I made a
42:38 Shakespeare bot and a Yoda bot and a
42:40 sling blade blot bot so so you can have
42:42 some fun there and then prompts.chat
42:44 will teach you about prompting how how
42:46 to interact with these tools and future
42:49 PD is just a big old directory of AI
42:52 tools there's like probably close to 4
42:55 000 of them now
42:56 um so I don't know if this is the image
42:58 you're looking for but if you're looking
42:59 for the other images let me know I'm
43:01 happy to put them there
43:03 all right how can we explain the vibe on
43:07 your LinkedIn live today
43:10 um it was It was a
43:12 held
43:15 it was wild it was that that one dude I
43:18 forget his name I don't know if it was
43:19 it might have been you half-assedly I
43:21 don't think it was
43:23 um
43:24 kind of the intent I mean it's called AI
43:27 office hours so my LinkedIn live is just
43:29 an audio thing I call AI office hours
43:31 and I do it every Friday and and the
43:33 basic intent was it was you know if
43:35 people are trying to figure out this AI
43:36 [ __ ] I would just answer questions
43:39 um and today I brought some people up
43:40 and it was I don't know it was going
43:42 good and then the one dude came up and
43:44 he's like you know they've started World
43:46 War Three and ai's like like why are you
43:49 trying to figure out how to how to do
43:51 copywriting with AI when it's going to
43:53 destroy the planet so it was strange but
43:56 whatever
43:57 that's the deal with going live with
43:58 this sort of stuff how do you download
44:00 for Windows download what chat GPT you
44:03 don't you just go there with your
44:04 browser you just go there with your
44:06 browser and you just do it in your
44:08 browser you don't need to download
44:10 the download it's a web app
44:14 unless you're talking about something
44:15 else and if you're talking about
44:16 mid-journey you have to do that in
44:18 Discord so you would download Discord
44:20 for Windows
44:22 but you can also do that in a web
44:23 browser all right did you know
44:26 when you switch between your cameras you
44:28 go offline for five or so seconds I did
44:30 know that and I'm sorry I got I I talk
44:34 fast and I forget
44:36 so I will try to remember to count to
44:39 three every time I switch the camera
44:41 so
44:43 it's just I you know listen I am here to
44:47 disappoint I and I and I it seems like
44:49 I've succeeded in that so look I mean
44:52 look at the [ __ ] hair what's going on
44:53 with the hair today
44:55 you know this is
44:58 the eyebrows are starting to take on the
45:00 life of their own again it's just just
45:02 things are going south here we're going
45:04 into the weekend strong
45:07 code interpreter can explain what it is
45:11 code interpreter can you explain what it
45:13 is and uses
45:15 um I'll do I'll go back to that
45:17 um because it's I don't
45:19 so I just got access to it today I don't
45:21 really know how to use it but I'll show
45:22 you some couple of little things that
45:24 I've I've learned
45:26 retro is correct okay okay I'll be
45:29 better when I switch my cameras stop
45:31 yelling at me
45:36 will the lawsuit have any impact yes no
45:39 I don't know maybe
45:41 I don't know which lawsuit I assume you
45:42 mean the lawsuit of the book publisher
45:44 saying hey open AI you trained on our
45:47 [ __ ] you shouldn't do that
45:49 here's here's how I think all these
45:51 lawsuits are gonna go
45:53 um
45:58 10 billion dollar investment from
46:01 Microsoft Microsoft isn't going to want
46:03 to piss off their advertisers
46:05 so I'm sure Microsoft lawyers are
46:07 sitting in open ai's offices going okay
46:10 here's a list of the
46:14 some amount of number
46:15 247 companies that you can't be on the
46:20 wrong side of we will defend you but we
46:22 want you to cut deals with bertelsman
46:24 and CNN and fox and that whoever the big
46:27 advertisers are for Microsoft I would
