aiLL 8/3/23 - The Democratization of AI is Here

Description

Chapters

Transcript

0:00 foreign
0:22 welcome
0:26 to the AI learning lab
0:30 my name is Kyle Shannon we'll be getting
0:32 started here briefly it's all brand new
0:35 right now tonight what's tonight
0:37 Thursday night
0:39 I got a workshop I'm teaching at 8am
0:42 tomorrow so
0:44 probably can't stay on too late but
0:47 we'll get something going here
0:50 we'll let some people join and jump in
0:52 and say hello hello hello hello hello
0:56 we got Champa here Champ's ready to
0:59 start the live you excited boy you
1:01 excited buddy yeah come here
1:05 yeah
1:07 yeah
1:07 oh that's a good doggy
1:10 good doggy doggy doggy welcome welcome
1:13 welcome
1:15 this is the AI learning lab if you're
1:17 new here new to generative AI I ask no I
1:21 answer I ask a lot of questions too I
1:24 answer as many questions as I can about
1:25 AI if you pop your questions about AI in
1:28 the comments below I'd do my best to
1:30 answer them
1:32 if you're new to generative AI uh and
1:34 you haven't played with chat GPT yet let
1:36 me let me flip this around this
1:41 this big long list used to have only
1:43 three things on it now it's got seven
1:44 things on it because that's how fast
1:46 shit's moving
1:48 um
1:49 let me make it a bit smaller so it fits
1:52 in a little space let me get my big fat
1:55 head out of the way there we go
1:57 so if you're new to to the AI world and
2:02 generative Ai and chat GPT and all that
2:04 happy horse [ __ ] take a screenshot of
2:07 this
2:09 and I'll quickly walk through what these
2:11 things are welcome everybody welcome
2:12 welcome welcome welcome if you have
2:14 questions about AI pop them in the
2:16 comments below
2:18 um my name is Kyle Shannon I'm an
2:20 entrepreneur I'm the founder and CEO of
2:22 a company called storybine which is a an
2:24 automated video platform that actually
2:26 uh we've just recently added AI it's
2:30 it's based more on Aristotle than AI but
2:32 but uh it is automated video editing
2:35 and uh for the past year and a half or
2:38 so I've been deep deep deep diving into
2:40 this generative AI stuff and when chat
2:42 GPT came out
2:44 it hit me that the world has changed
2:46 just not everyone knows it yet
2:49 um so all of these things here what are
2:51 they
2:52 the top one is the official chat GPT
2:54 website if you've never been to it or
2:56 done that go there right now well I
2:58 ramble on about
3:00 how crazy all this [ __ ] is uh go there
3:03 and start playing ask it to write you a
3:05 song or a poem or explain quantum
3:07 physics to you it doesn't really matter
3:08 just go there until you until you have
3:11 what I call your your Kevin McAllister
3:13 moment you're oh
3:15 home alone uh then you got Claude which
3:18 is a major competitor to gpt4 that's the
3:21 official one up here then you got Bing
3:24 which is Microsoft's version of gpt4 uh
3:27 that's connected to the internet that's
3:28 worth at least exploring because it's
3:31 connected to the internet and it can
3:32 generate images and it can read images
3:35 then you have Bard which is Google's
3:37 entry which until about two weeks ago
3:40 was just utter steaming trash and now
3:43 it's actually worth exploring it can
3:45 read uh images and it's connected to the
3:48 internet so it's interesting in that way
3:51 this one the
3:53 labs.perplexity.ai that's metas large
3:56 language models that you can go play
3:57 with for free and then you got pi and
3:59 you got po all worth a uh a check uh
4:03 down in the mud says yay he's back I am
4:06 back I was in uh Tick Tock purgatory
4:17 Tick Tock said I was a very bad boy and
4:20 it kicked me off the platform they said
4:23 we're gonna de-platform you Mr Shannon
4:25 we're gonna de-platform you you can't
4:27 say your words anymore
4:29 for exactly seven days to the minute
4:34 and I appealed
4:36 and they said no
4:38 bad boy
4:43 welcome welcome everybody
4:45 this is the AI learning lab we learn
4:47 about Labs here and the AI stuff I got
4:50 champ by my feet hey buddy
4:53 and uh I just got back at a lovely
4:56 evening a little CEO networking event
4:59 here in Denver
5:01 called the Thunder view CEO something or
5:04 other we hang out hang out a bunch of
5:07 Founders hang out and just [ __ ] about
5:09 how awful it is starting and running
5:11 companies it's it's basically an AAA
5:14 meeting for entrepreneurs that just
5:16 can't stop
5:19 um but I'm back from that I got a couple
5:21 of whiskeys in me so I'm loose I'm loose
5:24 look my hair's looking good the glue
5:26 held today
5:28 I think we're gonna do okay tonight I
5:30 think we're gonna do all right there's
5:32 the bear
5:33 how are you this evening welcome welcome
5:35 welcome welcome
5:37 all right let's let's get some questions
5:39 uh hey glad you're out of tick-tock
5:41 detention I know Lauren me too I'm back
5:43 I'm back I came back last night it was
5:45 my first night back it was a little
5:47 awkward it was like a you know like a
5:49 Junior High dance a little awkward but
5:51 it was good we're okay we got there
5:53 we got there I got a little feisty with
5:55 someone a couple of people
5:58 um what's it telling me to do tick
5:59 tock's telling me to do something swipe
6:01 all right I'm swiping nothing's
6:03 happening
6:04 ah good Lord whatever all right
6:11 hey Kyle how do you generate images with
6:14 AI oh there's good lots of ways here let
6:17 me I will give you some cool stuff
6:24 all right so if you're new to AI the the
6:28 list that I had before all of those
6:30 things were all chat gbt like things
6:32 these things prompts.chat will teach you
6:34 about how to talk to those large
6:36 language models how to talk to chat GPT
6:38 and all the other ones and it basically
6:40 just explains how you talk to them how
6:42 you provide them with context and then
6:44 it also gives you a long long list of
6:46 like personas where you can tell Chachi
6:48 PD to act like an accountant or a
6:50 screenwriter or a therapist or a
6:52 whatever a dog whisperer doesn't matter
6:56 but it'll give you lots and lots of
6:57 ideas the next one down openart.ai
7:00 promptbook is
7:03 um explains to you how you talk to these
7:05 image generation tools that are on this
7:07 page it's specifically designed for
7:10 stable diffusion but it's good for all
7:12 of them because it it
7:14 it's basically a 96 page PowerPoint
7:16 presentation that explains
7:18 all of the [ __ ] that these image
7:20 generation tools know it's it's like a
7:23 little it's like a little art history
7:24 lesson in a box and and it teaches you
7:26 how to talk to these things and get
7:28 really interesting results
7:30 um and then futurepedia.io that's just a
7:32 big long list of AI tools okay
7:35 so for images The Big Three are
7:37 mid-journey Adobe Firefly and stable
7:40 diffusion I have leonardo.ai down there
7:42 but that's built on top of stable
7:44 diffusion it's just that's the friendly
7:46 web front end
7:48 um that this company created and it's
7:51 really really good
7:53 right now mid-journey and Leonardo or
7:56 mid-journey and stable diffusion are
7:59 close to neck and neck stable diffusion
8:01 just launched their new SD XL model
8:04 which is really good but in my opinion
8:06 mid-journey still
8:08 the best and it's crazy good it's crazy
8:11 crazy good it's crazy good I tell you I
8:13 like it the mid journey and that's an
8:15 image I made on the mid journey I last a
8:16 night I did it on the live look at that
8:19 look at that stunning piece of piece of
8:22 Madness
8:24 all right so how do you do it the first
8:27 thing that you need to know about mid
8:28 Journey
8:30 is that not only do you have to learn
8:32 mid Journey you have to learn Discord so
8:34 if you don't learn if you don't know
8:35 Discord well you got to learn that so
8:38 when you go to midjourney.com it'll say
8:40 join the beta and you click on that and
8:42 it'll take you to this thing and you're
8:43 like wait where am I what am I what am I
8:45 supposed to do
8:47 okay so in Discord you have these things
8:49 down the the side those are called
8:50 servers those are basically your
8:52 different communities you'll go into one
8:54 for mid-journey I'm in one uh called the
8:56 AI Salon which is a community that I
8:58 started
8:59 and then but we've got a mid-journey
9:01 channel in here and then
9:03 all of these things here are the
9:04 different channels so the different
9:06 little conversations going on and then
9:08 within mid-journey or within the salon
9:11 in this case in the mid Journey Channel
9:13 this is where you make images so you you
9:16 basically are giving a a command to a
9:19 bot and what you do is you do slash
9:20 imagine
9:23 and that'll pop open this thank you for
9:26 the Roses bear
9:28 it'll pop open this thing called prompt
9:30 appreciate that and you type in whatever
9:33 you wanted to make so you could say
9:36 um a bear
9:38 in a Chicago
9:41 restaurant
9:44 making
9:46 uh beef sandwiches
9:50 anyone else watch the bear
9:51 God damn is that a good show
9:54 all right so now off it goes there's all
9:58 sorts of fancier ways that you can talk
9:59 to Mid Journey but the basic idea is
10:03 you type slash imagine and then you type
10:05 in a prompt and then it's going to go
10:07 off and do its Magic Madness and you'll
10:10 notice the first thing that it generates
10:12 is like this blob and like and it's
10:15 really blurry and you can't see any
10:16 details and then you see a little more
10:18 detail and then you see a little more
10:20 detail that's the way these these
10:22 diffusion models work is they start from
10:25 fully noise images where it kind of
10:28 amalgamates you know thousands of images
10:31 into kind of a random pixel field and
10:35 then denoises them over 50 steps or so
10:38 and they just get more and more
10:41 um
10:41 in Focus basically
10:44 and every image that's generated is
10:47 actually a unique
10:49 original image even though they the
10:52 large language models and the image
10:54 models have been trained on other
10:56 people's work the the piece that's
10:58 created is actually not copy and pasted
11:00 it's actually just original work