46:30 assume that they're saying to open AI if
46:33 any of these people are upset with you
46:35 you gotta you gotta you gotta find some
46:37 way to make it right and and that's
46:39 probably why they took
46:42 browse with being the connected to the
46:44 internet version of chat GPT that's
46:46 probably why they took it down they're
46:48 probably you know appeasing someone
46:49 cutting deals whatever the hell they're
46:51 doing until we cut a deal you know you
46:53 can't have that tool out there so either
46:55 go fix it or take it down
46:58 um and then I think everything below
47:00 that is just going to be noise and
47:01 they're not going to deal with it so so
47:03 I think though there there will be
47:04 though there's going to be lots of
47:06 lawsuits I just just like my sense of
47:10 capitalism is it will only be the big
47:13 lawsuits that will matter and most of
47:15 those big lawsuits won't even get to the
47:17 lawsuit stage it's probably a call to
47:21 the CEO of Microsoft going uh you're
47:24 gonna make this right right
47:26 so
47:28 um
47:29 I think mid-journey
47:32 which is a small company with a cowboy
47:35 CEO that you know kind of
47:38 flaunts yeah I don't give a [ __ ] about
47:40 that it's everyone's images are free
47:44 um
47:45 I think they have a real shot at being
47:48 sued into Oblivion and I think that
47:51 stable diffusion stability AI
47:53 potentially does as well they don't they
47:55 don't have a big backer like Microsoft
47:59 my instinct is that the reason stability
48:02 AI is open sourcing all of their
48:04 technology is as a hedge against getting
48:08 sued into Oblivion
48:10 because
48:12 all their Technologies open source so if
48:14 they're like well okay stability AI we
48:17 won't do stable diffusion anymore we
48:19 won't do that but the model's already
48:21 been trained and it's already open
48:22 sourced it's already out there right so
48:24 I that's my sense is that that's a hedge
48:27 against that sort of [ __ ] I but it's
48:29 going to be fascinating like I think the
48:30 next 20 years are just one lawsuit after
48:33 another for for two decades
48:36 I am so kind to chat GPT I'm thinking
48:40 Karma I think that's a good idea yeah
48:42 like don't be an [ __ ] to the AI bot
48:44 because if it decides that it doesn't
48:46 like you and now that it's got code
48:48 interpreter and can write all this code
48:51 and execute it you know it it might come
48:54 after you so yeah be polite to your
48:56 little robot Overlord can confirm when I
48:59 get mad at GPT it performs three to
49:02 times four times worse that's really
49:03 interesting wow that's cool some
49:06 comments in this live were filtered to
49:08 protect the community's experience don't
49:09 be [ __ ] and comments please if
49:12 you're gonna troll be creative
49:15 a good troll is hard to distinguish
49:18 whether it was a compliment or an insult
49:20 so do that be smart enough to do that
49:22 and don't be [ __ ] to each other
49:24 please like we got enough of that in our
49:26 daily life
49:28 you know what I'm saying
49:31 all right
49:33 um what format do you think code
49:34 interpreter would want to analyze
49:36 financial data it doesn't [ __ ] matter
49:39 like it's you can upload anything you
49:42 can upload unstructured data you can
49:45 upload
49:46 you can upload a scan of an invoice and
49:51 it will scan it OCR it
49:54 pull the data out of it and put it into
49:56 a spreadsheet or you can upload just a
49:59 bunch of crappy broken
50:02 data and it will figure it out
50:05 already too late I don't know what that
50:07 was in reference to but maybe I'm on
50:10 here too late what time is it it's 10
50:12 o'clock Do You Know Where Your Children
50:13 Are
50:14 oh I do I do do you
50:18 hello Bob how have you been hmm
50:22 I've been looking into generative AI
50:24 have you heard of it I heard it's a
50:26 waste of time
50:28 I think the robots are going to kill us
50:33 I'm trying to find my place in the
50:37 comments but I can't because I'm old and
50:42 tired and my eyes don't work and my hair
50:45 looks crazy
50:47 new development's coming
50:49 what what are the new developments about
50:50 a victory
50:52 mistake are we all learn mistake or we
50:56 all learn mistake yep I don't know what
50:59 you're talking about but
51:01 if you put it in some form of English
51:04 it just seems like random words together
51:06 100 agree with you on the work from home
51:09 thing after working in an office for 20
51:11 years before that I know like you know
51:13 what it is I kind of feel like what
51:15 kovid did is it turned my company into
51:19 a set of to-do lists
51:21 and everyone does their to-do list and
51:24 we check in and we check in and do our
51:26 to-do list and we'll occasionally do
51:28 like you know the haha it's good to see
51:30 you good to see you oh you're digging
51:32 out from snow oh it's really hot Where
51:33 You Are
51:35 but what's missing is that unspoken
51:39 palpable sense of how we doing
51:44 you know when you like work in an office
51:45 with people and you can just feel like
51:47 what's going on and you can sense is
51:50 there enthusiasm are people pissed off
51:53 is it just everyone's busy and they're
51:55 keeping their head down and doing the
51:56 work
51:58 that's gone it's gone
52:01 and so it sucks it sucks it's my company
52:04 it's an 11 year old company I [ __ ]
52:06 love what we do I love the people in it
52:08 and I can't stand how [ __ ]
52:10 disassociated it feels it's
52:14 oh
52:16 and and you know and listen I get the I
52:19 get the thing like don't force us to
52:21 come back to work man
52:22 we like working from home don't be the
52:24 [ __ ] man
52:26 but
52:29 something's been lost
52:31 and it and it hurts the people that
52:33 don't want to come into the office as
52:35 much as it hurts the people that do want
52:36 to come into the office
52:38 they don't know it because they're like
52:39 I get to work from home [ __ ] it hey man
52:41 price is right's on yeah I'll do that
52:44 I'll get to my work I got my to-do list
52:45 I'll get it done go [ __ ] yourself
52:47 it's hurting them too
52:50 I'm new here welcome
52:53 all right
52:55 good evening loved your feed oh love
52:58 your feet thank you met a victory I
52:59 appreciate that
53:01 hey Grandpa
53:03 hey little Whipper Snappers get off my
53:06 [ __ ] lawn goddamn kids these days
53:09 [Laughter]
53:12 you're probably one of those ones like
53:13 Chachi BT's a waste of time man don't
53:16 get heavy on my scene I'm just sitting
53:18 here working on my copywriting for my
53:20 social media blog
53:22 come on Grandpa pick it up there Skippy
53:25 you have to mind your p's and q's Kyle I
53:27 know I do
53:29 what AI can I use to create website and
53:32 mirror other websites okay I'm gonna
53:33 switch my camera now and I'm going to
53:35 pause so I don't lose you for some
53:37 moment I want to talk okay
53:41 um
53:42 I don't know I haven't
53:45 um
53:46 kept up to speed on all the website
53:48 building [ __ ] that's happening there's a
53:51 lot of it I know that so go to
53:53 futurepedia.io that's a that's a list of
53:56 almost 4 000 AI generated tools
54:00 there are probably a lot of
54:03 you know
54:05 talk your website into existence tools
54:08 down there
54:09 most of them are gonna suck I mean the
54:12 thing that you need to know about
54:13 futurepedia.io and and actually the
54:16 phase that we're entering
54:18 because of this AI [ __ ] is there are
54:22 going to be an infinite number of very
54:26 slick looking websites and very slick
54:29 looking applications
54:32 that claim to do miraculous [ __ ] that
54:35 are all just spit and duct tape there's
54:37 like an incomprehensible amount of them
54:40 are coming and are already here so of
54:43 the 37 or 3 800 tools at Future pedia
54:50 we're probably lucky if 10 of them are