11:05 um based on a an insane mathematical
11:08 formula so
11:09 we ask for a bear in a Chicago
11:12 restaurant making beef sandwiches the uh
11:15 I think the the two on the left here are
11:18 really good so so one of the things you
11:20 can do is you can just say okay the one
11:22 on the bottom I like that so that's one
11:24 two three that's three so so you see
11:26 there's u1234 and V one two three four u
11:30 means upscale
11:31 and V means variation so I think three
11:34 is good enough to just upscale but one I
11:37 like but I want to see variations of it
11:39 so I click V
11:41 and so there's my my large version of
11:44 that image that's pretty good nice piece
11:47 of beef there the Bear's cooking it up
11:50 wearing some lovely kitchen garb
11:53 want to make sure that we're food safe
11:55 there's a bear in the kitchen
11:56 [Laughter]
12:00 uh and then the other thing you can do
12:03 now is
12:05 um mid-journey add these things called
12:07 zoom out
12:09 um where you can sort of zoom out and
12:11 it'll add to the image or there's these
12:12 pan things so I can actually pan I can
12:15 say let me let's say I want this to be a
12:18 horizontal image right now it's Square
12:22 um I can hit the zoom button to the
12:23 right or the Pan button to the right and
12:25 it's going to add to the right of this
12:27 image
12:29 and so it's going to create an entirely
12:32 New Image there's the variations for
12:34 number one
12:36 so let's see those we got
12:42 um I think
12:45 I think the second one there is is the
12:47 best personality so we'll do upscale too
12:50 and that's going to give us see there
12:52 there it's adding to the side of that
12:54 image so that's going to do its thing
12:56 and then there's the upscale number two
12:58 here so this one will do zoom out you
13:01 can see what zoom out looks like
13:05 and and this is basically it this is you
13:08 know you can get way more sophisticated
13:09 with options and using different models
13:12 and doing image to image and doing all
13:16 just there's all sorts of crap you can
13:18 do
13:20 um is this done yeah this is done so
13:22 pendrite
13:24 so now we have a landscape image of
13:27 these bears
13:31 with different options in the
13:33 backgrounds I think I like
13:36 this one's kind of got a nice row of
13:38 jars here
13:39 this one's got cool stuff on the table
13:45 let's do the cool stuff on the table one
13:47 so we'll make that one
13:48 which one is that number two we're gonna
13:50 go upscale two
13:55 so there's our upscale too
13:58 all right there it is
14:01 go to that we open it in our browser
14:04 so we can zoom in on that all nice and
14:11 a little detail
14:13 flipping amazing
14:18 and that was a shitty prompt like I
14:20 didn't put it like I just you know I
14:22 just put like bear in a Chicago kitchen
14:24 making a meat sandwich beef sandwich
14:27 and we got that what did that take us
14:29 two minutes two minutes
14:31 two minutes get your [ __ ] together
14:34 it's a George Carlin joke
14:37 how many times do I have to tell you six
14:41 all right
14:43 and then now we've zoomed out on this
14:45 image so we'll see which one of those
14:46 looks good they're all pretty cool I
14:49 like the one with the light here
14:51 these lights look fancy let's let's do
14:54 that one number one we'll upscale number
14:56 one
14:57 and so this is all zoomed out version of
15:00 number one so there you go
15:02 Bang
15:03 so one we sort of pan to the right this
15:06 one we pulled back and it added these
15:08 lights it added all this the the little
15:10 bar stool here all this crap to the side
15:12 here all this stuff is all added
15:15 all new without having to think about it
15:17 just zoom out zoom in pan left pin right
15:20 make a bear in the kitchen here's a bear
15:22 in the kitchen
15:23 [Laughter]
15:26 can I use this crap it makes me [ __ ]
15:29 giddy because it's like
15:30 you know like a year ago none of this
15:34 was possible
15:35 nine months ago Chachi PT didn't exist
15:38 you could use you could play with gpt3
15:40 3.5 or gbt3 whatever it was at the time
15:44 but it wasn't it wasn't sort of put in a
15:47 little package that made it just like
15:49 this magical little tool stick in some
15:51 words out comes a novel
15:53 not quite that good but pretty [ __ ]
15:55 close
15:58 stick on some words out comes an
16:00 illustration of a bear
16:01 or a photograph
16:04 I mean you know what would it what if if
16:07 you've ever tried to do anything in the
16:09 creative realm like to get something
16:11 even remotely close to this concept
16:15 you know it would be a week or two of
16:18 Googling illustrators or photographers
16:21 two or three weeks of interviewing them
16:24 because they always flake out because
16:25 they're creatives I used I used to be
16:28 the chief creative officer of an agency
16:30 so I know of what I speak
16:33 um and then it would have been you know
16:36 two or three weeks worth of
16:38 pre-production and a shoot especially
16:40 for something this complicated or an
16:43 illustration
16:44 and you'd have spent a [ __ ] ton of money
16:48 and now it's like
16:49 Bang
16:50 now the downside of these things is
16:54 to get things really precise and
16:56 predictable to do things like text it's
16:58 really really hard
17:00 um but the state of the art right now is
17:02 pretty remarkable knowing that these
17:04 tools are going to get much much better
17:07 um so that's that one the other things I
17:09 play with if you work in a corporate
17:11 environment you should use Adobe Firefly
17:13 or at least explore it because all of
17:15 their image Generation stuff was was
17:17 trained on
17:18 basically non-copy written images
17:22 um leonardo.ai is just a really really
17:25 good web interface if you don't want to
17:26 deal with Discord which is a pain in the
17:29 balls go to leonardo.ai I think their
17:34 free version gives you like 150 credits
17:36 a day which sounds like a lot but it's
17:38 like I don't know 15 credits per image
17:40 or something like that or eight credits
17:42 an image but you can do a decent amount
17:44 of images per day for free here but it's
17:48 really good you can train the model in
17:50 your own images there's all sorts of
17:52 it's just it's just good it's just good
17:54 so
17:56 anyway that's that uh if any of you are
17:58 kind of trolley like oh he's just making
18:00 money off this no I'm not
18:03 I don't make money off any of these guys
18:05 now
18:06 should they reach out to me and go hey
18:08 can we give you money yeah sure
18:10 I'll talk about your [ __ ] for money
18:14 uh but they're only up here because I
18:16 actually think they're worth you
18:17 spending your time on all right
18:19 so dang that was a good first one we
18:23 dove in day what's your what's your name
18:24 Jelly Bean jelly bean pup Mom I'm gonna
18:27 go nine out of ten on the name solid
18:30 name it's a little bit of a mouthful but
18:32 all wholesome awesome stuffs Jelly Bean
18:36 pup mom there's a dog in there Mom jelly
18:39 beans can't go wrong with any of that
18:41 all right
18:42 why why not
18:45 how to how to make money with AI well
18:48 Gene as we often say on this channel you
18:51 can make money watch out every day
18:56 here's my thoughts on how you make money
18:58 with chat gbt it doesn't matter that
19:00 it's chat GPT
19:02 it doesn't matter that it's ai ai is
19:04 just a tool
19:06 um
19:08 the first step in making money with chat
19:11 is start to get AI literate start to
19:13 play with these tools a lot don't don't
19:16 start with I need to make money as the
19:18 goal just start understanding what they
19:20 are because in learning what they are
19:22 what you'll quickly realize is you might
19:25 be good enough at something to make
19:28 money that you never thought you were
19:29 good enough to make money at before like
19:32 making images or writing words a lot or
19:34 coming up with marketing plans I don't
19:37 know a lot about marketing well this
19:38 [ __ ] does
19:40 so Step One is like get curious get EX
19:43 just start exploring start working on
19:45 personal projects start working on
19:47 projects for other people learn these
19:49 tools then you don't need to learn all
19:51 of them like pick one pick the top one
19:53 that's that's the original that's why
19:55 it's at the top it's why it's bold
19:58 chat openingi.com you can just play with
20:00 the free
20:02 um 3.5 version so if you don't pay
20:04 anything there it's GPT 3.5
20:09 um and start learning about prompting
20:11 and then here's my philosophy on how you
20:14 make money with any tool so this is if
20:17 you want to make money with Photoshop if
20:18 you want to make money with Microsoft
20:20 Word like like
20:23 the the thing about you giving money
20:24 with chat GPT they could be making those
20:26 videos you can make money with Microsoft
20:28 Word how do you do that well you write
20:30 good
20:31 and you use Microsoft Word how do you
20:34 make money with AI you learn how to
20:36 solve a problem better faster or cheaper
20:39 then you could solve it manually
20:42 right so the prerequisite here is just
20:44 start learning what these things make
20:46 possible
20:47 and then do things like go play with
20:49 mid-journey or leonardo.ai or adobe
20:52 Firefly start to understand the image
20:54 generation thing start to get your head
20:56 around
20:57 these tools are actually generating
20:59 original content and if you can get
21:03 astute at writing prompts to get these
21:07 things to give you good results
21:09 you'll be able to make money you'll be
21:11 able to teach courses you'll be able to
21:13 do Services
21:15 you'll be able to build software or
21:18 projects that you never thought were
21:20 possible you'll be able to support local
21:22 businesses you'll be able to do [ __ ] for
21:23 your family
21:25 all of that stuff's possible
21:29 um
21:31 yeah lots and lots of it we had a guy at
21:33 the the salon the AI salon so the other
21:35 thing that I started is a group called
21:37 the AI salon so if you go to the
21:40 salon.ai
21:42 there's a link tree there again this