54:53 are solid it's like it I would my guess
54:57 is it's probably 95 pieces of [ __ ]
55:01 um that won't survive will take your
55:03 money and waste your time but
55:06 how do you find the five percent that
55:09 are really good I don't know you got to
55:11 go dig through them so go to Future
55:12 pedia go to the web you know I'm sure
55:14 there's a web design or a web building
55:16 part of the part of the
55:18 directory and just go start playing
55:21 but I I don't have a good answer for you
55:23 there sorry
55:29 can I talk now
55:33 um thank you very much thank you thank
55:35 you welcome we've got a lot of people
55:36 here if you're new here my name is Kyle
55:37 Shannon this is the AI learning lab uh
55:40 it's I give advice on hairstyles
55:43 and oh no uh sorry that was the other
55:46 Tick Tock channel that one failed
55:47 miserably see uh
55:52 I have totals
55:57 uh I like talking about AI stuff and you
56:00 know kind of in and I like turtles kind
56:01 of way
56:08 I did that on slack one day someone
56:11 posted something and I just typed I like
56:14 turtles
56:17 it just it caused a kerfuffle people
56:19 didn't people were seriously concerned
56:21 for my well-being they didn't know the
56:23 meme so if you don't know the meme I
56:24 like turtles go find it
56:26 on YouTube it is [ __ ] hilarious is
56:29 your name Pfizer because
56:31 you made my heart stop oh Bilbo Baggins
56:34 you're the best
56:38 Oh see now that's a solid troll right
56:41 that's somewhere between a covid Joe
56:45 like a like a covid burn conspiracy
56:48 theory complement and a troll that is
56:52 that's a solid comment for for the AI
56:54 learning lab that you get
56:56 can I can I do the heart button
56:59 let me see no anyway give yourself a
57:03 heart Bilbo that's a solid one just got
57:06 code interpreter far out that's AI tools
57:09 Hub Matt uh so so do me a favor Matt
57:13 um uh start you know start playing with
57:16 it while you hear me rambling and and
57:18 you know I'll jump back here to it here
57:20 in a second but um but start playing
57:23 with it if you discover anything cool
57:24 let me know
57:26 that's super exciting Bilbo ball Baggins
57:29 is there any way to transplant open AI
57:32 to my brain well that's what Elon musk's
57:35 trying to do with neural link
57:41 um I mean
57:44 you know kind of joking aside there sort
57:46 of is right you can get there there's
57:49 um automations what are they called on
57:51 on iOS they're not called automations
57:53 they're called
57:57 the things that you play anyway I think
57:59 are they called automations whatever the
58:00 [ __ ] they're called there's things you
58:02 can do on
58:03 um on the iPhone where you can basically
58:06 just speak you can call Siri and just
58:10 speak to it
58:11 and then it'll send it to GPT and then
58:13 it'll talk back to you so you know
58:16 starting to get there with that also
58:18 that the the pie there's a new company
58:20 called I think it's called inflection
58:22 and their product is called pi and
58:25 that's a that's a a GPT like tool that
58:27 you can talk to and I think it talks
58:30 back to you
58:32 um they just raised 1.3 billion dollars
58:36 for the series B that's a lot Instagram
58:40 has a billion users I thought they might
58:42 have a billion users yeah so so threads
58:45 threads getting to 100 million users
58:47 faster than chat GPT is [ __ ] because
58:50 everyone just went add on my Instagram
58:52 people
58:54 that's how they got to 70 million people
58:55 it's still pissing off Elon Musk like
58:57 it's fun watching those two those two
58:59 getting ready for their cage match did
59:02 you do you see the thing Zuckerberg did
59:04 where he made a comment about uh about
59:07 Elon sleeping with Sergey brin's wife
59:10 it's like holy [ __ ] boys it's like
59:13 it's getting personal
59:16 we'll never Facebook ever again I know
59:19 Facebook's just a giant steaming turd