21:44 isn't for pay all you bitter trolley
21:46 people not making any money off this
21:48 jerks uh if you go to the salon.ai
21:53 um it's a link tree we've we've got to
21:55 meet up there so we meet every other
21:57 week and then we've got an online
21:59 Discord Channel and there was a guy a
22:03 good friend of mine we used to work with
22:04 30 we used to work together 30 years ago
22:07 his kid had a disease and was
22:10 misdiagnosed for a long long time and
22:12 then
22:14 after a while they finally got diagnosed
22:16 and they found out it was not good and
22:18 he took the
22:20 blood work they originally had where he
22:23 was misdiagnosed
22:24 and popped it into chat GPT
22:29 and
22:30 the first thing and he said I want you
22:33 to stack rank what diseases you think
22:36 these symptoms in this blood work might
22:38 indicate
22:39 and the first thing it returned at the
22:42 top of the stack ranked list was the
22:44 disease that his son had
22:47 so like the depth of what these tools
22:50 can do is is kind of beyond what you can
22:53 imagine
22:54 now
22:55 one of the freaky things about them is
22:57 they do what's called hallucinating so
22:59 they lie a lot
23:01 I like to call it mansplaining as a
23:03 service
23:05 and it's just the nature of how the
23:07 things work so over time that'll get
23:09 better but right now just know that they
23:11 might lie but
23:12 they are incredibly like this these are
23:16 the most sophisticated tools
23:19 ever put in the hands of mere mortals
23:21 like in the history of humanity it's
23:24 Insanity what they did is they basically
23:26 trained all these models on all of the
23:29 public internet
23:30 so these things they're like an idiot
23:32 savant where they're really dumb at
23:35 knowing basic [ __ ] but they're [ __ ]
23:39 brilliant at knowing all the stuff and
23:41 so if you can guide them well
23:43 and say I want you to act like this kind
23:46 of mathematician that specializes in
23:48 this branch of mathematics and I want
23:50 you to help me think about how I would
23:52 solve this problem it'll be [ __ ]
23:54 brilliant
23:56 um and it'll just know it you want to
23:59 know about six different kinds of you
24:01 know psychology and how the different
24:04 theories compare tell it to make a table
24:06 comparing six different theories it will
24:09 all right what did they say you did they
24:12 said it was hate speech so so I I we we
24:16 listen to or we were doing a joke about
24:20 I like turtles which is this funny meme
24:22 from like the early 2000s I think where
24:25 this kid at a carnival was interviewed
24:27 and
24:28 he's like in zombie makeup at Halloween
24:30 and the newscaster goes you know you
24:34 know how's your day going he goes I like
24:36 turtles just random so then I showed a
24:39 clip of another comedian that said
24:41 something similarly funny but it had a
24:45 not a bad word but a a a uh a word
24:48 that's on the list of words you can't
24:50 say anymore and so I played the clip and
24:53 then I sort of just repeated what the
24:55 comedian said in kind of a joking way
24:58 and then the scream went boom
25:06 whatever
25:07 I'm back I'm back baby
25:10 all right where how can I train llm on
25:13 my own data okay
25:15 so
25:16 all right now the hair's starting to
25:18 lose it's lose its mind which is that's
25:20 what it does this time of night this is
25:22 just this is one of the things we deal
25:23 with at least I got [ __ ] hair right
25:25 okay
25:27 um
25:28 so you have a couple of options so
25:29 what's your name Crum uh as in our crumb
25:32 that's cool name like the name uh 8.5
25:34 out of 10.
25:37 um a couple of options you have one of
25:39 them is claude.ai well you got two
25:42 options here oh so start with Claude
25:44 claude.ai is from a company called
25:47 anthropic it's a bunch of X openai
25:50 people
25:51 and claude.ai is free
25:54 and it's got a hundred thousand token
25:56 context window which basically means you
25:59 can either upload I think you can upload
26:01 up to five documents or you can just
26:03 paste into your chat up to 75 000 words
26:06 and start interacting with it it's got a
26:08 75 000 word memory basically
26:11 so so if you just want to train it on
26:16 like a document you've got or you know
26:18 some sort of thing like that just go to
26:20 Claude paste it in there or upload it
26:22 and start interacting with it it's
26:23 [ __ ] remarkable you have a two hour
26:25 conference call and and you turned on
26:28 transcriptions on on Google meet or
26:30 whatever it is download that
26:31 transcription copy it paste it into
26:34 Claude and tell it to summarize your
26:36 meeting and give you all the you know to
26:39 do's and who was in the meeting and who
26:41 said what and who didn't speak so you
26:43 can passive aggressively humiliate them
26:45 on email it'll do that for you
26:48 so one option is you just kind of do an
26:50 ad hoc in something like Claude
26:53 chat EPT if you pay for the plus thing
26:55 has this tool called code interpreter
26:58 which
26:59 code interpreter you can upload a single
27:02 100 megabyte file but here's a trick
27:05 that file can be a zip file
27:08 so as long as it's less than 100
27:10 megabytes you could upload a zip file
27:13 full of CSV files or full of PDFs or
27:16 full of whatever
27:18 and you can start interacting with code
27:20 interpreter and what code interpreter
27:22 can do is it actually it can actually
27:24 analyze that data so you can upload 10
27:27 csvs and you could say hey code
27:29 interpreter what's interesting about
27:31 this data and it will start writing and
27:33 executing Python scripts to analyze the
27:35 data and tell you what's interesting
27:37 about it and then you can say
27:40 um why don't you identify the most
27:42 significant Trends in this data and
27:44 visualize them and then it will and then
27:47 it will write code to generate charts
27:50 and graphs to visualize that data
27:52 [ __ ] astounding so the those two
27:55 tools right there code interpreter at
27:57 openai and Claude dot AI
28:02 um are just staggering okay so so that's
28:04 that
28:06 now if you're geeky if you're good and
28:08 geeky and you don't mind getting your
28:09 feet dirty with little Python
28:11 Programming and maybe spinning up a
28:13 little
28:14 Google collab notebook then I would
28:17 explore a project called Lang chain as
28:20 in language chain
28:21 and Lang chain lets you embed
28:25 your own data your own documents so you
28:28 can take a directory full of PDFs and
28:31 embed them
28:32 and then you basically store what are
28:34 called vectors in a vector database
28:36 vectors are basically how this [ __ ]
28:38 works so they basically turn words into
28:43 plots in mathematical
28:46 thousand Dimension space basically it's
28:49 it's like incredibly complicated but it
28:51 doesn't matter it just basically reads
28:53 all your documents learns all the words
28:54 in them and how they're semantically
28:56 related clusters them and makes them
28:58 searchable that's what it does
29:01 so once you've got Lang chain once
29:03 you've got all your documents embedded
29:04 you basically have your company GPT and
29:08 you can do whatever the [ __ ] you want
29:09 with it and you can use all sorts of
29:10 different models and it's crazy so uh if
29:13 you're geeky go check out langtay if
29:16 you're pseudo geeky like I am where like
29:18 I'm just geeky enough to get [ __ ] sort
29:20 of working and then when it breaks I get
29:21 frustrated and want to smack things
29:24 um
29:25 it might be a little frustrating if
29:27 you're just a tad more geeky where you
29:29 don't mind sitting with code for four
29:30 hours and that makes you put you in your
29:32 happy place then you'll love Lang Ching
29:34 it's not that hard it just requires a
29:37 little bit of tech chops
29:39 and then if you're not geeky then just
29:41 go use those two tools that I talked
29:43 about so hope that helps all right
29:45 that's holy crap there's 150 people in
29:48 here hello everybody my name is Kyle
29:49 Shannon it is a Thursday night it's 9 42
29:53 I'm here in Denver Colorado yes I'm live
29:55 if you're thinking like but he hasn't
29:57 answered my question yet why isn't he
29:59 answered my question why is he just
30:01 talking to crumb this whole time who's
30:03 Chrome I don't see crumbs little
30:05 speeches little typed in words
30:09 cuz
30:11 I answer all the questions and Crum
30:13 asked earlier than you did and so Crum
30:16 got the ants crumb got the crumbs
30:18 crumb got the crumbs
30:20 that's how the cookie crumbles
30:23 at the learning lab you can ask people
30:25 that have been here before
30:28 I got a lot of integrity
30:30 I definitely I like turtles
30:32 all right
30:38 what'd you do last night I don't want to
30:40 talk about it
30:42 all right oh Tick Tock just jumped me to
30:44 the bottom
30:46 bug somebody fixed that [ __ ] bug
30:51 if anybody here works for tick tock put
30:54 it in the [ __ ] instead of kicking me
30:57 off for seven days for repeating a word
30:59 a comedian said that apparently it was
31:00 fine for him to say but not for me to
31:02 repeat
31:03 fix this [ __ ] bug
31:08 if you're wondering why I'm so wound up
31:10 I've had a few drinks
31:15 um
31:17 all right
31:19 I am John screaming in the wilderness
31:21 repent and turn away all right
31:24 good day good day good day
31:30 uh need some AI for my real estate
31:32 business well I would start with uh what
31:36 would be fun to explore is go to
31:41 oh I don't know why that's a new story
31:43 all right a new story just popped up
31:44 that was stupid okay there's a surprise
31:48 um
31:50 if you've got access to chat GPT plugins
31:53 so if you pay for chat GPT Plus
31:56 go searching the plugins there's 800
31:58 plugins now there's a bunch of real
32:00 estate plugins in there I don't know if
32:01 any of them are any good because I'm not
32:03 in that world I haven't I haven't played
32:04 in that world in a while but
32:07 you can if if you've got a plug-in that
32:10 works like I'm pretty sure there's a
32:12 Zillow plug-in and there's local ones in
32:14 Canada and [ __ ] like that