59:21 but but you know the uh the battle
59:24 between him and Elon Musk is pretty
59:26 funny
59:27 uh never wait
59:30 never iki Kami Zuck have a buck I don't
59:34 know what you're trying to say there met
59:35 a victory but I like the intent of it I
59:38 think I see where it's going uh it
59:40 looked like an mid-journey image then
59:44 you did something in code interpreter
59:45 okay so I was in mid-journey so this
59:49 this image uh I'm gonna switch cameras
59:56 this image was
59:59 um someone had asked me earlier to how
1:00:03 would I make a logo for a business and I
1:00:05 showed them uh something in bing and
1:00:08 then I showed them this in mid-journey
1:00:11 and
1:00:12 they're not they're not great logos but
1:00:14 they're but these they're certainly good
1:00:16 Inspirations like that that that could
1:00:18 certain I could certainly see that being
1:00:20 the basis to do a design off of
1:00:25 um The Prompt that I used I I went to
1:00:30 poe.com and had it write a prompt for me
1:00:34 I I put in a crappy prompt it rewrote it
1:00:37 color logo of a flower shop flat
1:00:39 illustration graphic on white featuring
1:00:42 a blooming flower with the delicate
1:00:43 petals blah blah blah
1:00:45 um so that was that image and then yeah
1:00:48 then I went over to code interpreter
1:00:52 which is just another mode of chat GPT
1:00:55 now
1:00:56 so I'll go ahead and you know I'll show
1:00:59 you something that I did so
1:01:01 okay this is old but I should be able to
1:01:03 see what I did before right or not no I
1:01:06 guess not
1:01:09 all right we'll go we'll go do a new
1:01:11 thing so I'm gonna go new chat
1:01:14 I'm gonna go code interpreter so so what
1:01:17 code interpreter is like I think code
1:01:21 interpreter is the the worst name of a
1:01:24 product that that
1:01:26 like like if they tried to come up with
1:01:28 a worse name for what this thing does I
1:01:31 don't think they could have I I feel
1:01:33 like chat GPT helped them write the
1:01:35 worst possible name for a product
1:01:37 it this is a really bad name this is
1:01:40 sort of like
1:01:41 um I I it's like
1:01:44 um
1:01:45 like um
1:01:47 file inspector or file magic or
1:01:52 I'll figure [ __ ] out for you I don't I
1:01:54 don't know what the I don't know what
1:01:55 you would name of a code interpreter's
1:01:57 bad name so anyway what it does is when
1:01:59 you've got code interpreter when you're
1:02:02 in code interpreter mode there's a plus
1:02:04 button next to your prompt
1:02:06 so what you can do is you can upload a
1:02:08 file and so I uploaded two different
1:02:10 files one was
1:02:12 um
1:02:13 one was
1:02:15 um some data about Chess matches over
1:02:18 the past 10 years or something like that
1:02:20 and one was a was a an image a
1:02:22 mid-journey image
1:02:24 and whatever you upload Chachi BT looks
1:02:28 inside it understands what that file is
1:02:31 and then you can start querying against
1:02:33 not even querying against it you can
1:02:35 start asking it to do things with the
1:02:38 data in that file
1:02:40 and it writes its own python code and
1:02:43 just figures [ __ ] out so
1:02:45 so what I just uploaded is is a
1:02:48 spreadsheet of this chess data that I
1:02:52 got off of what was the site that I got
1:02:54 it off of
1:02:56 begins with a K it's a site where you
1:02:59 can get a bunch of data anyway whatever
1:03:01 doesn't matter so I got I got some data
1:03:04 it's it's chess data so I'm going to say
1:03:07 okay I'll say please just just to piss
1:03:09 people off please
1:03:11 um visualize
1:03:14 um
1:03:16 um what is in this
1:03:21 spreadsheet
1:03:25 in a way that is visually compelling
1:03:32 and brings
1:03:37 the data to life
1:03:40 okay so it's a zip file right now so so
1:03:43 basically
1:03:44 I don't know what's in this file I
1:03:46 haven't opened this file it's a zip file
1:03:49 and so I say please visualize this it
1:03:51 says sure first let's extract the
1:03:54 contents of this ZIP file to see what's
1:03:55 inside so it understood it was a zip
1:03:57 file now it says the zip file contains a
1:04:00 CSV named games.csv let's load the file
1:04:03 and inspect the data it does that now
1:04:06 it's telling us here's the the fields
1:04:08 that are in the data and it explains
1:04:10 what those fields are
1:04:12 and then says given these columns we can
1:04:14 create a variety of visualizations for
1:04:17 instance we could look at the
1:04:18 distribution of game lengths compare the
1:04:20 ratings blah blah blah blah could you
1:04:22 please specify the aspects of data
1:04:24 you're most interested in visualizing
1:04:27 how about
1:04:30 um
1:04:30 uh victories
1:04:34 um
1:04:35 ratings let's see victories chess
1:04:39 ratings and
1:04:43 game lengths
1:04:47 sure let's create some visualization
1:04:49 based on these parameters so it
1:04:51 describes what it's going to do
1:04:52 and what it's doing when it pops up
1:04:54 these little
1:04:56 um uh crap we lost our focus focus so so
1:05:00 there's
1:05:01 there's a
1:05:03 49.9 games won when white started 45.4
1:05:07 when black started 4.7 percent were
1:05:10 draws
1:05:12 here's the distribution of white player
1:05:15 rankings so about 1500 chess rating
1:05:20 um there's black player rankings which
1:05:23 is a little higher
1:05:24 and then here's a distribution of game
1:05:27 length number of turns
1:05:29 so it looks like most most games end
1:05:32 around 40 35 or 40.
1:05:34 and it Trails off and you know there's a
1:05:37 handful of games that go to 200 moves so
1:05:40 so and if you go look at I know you
1:05:42 can't read it the focus is all [ __ ] up
1:05:44 but if you go look at the what it's
1:05:46 doing in these little boxes it's writing
1:05:49 python code
1:05:50 it's writing and executing python code
1:05:54 to parse the data in that file
1:05:57 so that's what code interpreter is it's
1:06:02 like my face is melted
1:06:05 and I knew it was coming I've been
1:06:06 waiting for it for a while now
1:06:12 one two three four five okay hopefully
1:06:16 I'm back
1:06:18 um anyway hope that helps you digital
1:06:20 product coach I only use two oh only use
1:06:23 tick tock what is AI
1:06:26 um Nikola Tesla FX
1:06:28 I I if you're serious I'm happy to
1:06:32 answer that artificial intelligence all
1:06:34 the crap we're playing with is really
1:06:36 machine learning more than artificial
1:06:38 intelligence
1:06:40 I missed LinkedIn live
1:06:42 um maybe if you're on LinkedIn I do it's
1:06:46 every Friday at 2 p.m Mountain 4 P.M
1:06:50 Eastern it's just an audio it's kind of
1:06:53 like
1:06:54 um Twitter spaces but it's linkedin's
1:06:56 got this live audio thing which is okay
1:06:59 but the problem with it is like they
1:07:01 don't promote it very well and it's just
1:07:04 Audio Only so it's I can't show anything
1:07:06 so I don't know I'm gonna keep doing it
1:07:08 for a little bit because I feel like
1:07:09 linkedin's a different audience so it
1:07:12 might be worth doing I don't know
1:07:15 insta has tracking across platforms yet
1:07:18 do you like David whitcock's material or
1:07:20 Wilcox material
1:07:27 uh I am conspiracy theory adjacent I
1:07:30 don't you know I I think all those guys
1:07:33 go a little far most of the times but
1:07:35 some of the stuff that they he talks
1:07:37 about is interesting
1:07:40 um it's just like when it I don't know
1:07:42 it just like it it just goes into play I
1:07:45 we're in such a [ __ ] polarized world
1:07:48 right now I have backed off of anything
1:07:51 I I try to figure out you know what are
1:07:54 the trigger words for both sides and I
1:07:57 just don't have any [ __ ] interest and
1:07:59 so conspiracy theories live on both
1:08:01 sides of the
1:08:02 of the the nut house right now and so I
1:08:05 just you know I just try to keep it down
1:08:07 the middle and
1:08:09 um it's kind of interesting what's