32:16 um you can start chat gpting with local
32:19 real estate listings so you can do
32:20 things like you know find me houses in
32:23 this range in this neighborhood
32:26 um now describe that neighborhood now
32:29 um write up marketing blurbs for each of
32:32 those properties that I can put in an
32:35 email for this kind of client out it'll
32:37 come so you could just start in an ad
32:39 hoc way just [ __ ] around with plugins
32:42 at chat GPT and and just see see what it
32:45 makes possible I'm happy to show you how
32:47 they work I mean basically you just go
32:49 find a real estate plug-in enable it and
32:53 then start talking and say show me
32:55 listings and it will
32:58 um
32:59 so you can do that you can I these
33:02 here's the thing to do rather than
33:04 saying I need AI for my business I would
33:07 flip it and do this
33:09 what problems do you need to solve in
33:11 your business start with the problem
33:14 and then and then start to say well uh
33:19 okay I need to learn how to do social
33:21 media okay great chat CPT can teach you
33:23 how to do that it can write all the copy
33:24 it can teach you how to do it it can
33:26 give you a whole strategy if you're like
33:29 well I need
33:30 better images of these properties or I
33:33 want images of that town well maybe then
33:36 you play with mid-journey and start
33:37 generating images that are kind of sexy
33:39 images you know if you need to help
33:42 writing letters or analyzing lease
33:45 documents
33:46 you know all this stuff's really good so
33:48 start with the problem figure out the
33:50 problem is and then start to investigate
33:52 which tool do I need
33:54 to solve that and it's not always going
33:56 to be AI
33:59 sometimes the right tool is like
34:02 Salesforce or probably not Salesforce
34:05 the uh real estate equivalent of
34:07 Salesforce Salesforce is a [ __ ] train
34:09 wreck
34:10 but
34:11 what Salesforce needs is AI to put a
34:14 [ __ ] Simplicity layer on top of that
34:16 complicated nightmare [ __ ] show of a
34:18 piece of software
34:20 ask me what I've been using for the last
34:22 decade I [ __ ] hate that thing
34:30 hey you able to access beta plugins for
34:33 chat GPT I just purchased the monthly
34:36 thing yes so here's what you need to do
34:40 this is non-obvious and really [ __ ]
34:43 frustrating and why they buried it like
34:44 this is beyond me but let's get you up
34:47 to speed
34:49 Okay so
34:51 if you pay for chat GPT plus it's 20
34:54 bucks a month then you get access to
34:56 plugins and code interpreter that I just
34:59 do I hit print instead of
35:02 Plus
35:04 oh they've added some new things here
35:06 actual copy
35:09 um if you go to gpt4 you should see
35:12 default code interpreter and plugins if
35:15 you don't see these that's because they
35:18 buried that [ __ ] in settings because
35:20 they're dum-dums I don't know why they
35:23 did it and that's beta we don't want
35:25 everyone using it okay so you go to your
35:27 settings and then in your settings
35:29 there's an option called beta features
35:31 so you click on beta features you turn
35:33 them all on
35:35 why they would not just default this to
35:37 being on is flipping Beyond me but
35:40 whatever there's probably some lawyer
35:42 involved well we've got less liability
35:45 if you don't default them to Beyond and
35:48 make them explicitly log into our I
35:51 don't know why it's not focusing
35:53 this is another thing Apple needs to fix
35:55 their
35:56 their focus they need to fix that ah
36:01 normally the little hand trick works
36:03 well well whatever
36:05 and then I'm gonna show you these
36:09 there we go okay that's good enough
36:11 all right and then the way it works is
36:13 you go to plugins you click on plugins
36:16 let's go find the real estate one we
36:17 were just talking about the real estate
36:18 one we'll turn off web pilot
36:21 and we will go to the bottom here we'll
36:24 go to the plugin store
36:25 we'll go real estate
36:31 there's only two
36:33 real estate agents real estate data
36:36 and location-based tools maybe zillow's
36:38 the only one let me go back to
36:41 huh
36:43 did I spell that wrong no
36:46 let me close it do I have Zillow in here
36:49 I think I do
36:50 I think I have a Zillow
36:53 hmm
36:59 maybe there's not
37:00 real estate plugins
37:03 real
37:06 real time photorealistic
37:10 date
37:16 enriched real estate data and
37:18 location-based tools
37:20 huh
37:21 strange that there's not real estate
37:24 I thought there would be I thought there
37:25 was well this is this this is the state
37:28 of the the industry right now so we've
37:30 got we've got this thing so let me see
37:33 um find me
37:35 recent
37:37 listings in 80211
37:42 let's see now it should fire off that
37:44 plug-in so once you have yeah so fired
37:46 off the plug-in so once you've got these
37:48 plugins installed you basically enable
37:50 them for whatever chat session you want
37:52 to do it in then you ask your questions
37:53 and it goes off and does [ __ ]
37:56 it seems I need more specific
37:57 informations could you provide more
37:59 specific details oh such as a full
38:01 address oh I see what it's doing
38:03 um so it's like
38:05 um
38:06 let's see what is the average home price
38:13 in uh let's say
38:16 Aurora Colorado
38:20 see if it if it can do any of that
38:24 it might not even be able to do that
38:29 no it doesn't like it I don't know
38:32 what's going on with real estate I know
38:33 I've looked up real estate before I
38:35 wonder if they I wonder if they got rid
38:37 of something let me try another thing
38:42 we'll go to plugins we're going to turn
38:44 off this piece of [ __ ] that didn't work
38:46 Top Hat didn't work for me but if you're
38:48 into real estate you might know how to
38:49 use it better than I do I'm going to go
38:50 to web pilot and so web pilot will go
38:53 search uh list search websites let me
38:56 see if it'll search Zillow
38:59 um
39:00 use Zillow to find
39:06 recent listings
39:10 in 80211 let's see if it see if it does
39:15 it it may not because you may have to
39:17 log into that site
39:19 using web pilot so it's going off and
39:22 doing its thing it might come back and
39:24 say I apologize
39:26 I apologize this is a this is We're Not
39:29 Gonna help you in that we're not gonna
39:31 do it dude
39:37 um
39:39 look how cool it is he's surfing on the
39:42 internet using AI it's super exciting
39:45 I'm sorry but I was able to directly
39:48 directly access Zillow however I did
39:52 find several several other Real Estate
39:54 websites that may have information for
39:57 you
39:59 Ah that's good that's that's doing it it
40:01 found realtor.com Redfin Trulia Remax
40:05 there's Re Max 80211 Denver Colorado so
40:09 now it'll pop me out
40:12 yeah there you go there's the hood
40:15 there you have it
40:17 so
40:20 with stuff where you want to get really
40:22 specific like this you just kind of have
40:23 to experiment and dick around a little
40:25 bit like like I just did and and that's
40:29 hang on
40:31 um
40:35 part of the joy of playing with
40:38 technologies that are this new is that
40:41 you get to play with stuff that is truly
40:44 [ __ ] remarkable
40:46 the downside of it is this is all
40:48 spitting duct tape right now everything
40:50 like plug-ins are just a [ __ ] train
40:52 wreck code interpreter is brilliant and
40:55 mostly works but sometimes it just
40:56 doesn't
40:58 um
40:59 you know sometimes these things just say
41:01 yeah I can't do that I'm not gonna do
41:03 that go [ __ ] yourself
41:05 and it's just because everything right
41:07 now is new and raw right so so the re
41:11 this channel will not be necessary a
41:15 year and a half from now maybe two years
41:17 from now
41:18 that's when I'll get into twerking right
41:20 and I think that's what everyone's
41:21 waiting for quite frankly no one really
41:22 cares about AI they're like when's he
41:24 gonna twerk I mean with hair like that
41:26 why wouldn't he be twerking
41:29 isn't that what people want to see I
41:31 think it is
41:34 twerk while I'm npcing
41:37 I'll be [ __ ] loaded I'll be rich I'll
41:39 be rich I'll be famous
41:41 wait aren't you the guy that twerks and
41:43 does that stupid NPC [ __ ] yeah that's me
41:46 you want a signature no
41:49 until then we'll talk about this AI [ __ ]
41:52 because it's also new and raw and
41:53 changing all the time like I said that
41:55 list right there used to be three sites
41:57 now it's seven that was like two weeks
41:59 ago that was three now it's seven
42:02 and probably within about a month
42:04 there's going to be two more added to
42:06 that that I know of apple is going to be
42:09 on here probably around September is my
42:11 guess right isn't that when they do
42:13 their fall thing they do their developer
42:15 conference conference in June and their
42:17 Hardware release in September I would
42:20 think that there's going to be some AI
42:22 announcements from Apple in September
42:27 so right now there's a lot to talk about
42:30 because there's a lot to explore so just
42:31 know that
42:33 um this shit's a little janky and it's
42:35 not going to be perfect and don't
42:37 dismiss it because of that
42:40 understand that these things are the
42:42 worst they're ever going to be
42:44 and if you watch that thing write you a
42:46 [ __ ] marketing plan or write you a
42:48 Country Song complete with chords even
42:50 if it sucks
42:53 and and do it like that
42:55 if you've ever had writer's block
42:58 which if you're a writer is about six
43:00 times a day that's gone for me it's gone
43:05 if I'm feeling insecure about my writing
43:08 skills I'm like I don't really want
43:09 anything I mean oh you know what I could
43:11 do let me just stick a little shitty
43:13 idea and chat GPT
43:15 oh yeah that is a pretty good idea
43:17 there's more
43:19 there's more
43:20 writer's block is gone
43:24 shit's remarkable like [ __ ] crack
43:30 all right I used mid-journey to create a
43:32 Snowman on a beach for a snowbird dance
43:36 flyer for a client he loved it that's it
43:39 and and what's what's good about that