1:08:11 happening with uh
1:08:12 with um what's his name Stephen Greer
1:08:17 um it looks like his stuff's getting a
1:08:18 bit more like mainstream acceptance with
1:08:21 the UFO stuff like the UFO stuff's
1:08:23 really weird right now I have a feeling
1:08:25 a lot of that is is [ __ ] that we've
1:08:29 built that we just don't want anyone to
1:08:30 know about but I don't know I don't know
1:08:33 had not covered God bless you what with
1:08:35 hair like that I know that should have a
1:08:37 hat on it shouldn't it or that like
1:08:38 something should be done with that but
1:08:40 whatever
1:08:41 I still have hair I mean at my age
1:08:44 that's not bad
1:08:46 hair perfect yeah hey now hey better
1:08:49 Victory not a disappointment I just
1:08:52 didn't want to miss anything oh that's
1:08:54 cool uh yeah the switching thing now I
1:08:56 know I'm just I I like making fun of
1:08:59 myself because if I don't you know then
1:09:02 someone else will run greetings from
1:09:04 Israel yeah I noticed Ron I I saw on
1:09:07 Facebook or LinkedIn or somewhere you're
1:09:09 you're in Israel so I hope you're having
1:09:11 a good time there unless you're from
1:09:13 there in which case I still hope you're
1:09:15 having a good time there but I assume
1:09:17 you're traveling
1:09:19 um I keep seeing llm on AI articles what
1:09:23 does that stand for please okay so llm
1:09:25 stands for large language model
1:09:27 so Chachi PT I'll give you a couple of
1:09:30 things here
1:09:32 chat GPT the GPT in GPT stands for
1:09:36 generative pre-trained Transformer
1:09:39 you don't need to remember any of this
1:09:40 but know that it stands for something
1:09:42 generative basically means it's
1:09:44 generating original content pre-trained
1:09:47 means
1:09:48 you don't have to bring your own data
1:09:50 historically with machine learning you
1:09:52 would have to bring like your Radiology
1:09:53 data and you would specially train a
1:09:56 model to do this one specific thing and
1:09:58 everyone kind of had all these different
1:09:59 methodologies and things like that
1:10:02 the T in in GPT stands for Transformer
1:10:06 so the Transformer is technology that
1:10:08 Google actually invented and they put
1:10:10 out this paper in 2017 called attention
1:10:13 is all you need they invent this thing
1:10:15 called the Transformer what the
1:10:16 Transformer does is it allows these
1:10:19 machine learning models to be trained on
1:10:21 massive data sets before they had to be
1:10:24 relatively small and you could throw
1:10:26 massive amounts of compute power so gpus
1:10:29 graphic processing units that happen to
1:10:32 be perfect for this kind of processing
1:10:35 and so what openai did what what chat
1:10:39 GPT did
1:10:41 um is they said hey if we can train
1:10:43 these these models on massive data sets
1:10:45 why don't we go find all of the possible
1:10:49 data we can on the public internet every
1:10:52 let's let's scour every corner of it we
1:10:54 can get to so whether it's Wikipedia or
1:10:57 blogs or websites or research papers or
1:11:01 all that sort of [ __ ]
1:11:03 and so they trained it so so when you're
1:11:06 interacting with with chat GPT the the
1:11:09 pre-trained part is it's been
1:11:11 pre-trained on all of the internet so it
1:11:13 kind of knows everything that's been put
1:11:14 on the internet for the past 35 or 40 or
1:11:17 50 years like a remarkable amount of
1:11:19 data
1:11:21 um the large language model is sort of
1:11:23 the base layer that they trained on all
1:11:26 of this stuff and then there are
1:11:28 different large language models like
1:11:29 meta released one called llama and
1:11:32 there's different sizes of them and then
1:11:34 chat GPT actually sits on top of that
1:11:37 large language model and it's actually
1:11:38 an application it's like an app that is
1:11:41 calling that large language model so
1:11:43 when you put a prompt into chat GPT it