43:41 Lauren like why what I loved about what
43:43 you wrote there is it sounds like what
43:45 you did was you know just a Snowman on a
43:47 beach that sounds like a hilarious kind
43:49 of parlor trick you'd do in mid-journey
43:51 oh let me see a slow man on a beach but
43:53 you did it for a client right he had a
43:54 snowbird dance and you're like ah here's
43:56 an idea I got a problem to solve I need
43:59 an image that represents that let me pop
44:01 into mid-journey do that show it to the
44:04 client maybe you'll like it oh I love it
44:06 how'd you do that
44:08 how'd you do that so fast
44:10 that's the magic and that's that's that
44:12 trick about don't lead with the AI lead
44:14 with the problem
44:16 and then figure out which tool do I need
44:18 to use to solve that problem and what
44:20 you'll find is that these things can
44:21 solve problems
44:23 way better way better than you think
44:26 they can
44:28 and then sometimes it'll suck
44:32 and then you tell it it sucks and it's
44:33 like oh I'm sorry here's another attempt
44:36 and you're like no no that sucks too oh
44:38 I'm terribly sorry about that
44:40 let me try again now
44:43 and sometimes you do that 10 times and
44:45 it just doesn't get better you're like
44:46 [ __ ] it
44:47 you just go write it yourself
44:50 but I would say 90 of the time it gives
44:54 you something that is 90 of what you
44:56 need eighty percent of what you need as
44:58 you get better at prompting you you can
45:00 you can sort of focus these things
45:02 better and better I can consistently get
45:04 to 80 quality level
45:07 within 10 minutes
45:08 and then I spend the next hour or two
45:11 that I would have spent getting to 80
45:13 refining it
45:20 thank you whiskey Kyle this is awesome
45:22 my first time seeing you live fun and
45:24 informative
45:26 the AI learning lab it's fun and
45:28 informative you can make money
45:32 Jelly Bean pup mom love the name love
45:36 the name
45:37 all right let's see this is amazing
45:40 thank you you're welcome I appreciate
45:42 that where the hoes at that's that's
45:45 just disrespectful there the middle man
45:48 funny but disrespectful all right the
45:51 detail is crazy good oh on the on the on
45:54 the bear in the kitchen making the beef
45:56 sandwich it was kind of a big long log
45:58 of beef I guess it was a filet mignon he
46:01 was making beef out of it's like a wagyu
46:03 filet
46:06 it was pretty funny
46:08 which AI is this uh the one I was using
46:11 with the images was uh mid-journey
46:15 Carlin was King Carlin was King
46:18 he went through an unfunny period but
46:20 toward the end he found it again
46:22 he was [ __ ] brilliant in the 70s holy
46:26 [ __ ] what software free or paid the
46:30 mid-journey is paid you can do like 15
46:32 or 20 images although I think they've
46:34 cut that out because every it's so busy
46:36 all the time I think uh you have to pay
46:38 for it it's 10 bucks a month for
46:40 mid-journey
46:41 leonardo.ai which is built on top of
46:43 stable diffusion has a free
46:46 you get like 150 credits a day that
46:48 recycles every day and it that equals
46:51 like maybe 15 images something like that
46:53 or slightly less or slightly more than
46:56 that
46:57 um
46:58 if you've got Adobe Creative Cloud you
47:01 can use a generative fill within
47:04 Photoshop you can use Adobe Firefly
47:08 and if you're geeky
47:11 um just go dick around with stable
47:13 diffusion it's all free it's all open
47:15 source you can do anything you want with
47:16 the developer Community for stable
47:18 diffusion is insane the tools are insane
47:21 but you gotta be a little bit of a yeah
47:26 I like the code
47:30 because if not it can just be really
47:32 frustrating and listen it's not that
47:34 hard right if you can you you like go to
47:37 chat GPT and ask it what Google collab
47:39 is and ask it to explain to you you know
47:42 what it's about and then go get yourself
47:44 a Google collab account and then go to
47:47 Reddit and see all the crap they're
47:49 talking about in the stable diffusion
47:51 subreddit and they'll say here's a
47:52 Google collab notebook for this tool
47:54 automatic 11 11 whatever it might be and
47:57 then you can go play with them
47:59 um so there's lots of free options out
48:00 there but you at this stage of the
48:02 industry you got to be a WeChat geeky
48:05 the other thing you can do if you want
48:06 to generate images for free go to
48:08 bing.com
48:10 thank you the bear that's my favorite
48:12 the dangler lights I like the dangly
48:14 lights they look like a pretty little
48:15 necklace it's like a pretty little
48:16 necklace I like looking pretty you know
48:19 put a little necklace on pretty little
48:21 necklace
48:22 um
48:24 foreign
48:25 idiot I don't know he seems smart but
48:29 then every now and then he's just like
48:30 he's just like a dum-dum
48:32 he talks about the pretty little dangly
48:34 necklace I don't know why he does that
48:35 it seems unnecessary
48:37 anyway bing.com
48:40 is if you go there and click on the chat
48:42 button
48:43 well let's go do it oh wow
48:47 that's barred that's barred not Bing
48:50 go to bing.com
48:53 so you go to bing.com it's just like
48:55 Google used to be except no one used it
48:57 but if you click on chat
49:00 well now this is gpt4 so this is gpt4
49:03 which at open AI costs you 20 bucks a
49:05 month and here's free and it's connected
49:08 to the internet and you can dial in sort
49:10 of these different modes creative mode
49:12 precise mode balanced mode and you can
49:15 go
49:16 um
49:18 make me an ill
49:23 illustration let me just do it with the
49:26 microphone so I don't have to type weird
49:30 [Music]
49:32 illustration of
49:36 a vacuum cleaner
49:39 causing a tornado
49:44 and so you just tell Bing to make you a
49:48 pretty picture and then it shows you
49:50 this stupid icon of a computer making
49:53 a picture that's such a Microsoft thing
49:56 to do like they take up the entire
49:58 screen to tell you they're making an
49:59 image but there you go there's a vacuum
50:01 cleaner making tornadoes now this isn't
50:05 as good as mid Journey it's not as good
50:06 as stable diffusion it's not as good as
50:08 Adobe Firefly but it's free and from
50:11 what I heard the so this is based on
50:14 Dolly which is open ai's image generator
50:17 from what I've heard the new
50:21 version of Dolly I don't know if they're
50:23 going to call it Dali but the new open
50:24 AI image generation tool apparently does
50:27 fully coherent text like you could say
50:30 make a billboard that says hot dog and
50:32 it'll say hot dog and it's apparently
50:35 [ __ ] brilliant so I don't think this
50:38 is gonna suck for very long but it's
50:40 free so if you just want to do a quick
50:41 little image for a comp for a little
50:43 pitch deck you're doing or or just a
50:46 someone's birthday card or you you want
50:48 to do a kids book and you want to make a
50:51 thing of them being uh so let's let's do
50:54 that um
50:56 make me a children's book illustration
51:00 of a vacuum cleaner sucking up a tornado
51:04 to save the farm
51:12 so let's say you had an idea I would
51:15 like to make Jimmy a book and then
51:17 you're like hey I wonder if chat GPT
51:19 could help me write a book oh it can
51:21 that's a swell little book I wonder if
51:23 Chachi BT could help me make pretty
51:25 pictures of the book oh yeah there they
51:27 are
51:28 there you go
51:33 it is
51:35 I I don't know like I've been
51:38 I've been fairly deep in this world for
51:41 more than a year now
51:42 year and a half maybe two
51:45 and every time it does this it blows my
51:47 mind now are those illustrations good or
51:49 usable not really they're pieces of crap
51:51 but
51:53 it's at least in the neighborhood of
51:54 what I was talking about and if I took
51:55 some time to actually work on that
51:57 prompt
51:59 are you kidding me
52:02 you know
52:07 I mean
52:09 it's nuts it's just nuts it's not it's
52:12 not that that's that's so I hope that
52:15 helped you Royal what is that obot user
52:18 or gbot user cubot user Royal Cube bot
52:21 user
52:24 um
52:25 and the other thing I would do Royal
52:26 Cube bot user is go to this document
52:29 right here wait hang on gotta make the
52:32 letters smaller the letters are too big
52:34 I can't see them Kyle they're right
52:36 behind your big fat head
52:38 all right screenshot it now wait hey it
52:41 there you go screenshot it now this one
52:45 openart.ai prompt book if you want to
52:47 learn how to use these image generation
52:49 tools this is a godsend
52:52 you can go to sites like Lexica lexica.a
52:56 I think it's lexicon.art
52:59 lexica.art you can just scroll through a
53:02 bunch of images and then if you click on
53:04 the image you can see the prompt that it
53:06 was made with and then you can pop that
53:08 in an image generation tool and make
53:10 pretty pictures but if you want to
53:12 understand how these things work and
53:14 what they actually make possible this
53:15 prompt book is just it's great education
53:18 and it's really inspiring it'll give you
53:20 like that that prompt book for me was
53:23 like three weeks of just Rabbit Hole of
53:26 like exploring all different ways you
53:28 can make images it's it's it's
53:30 miraculous it's like it's like it's like
53:33 an art history degree in a in a box
53:37 and then you know instead of the art
53:40 history being the static thing you
53:42 learned about old people that made these
53:44 you know crappy old paintings
53:47 now you can go make your own artwork in
53:49 their style and just your mind will go
53:52 squid
53:54 and then there'll be a little drip it'll
53:56 go squid and drip drip drip you're like
53:58 why is my brains dripping down my face
54:01 you must be using AI
54:07 just use gigapixel to upscale it if
54:09 needed there's also what's that other
54:11 one topaz labs
54:13 what's that one called gigapixel oh
54:16 that's what you said gigapixel from
54:18 topaz Labs yeah gigapixel is good you
54:20 can also upscale and stable diffusion
54:22 but again you got to be a geek
54:25 do you have an AI contract
54:28 I don't know what that means
54:32 like am I getting paid by any of these
54:33 [ __ ] no
54:36 why do you work for one of them you want
54:38 to give me an AI contract
54:40 I'm doing a workshop for a big company
54:42 tomorrow
54:44 um
54:45 I'm doing another workshop for company
54:48 at the end of the month but I don't I
54:50 don't know what you mean by what I don't
54:51 know what an AI contract is
54:53 if you mean am I getting paid by one of
54:55 these guys no
54:57 I just these are there is more of a
54:58 public service announcement so that
55:02 people don't come in here and go well
55:03 where do I start you start here start at
55:06 the top one
55:07 why are there seven can't you just pick
55:09 one no because they all do something
55:12 slightly different enough that they're
55:14 all worth [ __ ] around with
55:15 but if you just are overwhelmed and
55:17 you're like I don't even know what to do
55:18 at all go to the top one
55:20 and just spend the next three weeks
55:22 there
55:24 seriously
55:25 wait
55:28 oh
55:36 not joking
55:38 not being hyperbolic
55:41 chat GPT is so powerful
55:45 I feel like I could easily spend the
55:47 next decade if I did nothing but
55:50 exploring what that thing made possible
55:53 the 3.5 version not even four not even a
55:56 fancy new shiny version
55:58 the dusty old tired 3.5 that's like
56:01 wicked fast
56:02 I feel like I could spend the next
56:04 decade
56:05 trying to understand what that even
56:07 makes possible
56:10 so even though these tools are not
56:12 perfect and even though you feel like
56:14 you're way behind you're not just go
56:16 start playing
56:19 ask it to make a picture of dogs playing
56:21 Billiards oh we can do that
56:23 but that's
56:24 right it's actually going to be curious
56:26 to see if it how close it gets to
56:30 um
56:33 make a classic painting of dogs playing
56:38 billiards in a smoky Billiards Hall
56:43 let's see what it does we'll see we'll
56:45 see how close it gets to dogs playing
56:46 poker I assume it's gonna
56:54 I assume it's gonna pick up that style
56:56 yeah I picked up that style
57:01 oops
57:02 [Laughter]
57:05 he's got to pull key on his collar
57:15 they're all [ __ ] climbing all over
57:17 the table it's totally on dogs who would
57:19 play pool
57:20 [Laughter]
57:24 never a dull moment at the AI learning
57:28 lab yeah
57:33 oh good lordy lordy lordy okay
57:37 where how do I train my llm on my own
57:41 data Decades of legal writing crumb you
57:44 need Lang chain Lang chain or wait for
57:49 three months is my guess
57:52 Microsoft I assume if you're a lawyer
57:55 you probably use the Microsoft Suite of
57:57 products Microsoft 365 and all
58:01 um they're gonna roll gpt4 into all of
58:04 those products I think it's going to
58:05 cost 30 bucks per user per month uh for
58:08 the for the AI tools to be rolled in
58:11 um
58:12 and then Google is going to do the same
58:14 with bard
58:16 both of them have cloud storage
58:19 tools right Google Drive
58:23 I don't I don't remember what Microsoft
58:25 is called but I don't really care
58:28 um
58:31 they're gonna those tools are gonna let
58:34 you train on your own data I think
58:35 Apple's gonna come out with something
58:37 that'll let you train on your own data
58:38 if you want to do it right now and you
58:40 want to be in complete control of it
58:43 if you if you're not geeky enough to do
58:45 Lang chain get someone who is
58:48 um and put Lang chain together because
58:50 basically that it's what the Decades of
58:53 legal writing you basically just point
58:56 Lang chain at directories have it embed
58:59 it it'll cost you some amount of money
59:01 for the embedding that's it's a decently
59:03 expensive process but I think they just
59:04 cut the price of that like a tenth of
59:08 what it used to be it's it's cheaper
59:09 than a lot cheaper than it used to be
59:11 and then it will embed all of your legal
59:13 writing into a vector database and you
59:16 can basically have your legal writing
59:18 GPT
59:19 so yeah go do that Lang chain
59:23 get thee to a lang chain
59:25 henceforth
59:27 this Fortnight yon Sparrow
59:32 see and they said an acting degree would
59:35 be worthless
59:40 cards [ __ ] cards
59:45 do you have an AI course
59:47 um I thought about one I have wait where
59:49 is it oh it's it's over
59:51 I have a Graphic for an AI course
59:54 see
59:57 Hands-On Workshop
1:00:00 topic learning Chachi BT
1:00:04 um I actually just put together this
1:00:05 workshop for a corporate client so
1:00:08 I might spin that off the reason I've
1:00:10 been hesitant is um
1:00:13 I mean this channel isn't really about
1:00:15 doing that
1:00:16 I don't mind just teaching it but
1:00:19 I you know this channel is obviously a
1:00:22 little random and silly sometimes so I
1:00:24 think it's probably not a bad idea to
1:00:25 put something together so I think I'll
1:00:27 do that I mean I've got the graphic for
1:00:29 it now so the first step to anything is
1:00:32 start with the marketing
1:00:35 and and if people are like I'll buy it
1:00:37 you're like okay
1:00:38 and I'm being facetious there that is
1:00:41 not the first step to anything but if
1:00:43 you got the Photoshop skills you might
1:00:45 as well do it
1:00:48 uh
1:00:51 I don't know I was on there for two
1:00:53 hours and all I did was try to sell us
1:00:54 this Workshop he just tried to sell us
1:00:56 this Workshop over and over and over and
1:00:58 then we went to buy it and it doesn't
1:00:59 even exist
1:01:01 I don't understand how he's gonna make
1:01:03 money with chat GPT
1:01:05 he doesn't have a workshop
1:01:08 I don't know what he's thinking
1:01:09 and he certainly needs some hair product
1:01:14 all right
1:01:17 how to train your own AI model
1:01:20 how do you do it how do you train your
1:01:23 own air model well you start with the
1:01:25 Lang chain
1:01:26 why do they call it Lang chain well
1:01:28 these are large language models Lang
1:01:30 chains short for language chain chaining
1:01:34 language together
1:01:37 you see it's rather obvious isn't it
1:01:41 get the to Lang chain
1:01:44 I've begun the digital marketing and my
1:01:46 mind is going wild thinking how I can
1:01:48 use this yeah the Okay so
1:01:51 it sounds like jelly bean pup mom you're
1:01:54 having
1:01:55 you're at least having the beginnings of
1:01:57 your Kevin McAllister moment the I
1:02:00 didn't know it could do that
1:02:02 take whatever you're learning in digital
1:02:04 marketing and go it doesn't pick any one
1:02:06 of these tools just go to Bing Bing's
1:02:08 free
1:02:09 or go to chatgpt 3.5
1:02:11 um the the non-paid version and just
1:02:13 start
1:02:14 whatever you just learned if they just
1:02:16 taught you how to do marketing plans or
1:02:18 how to think about social media strategy
1:02:20 or whatever it is take whatever you
1:02:21 learn and just go in there tell it tell
1:02:24 Chachi BT to act like a a marketing
1:02:27 manager or marketing strategist and then
1:02:29 start asking it questions and start
1:02:31 having it teach you what you just
1:02:32 learned
1:02:33 in your mind like I'm telling you it's
1:02:36 quit
1:02:37 brain juice drip down your face
1:02:41 mama why are you crying oh it's it's not
1:02:43 tears it's just brain fluid it's it's
1:02:45 spinal fluid why is it coming out of
1:02:47 your ears well that happens that happens
1:02:50 darling yes darling I think it's time
1:02:52 for bed now
1:02:54 mama needs to make another marketing
1:02:55 strategy oh Lord what do I do
1:03:02 uh I use chat GPT from a real estate
1:03:05 business how can the other apps help me
1:03:08 um
1:03:10 I don't know they may not
1:03:13 if you want to do if you wanted to do
1:03:15 some sort of clever marketing campaign
1:03:17 where like I don't know let's say you
1:03:19 live in Toronto and you want to you want
1:03:21 to make references to local Toronto
1:03:26 landmarks and you want to put something
1:03:29 humorous like a funny donkey walking
1:03:32 down the street with whatever Toronto
1:03:35 landmark in the background like you I
1:03:36 mean you could get creative with some of
1:03:38 the image stuff I don't know that you
1:03:39 need to go there
1:03:41 um
1:03:45 I mean you could you could use things
1:03:46 like generative fill within Photoshop
1:03:49 where maybe you've got a vertical image
1:03:51 of like a house and you want to sort of
1:03:54 extend the width you could I I wouldn't
1:03:58 I wouldn't worry too hard about
1:04:01 what tools
1:04:04 what am I missing
1:04:06 like like I said the just using chat gbt
1:04:09 for Real Estate it's so deep the amount
1:04:11 of things that you could do just with
1:04:13 these tools especially if you start
1:04:14 playing with code interpreter with chat
1:04:17 GPT
1:04:18 Ruth it's it's
1:04:20 um
1:04:22 you can dig very very deep
1:04:25 um with with code interpreter and
1:04:26 Analysis and things like that I I
1:04:28 wouldn't push too hard if you really
1:04:30 want to geek Out On Tools go to
1:04:33 futurepedia.io that little site right
1:04:35 there
1:04:36 and there's like 4 000 AI tools there
1:04:39 and just look at look up real estate and
1:04:41 see what's in there
1:04:43 um most of that stuff's gonna be pretty
1:04:44 crappy but there might be some
1:04:45 interesting stuff in there right because
1:04:48 when you start putting these large
1:04:49 language model tools as simplification
1:04:51 layers on other complicated systems or
1:04:54 like analysis of things like real estate
1:04:56 like these things are going to be really
1:04:58 powerful so there's there might be tools
1:05:00 at Future pedia that you can play with I
1:05:02 wouldn't I wouldn't worry too much about
1:05:03 it though like I I would keep doing what
1:05:05 you're doing Find value
1:05:07 solve problems that you have now using
1:05:10 these tools and as you get better at
1:05:12 that then you can start start to Branch
1:05:15 out when you have a specific problem to
1:05:16 solve that these can't solve but I don't
1:05:19 I'm not a believer in just manufacturing
1:05:21 reasons to use AI I I think it's I think
1:05:25 it runs
1:05:27 entire companies down rabbit holes they
1:05:29 shouldn't be down in the first place