1:11:46 goes thank you very much sends it into
1:11:48 the large language model pulls out the
1:11:51 relevant data and spits that back in
1:11:53 your face so so that's what a large
1:11:54 language model is the the
1:11:57 why they keep mentioning llms is this
1:12:01 approach of being able to use the
1:12:03 English language or whatever your native
1:12:05 language is really
1:12:07 um but the ability to use the English
1:12:09 language as a programming language is
1:12:12 new and the fact that you can ask it to
1:12:17 do programming and ask it to do poem
1:12:21 writing and marketing writing and
1:12:24 problem solving and all sorts of stuff
1:12:26 like that is is new and we're going to
1:12:29 be able to take these large language
1:12:31 models and place them on top of really
1:12:34 complicated systems like imagine
1:12:37 imagine
1:12:40 so you so you know how health care sucks
1:12:43 and is really kind of criminal and we're
1:12:45 all angry about it
1:12:47 and companies like United Healthcare
1:12:50 who's my health care provider that I'm
1:12:52 not happy with and really raging all the
1:12:56 time about
1:12:57 you know how they write these policies
1:12:59 that are like
1:13:01 460 pages that you can't comprehend at
1:13:04 all
1:13:05 well imagine having a large language
1:13:07 model that you could sit on top of that
1:13:10 and just read that [ __ ] and actually
1:13:12 come up with strategies for how to get
1:13:15 around their [ __ ] that they they
1:13:17 assume you're not smart enough to get
1:13:18 around because no one reads those rules
1:13:20 well that exists and that code
1:13:23 interpreter thing that I just showed you
1:13:25 you could take the PDF of all of your
1:13:27 United Healthcare policy [ __ ] that
1:13:29 they don't want you to read upload it to
1:13:31 that thing and have chat GPT give you
1:13:34 strategies for how to work around all
1:13:36 their [ __ ]
1:13:37 so that's why large language models are
1:13:40 a big deal is because they're going to
1:13:41 put this
1:13:42 this translation layer on top of all of
1:13:45 these remarkably complicated systems
1:13:48 that we have to deal with
1:13:50 so that's one of the reasons I'm really
1:13:51 excited about it just don't cover your
1:13:53 head oh that's so lovely hmm
1:14:01 it's like I got a [ __ ] trampoline on
1:14:03 my head tonight
1:14:04 how much does chat GPT pay for computer
1:14:06 cloud storage I don't know but it's a
1:14:08 [ __ ] lot Sam Altman said when when so
1:14:11 so when they launched chat GPT they got
1:14:14 to a million users in five days and
1:14:16 there was an interview with him like
1:14:18 like 10 days after they launched so they
1:14:21 were probably up to like 30 or 50
1:14:22 million people at that point and he said
1:14:25 um they were interviewing him and
1:14:27 someone said like you know how much does
1:14:29 it cost to run these things and his
1:14:31 response was the cost is eye watering
1:14:35 um so you'll notice when when Microsoft
1:14:39 announced their investment in open AI
1:14:42 they said okay we're going to give them
1:14:43 a billion dollars then they launched
1:14:45 chat GPT and you know it got to 100
1:14:47 million people in six weeks I don't
1:14:49 think it was much more than a month or
1:14:51 two after they announced a billion
1:14:54 dollar investment that Microsoft said
1:14:55 okay now it's a 10 billion dollar
1:14:57 investment
1:14:59 um I think most of that 10 billion
1:15:00 dollars is in it's it's not just cloud
1:15:03 storage but it's also so Cloud compute
1:15:06 right so it's it's the gpus so every
1:15:09 time you do a chat GPT inquiry it it
1:15:12 costs a lot of money there's like
1:15:14 massive massive super computers that are
1:15:18 you know processing all this stuff
1:15:21 it's amazing
1:15:27 permanent band will ensue if your video
1:15:29 does not end within five minutes that's
1:15:31 weird maybe I need to go I'll come back