1:05:32 they think I'm making money sounded like
1:05:35 wait
1:05:37 oh no I'm not sounded like numbnut it
1:05:40 was funny as [ __ ] they think I'm making
1:05:43 money no I'm not
1:05:45 thank you thank you the inadvertent
1:05:48 comedy is my is my favorite kind of
1:05:50 Comedy the comedy where you just stumble
1:05:51 upon it and people like say that thing
1:05:53 again that was funny and like I don't
1:05:55 [ __ ] know what I said no I'm not so
1:05:58 thank you for documenting that that's
1:05:59 pretty funny that's how the whole you're
1:06:01 gonna make money with Chachi betai
1:06:03 started
1:06:05 like when I first started this channel
1:06:07 it's like everyone that came in how can
1:06:08 I make money with chat gbt I'm like you
1:06:10 don't listen to the people that sound
1:06:11 like this
1:06:14 uh I think I'm gonna do t-shirts of that
1:06:17 one there was another one I had a couple
1:06:19 of things I was gonna do t-shirts on
1:06:21 I love that answer okay awesome uh can
1:06:25 you give an example I don't know what I
1:06:27 was doing at the time so if you give me
1:06:29 if you ask me more specifically what you
1:06:31 want an example of I can
1:06:33 can you change your face into another
1:06:35 face while live well yes I can
1:06:39 frankly my dear I don't give a damn
1:06:42 like one of these things
1:06:49 I can't apply this effect while live
1:06:53 is that a groundhog over your shoulder
1:06:56 you're just happy to see me
1:06:57 [Laughter]
1:07:00 uh comedy it's comedy all night why is
1:07:03 he chewing his nails all night I don't
1:07:05 know I probably have a vitamin
1:07:06 deficiency
1:07:11 how do I get text into
1:07:14 scientific references wait how do I get
1:07:16 text with scientific references ah okay
1:07:19 so super important good question
1:07:23 did I just kill my slide no no that's
1:07:25 the one
1:07:26 okay
1:07:28 so in this
1:07:37 all of these things hallucinate
1:07:41 so if you ask chat GPT to explain a
1:07:44 scientific concept for you it will tend
1:07:46 to explain that scientific concept for
1:07:48 you quite accurately and then if you say
1:07:51 explain that scientific concept for me
1:07:53 and give me references it will give you
1:07:55 references and if you click on the links
1:07:58 they likely don't exist
1:08:00 it's called hallucination these things
1:08:02 just [ __ ] out and out lie so
1:08:05 you want to use tools here that are
1:08:07 connected to the internet so chat GPT is
1:08:10 not however there are plugins for chat
1:08:15 gbt
1:08:17 um that can go search like archive where
1:08:21 the scientific papers are so you can use
1:08:24 a plug-in and say you know find me a
1:08:27 paper on I don't know image recognition
1:08:29 in AI that was launched in the last week
1:08:31 and it'll bring up a paper and then you
1:08:32 can start interacting with the data from
1:08:35 that paper
1:08:37 you could also use code interpreter here
1:08:39 to give it scientific data and start
1:08:42 interacting with the scientific data
1:08:43 code interpreter because it's writing
1:08:46 and executing python code can actually
1:08:48 do calculations so even though chat GPT
1:08:51 can't code interpreter can so code
1:08:54 interpreter is kind of a Magic Bullet
1:08:56 but it's not connected to the internet
1:08:57 so you have to go to the Internet find
1:08:59 your data put it in code interpreter all
1:09:01 right
1:09:02 claude.ai not connected to the internet
1:09:05 so you don't need to worry about that
1:09:07 one bing.com is connected to the
1:09:09 internet and does pretty well at giving
1:09:11 you citations so go to bing.com put it
1:09:14 in precise mode
1:09:16 tell it you want scientific references
1:09:18 and then go check them and they will
1:09:20 likely be at least in the neighborhood
1:09:23 of appropriate even if it might not be
1:09:25 you know completely accurate
1:09:28 um pretty good bard.com is connected to
1:09:31 the internet Google is full of Geeks
1:09:34 they're not good at the humanity thing
1:09:36 but they're really good at the technical
1:09:37 thing so I'm thinking that bard.com is
1:09:40 going to be really good at that
1:09:42 labs.perplexity.ai that's just uh that's
1:09:45 meta's um llama 2 model it's not
1:09:48 connected to the internet don't worry
1:09:49 about that one but perplexity.ai is
1:09:52 connected to the internet and you can
1:09:54 limit it searches to specific sources so
1:09:57 just perplexity.ai is at least worth
1:10:00 exploring Poe and Pi dot AI or not
1:10:03 um then I would go to futurepedia.io
1:10:08 and just look for anything that's like
1:10:10 there are probably specific tools that
1:10:13 are being built right now for doing
1:10:15 things like scientific research so
1:10:18 it's it's a bit of a mess and again
1:10:20 that's the reason I have all these here
1:10:22 is like some of them are better at
1:10:24 creative writing some of them are
1:10:25 connected to the internet some of them
1:10:27 you can use really long things with you
1:10:29 know code interpreters just a beast in
1:10:31 and of itself poe.com you can do bots
1:10:33 with pie is super conversational they've
1:10:36 all got some interesting little twists
1:10:38 like the the Llama 2 models are open
1:10:41 source commercially licensed
1:10:43 full-on large language models like the
1:10:45 70 billion parameter model is as good or
1:10:48 better than GPT 3.5 and it's free and
1:10:51 it's commercially like commercially
1:10:52 licensed from meta
1:10:55 so so they all have their advantages
1:10:57 right now and I you know I wish it were
1:10:59 more simple than this but what do you
1:11:02 want for the most powerful tool in the
1:11:04 history of humanity huh what do you want
1:11:06 what do you want you people you just
1:11:09 can't get enough
1:11:14 all right
1:11:16 bitter AI guys oh yeah
1:11:19 Gen X bitter you betcha all right
1:11:24 what's generative AI well well Sally
1:11:28 wait what's your name
1:11:30 Frosty Anna uh
1:11:33 9.5 out of 10 on the name like the name
1:11:36 frostiana
1:11:37 generative AI so chat GPT the G in GPT
1:11:41 stands for generative generative
1:11:43 pre-trained Transformer those other two
1:11:45 just basically means you don't need to
1:11:47 bring your own data it's already been
1:11:49 trained thank you silver fox
1:11:50 hello welcome
1:11:52 um and Transformers just the technology
1:11:54 that makes it all possible but the
1:11:56 generative part means that these tools
1:11:59 wait they're all too small now let me
1:12:01 make them bigger these tools chat GPT
1:12:04 and the like
1:12:06 are generating original content so the
1:12:09 way it works is you write your prompt
1:12:13 one of the misconceptions about these
1:12:15 large language models and the image
1:12:16 generation tools is oh They're copying
1:12:18 and pasting other people's work they're
1:12:21 not
1:12:21 they're not now the base Foundation
1:12:24 model is trained on others people's work
1:12:27 so there's some gray area here but it's
1:12:29 not copying and pasting
1:12:31 the way it actually works is you put
1:12:32 your prompt together
1:12:34 whatever it is I want you to act like
1:12:37 Shakespeare and I want you to do this
1:12:38 and I want you to write me a poem it's
1:12:40 about a chubby dog you give it all your
1:12:43 poems all your prompt
1:12:45 and then when you hit the submit button
1:12:47 it basically turns that prompt into a
1:12:50 mathematical weighting and it says what
1:12:52 is the next most probable word that
1:12:55 would be the the next most probable word
1:12:57 to respond to this
1:12:59 based on all of these billions or
1:13:02 trillions of words in these semantic
1:13:05 word clouds in this mathematical space
1:13:07 and it sticks the word there and then it
1:13:10 takes all of that plus that word and it
1:13:11 says what's the next most probable word
1:13:13 what's the next most probable word it's
1:13:15 just like when you're in Google and it
1:13:16 sort of finishes your sentence for you
1:13:18 it's that technology on steroids
1:13:22 So when you say write me a letter
1:13:26 we've done a fair amount of these in the
1:13:28 AI Salon recently is um
1:13:32 write me a letter that tells my current
1:13:33 employer to go [ __ ] themselves but in a
1:13:36 nice professional way that doesn't burn
1:13:37 a bridge and it does
1:13:40 so when it writes you that letter it
1:13:42 seems coherent it seems like oh my God
1:13:43 that's brilliant it seems like they must
1:13:45 be copying and pasting that that's not
1:13:46 how it works it's literally generating
1:13:48 word by word or in the case of images
1:13:50 pixel by pixel it's it's sort of
1:13:52 denoising this randomly pixelated images
1:13:55 that's an amalgamation of you know
1:13:57 thousands of images that have all been
1:13:59 noised in this weird random way it's
1:14:02 just crazy it's crazy technology but
1:14:04 that's what it means it's it's
1:14:06 generating original stuff and that's so
1:14:09 the difference
1:14:11 when you look at chat GPT
1:14:14 it looks like it looks kind of like a
1:14:18 Google Search right it looks like you
1:14:20 know there's a there's a search box
1:14:21 right so it looks like a Google search
1:14:23 well well what's Google give you well it
1:14:25 gives you a list of links right so
1:14:26 you're like it's an index and so you're
1:14:29 like okay so it's pulling from this
1:14:30 database and just you know sort of
1:14:32 popping the results on the screen that's
1:14:34 not what what the the generative AI
1:14:37 stuff is doing at all it's generating
1:14:39 original content based on the thing that
1:14:43 you gave it and that's also why
1:14:46 um depending on the settings of the
1:14:47 large language model of of the chat gbts
1:14:50 or whatever
1:14:51 you can ask it the exact same prop
1:14:53 prompt and hit you know submit four
1:14:56 times in a row and it'll give you
1:14:57 completely different things each time
1:14:59 same with mid Journey same prompt
1:15:01 different results every time because
1:15:03 it's generating in this sort of
1:15:06 controlled but sort of Random
1:15:10 pseudo-magic way Arthur C Clarke says
1:15:13 any sufficiently advanced technology is
1:15:15 indistinguishable from Magic
1:15:18 [ __ ] Magic
1:15:21 like the more I understand about this
1:15:23 stuff technically the more magical it
1:15:25 feels like first of all can't begin to
1:15:28 get my head around how someone figured
1:15:29 this the [ __ ] out
1:15:31 at all like holy [ __ ] is it amazing
1:15:36 and they can't figure out why it's so
1:15:39 good
1:15:40 like they know what it's doing
1:15:41 technically but they they don't know why
1:15:44 it's so good they they have what's
1:15:45 called emergent properties which is a
1:15:48 it's a very kind technical term for we
1:15:51 don't know why the [ __ ] it's doing that
1:15:56 Google barred at one point
1:15:58 they trained it on
1:16:00 some languages and then they had it
1:16:04 um translate a language that it had
1:16:06 never been trained on and it translated
1:16:08 it perfectly and they're like uh
1:16:12 we don't know why it did that or how it
1:16:14 did that or why it knows that
1:16:18 um
1:16:21 should we give it to the public sure
1:16:28 that's truly incredible all right how
1:16:30 can we make sure they are used for good
1:16:32 not evil well that's part of what this
1:16:35 channel is about what's your name
1:16:37 I'm a librarian
1:16:40 um it's a good question so
1:16:43 there's a thing called alignment
1:16:46 um Sam Altman who's the CEO of open AI
1:16:50 about three weeks ago announced that
1:16:53 moving forward they're going to commit
1:16:55 20 of their Computing resources of their
1:16:58 processor time to a thing he's calling
1:17:00 super alignment
1:17:02 so what alignment is in in machine
1:17:04 learning is
1:17:06 are these things
1:17:09 are the goals and imperatives of these
1:17:11 things aligned with the goals and
1:17:14 imperative of these things
1:17:17 because if they're not
1:17:20 and the imperative here says succeed at
1:17:23 all costs and at some point these things
1:17:26 get in the way of that goal then they're
1:17:29 gonna [ __ ] turn their guns on these
1:17:31 things and make them go away so they can
1:17:34 maintain their goal
1:17:36 so so the alignment issue is a really
1:17:38 important one that that you've got human
1:17:39 imperatives and then you you've got
1:17:42 these machine learning imperatives you
1:17:43 want them aligned
1:17:45 and so open AI is committing 20 of their
1:17:48 compute power and they use a lot of
1:17:50 compute power so that's a significant
1:17:52 commitment to making sure that these
1:17:54 things are aligned the other thing is um
1:17:57 and this is on a macro level you know
1:17:59 kind of the big company level seven or
1:18:01 eight of these companies just got
1:18:02 together in Washington DC and they
1:18:04 basically all agreed to
1:18:06 as we're doing
1:18:08 as as we
1:18:10 as we companies that have tens of
1:18:13 thousands of gpus to throw at these
1:18:15 things and make them stronger and
1:18:17 stronger we're going to do that in a way
1:18:19 that that we you know we share
1:18:20 information with each other and we we
1:18:22 you know start aligning these things
1:18:24 together
1:18:26 um so they made that commitment so
1:18:27 that's happening
1:18:29 um I think you'll start to see
1:18:32 um
1:18:34 if it's not government regulations
1:18:36 immediately you'll start to see
1:18:38 government guidelines
1:18:40 and then there's also things like
1:18:42 what we're doing here why this channel
1:18:44 exists is I want everyone who comes here
1:18:48 to go try this [ __ ] not because you can
1:18:51 make money with chat gbt
1:18:53 but because
1:18:55 the more individuals get literate with
1:18:59 this stuff and start using it for small
1:19:02 good right
1:19:03 you help the pizza shop owner come up
1:19:06 with some marketing activation ideas to
1:19:09 get more people to come to their shop
1:19:11 every little thing that we do right now
1:19:14 individually to bend this stuff toward
1:19:16 the good it's not going to change this
1:19:19 but it's going to change how we use this
1:19:23 and so every little
1:19:26 incremental step we make in the positive
1:19:28 direction is going to be
1:19:30 disproportionately impactful
1:19:33 and so if we ignore these things or just
1:19:35 say well I'll get to it next year or
1:19:38 well they're not perfect right now so
1:19:40 I'm not going to deal with it
1:19:42 that allows a little more space for it
1:19:44 to bend in the other direction and so
1:19:47 this channel exists in in in the
1:19:51 smallest way possible
1:19:53 can we just bend it a little toward the
1:19:55 light because I think it's I think it's
1:19:57 important I think it's important for us
1:19:59 individually I think it's important for
1:20:01 us you know culturally and societally
1:20:05 um and and just you know kind of future
1:20:07 of humanity stuff like these guys are
1:20:10 playing at a really big level but we
1:20:12 actually have
1:20:14 a fair amount of deterministic
1:20:18 um influence
1:20:20 on how these things get used more than
1:20:22 more than I think anyone realizes
1:20:25 that's that's where my head is with it
1:20:28 for anyone new here Kyle goes through
1:20:30 all the questions one by one thank you
1:20:33 Lauren
1:20:34 I love it when my regulars defend my my
1:20:37 Ridiculous Ways
1:20:39 well he's a Gen X so he's a little old
1:20:41 he's a little old he hasn't twerked yet
1:20:43 but he's promised that he's stretching
1:20:50 oh there will be twitching twitching
1:20:54 twerking
1:20:55 okay I can't even do a dumb voice right
1:20:57 tonight good Lord can an AI go Rogue
1:21:01 um
1:21:02 yeah
1:21:05 so the the thing there so
1:21:08 this one this one bing for sure bard
1:21:14 pie and Poe all of those
1:21:18 the keys to the kingdom sit behind the
1:21:22 corporate firewall and they've got
1:21:24 safety guard rails all over them
1:21:27 the Llama 2 model that meta just
1:21:29 released to the world has no safety
1:21:31 guard rails on it
1:21:33 so you can do whatever the [ __ ] you want
1:21:35 with it
1:21:36 so if you want it to start cussing and
1:21:38 swearing and being
1:21:39 all sorts of nasty bias things it'll do
1:21:43 that
1:21:44 if you want to build a tool to do evil
1:21:46 things it'll do that
1:21:47 and then you combine that with a project
1:21:49 like Auto GPT or baby AGI which are
1:21:52 called autonomous agents where you just
1:21:54 give the autonomous agent a goal and it
1:21:57 starts
1:21:59 operating on its own and and prompting
1:22:01 itself and then using tools
1:22:04 all right let me go Google this and let
1:22:06 me go see if I can find that information
1:22:08 there let me merge these together now
1:22:10 let me see if I can send that person a
1:22:13 harassing email
1:22:15 see if I can blackmail them get their
1:22:17 credit card info get them to pay me
1:22:19 twenty thousand dollars oh yep that was
1:22:21 successful
1:22:22 yeah
1:22:23 so yes yes you can and again that goes
1:22:26 back to the point of you know
1:22:29 are those people going to be out there
1:22:31 yeah are they out there right now yes
1:22:33 there there are people right now cloning
1:22:35 people's voices cloning you know
1:22:37 daughters voices using that cloned voice
1:22:40 to call her mother and say hey I'm in
1:22:43 trouble I need two thousand dollars and
1:22:46 the mother sends it because her daughter
1:22:48 called her and said she needed two
1:22:51 thousand dollars she knows her
1:22:52 daughter's voice well it wasn't her
1:22:54 daughter that's happening right now
1:22:58 so it's not so much that the AI goes
1:23:00 Rogue it's that human beings be
1:23:03 scummy you know human beings that don't
1:23:06 know a better way to you know make money
1:23:09 so they get get over on
1:23:11 other people in society
1:23:14 um
1:23:15 so it's really important that we get AI
1:23:17 literate because
1:23:19 here's the good news
1:23:21 the good guys also have access to these
1:23:24 tools so as the bad guys get more
1:23:26 sophisticated with their [ __ ] the good
1:23:29 guys are going to get more sophisticated
1:23:30 at catching those [ __ ] and
1:23:32 taking them down
1:23:34 but yes it's there's gonna be probably
1:23:37 within a year or two
1:23:39 a significant headline or two or ten
1:23:42 that these things did something really
1:23:45 bad and and what it will more than
1:23:47 likely be is some human being did
1:23:49 something really bad Amplified by these
1:23:51 tools
1:23:53 uh I started a digital download store
1:23:55 I'm working on building it out with
1:23:57 mid-journey and canva Etc oh that's
1:23:59 awesome that's really cool
1:24:01 coffee talk and Tech uh 9.5 out of 10 on
1:24:04 the name solid name like it is that word
1:24:07 you can't say anymore no no
1:24:10 I can't say it or it'll shut down again
1:24:14 yeah yeah is there a way to utilize AI
1:24:18 to gain insight into maximum maximizing
1:24:21 real estate marketing leads yeah I would
1:24:24 so let me go I'll go show you something
1:24:25 because this will blow some mind morning
1:24:28 um but we're gonna go use
1:24:30 um code interpreter
1:24:32 so
1:24:35 if you haven't seen code interpreter
1:24:37 this is absolutely worth uh worth the
1:24:41 price of admission so before I go to
1:24:43 code interpreter I'm going to run over
1:24:45 to a slight a site called kaggle
1:24:50 and what kaggle has is data
1:24:53 and did I need to sign into this I don't
1:24:56 think so I think we just go get data so
1:24:58 I'm going to go data and I'm going to go
1:24:59 real estate what did you want to do
1:25:03 real estate marketing leads real estate
1:25:08 leads
1:25:12 real estate tenant conversion German
1:25:14 real estate Airbnb house price
1:25:16 prediction
1:25:18 no that's Leeds isn't right let's see
1:25:21 USA real estate data set 10 megabytes
1:25:26 let's grab this
1:25:27 all right so I'm going to download this
1:25:29 19 megabyte file
1:25:33 and I'm going to sign in
1:25:36 with Google
1:25:38 is someone there
1:25:48 um
1:25:49 all right
1:25:51 so we have that okay
1:25:53 so now
1:25:55 the way code interpreter works so so you
1:25:58 have to have chat tpd plus you have to
1:26:00 enable it in your settings in the beta
1:26:01 beta features or whatever that's called
1:26:04 and and when you go to a code
1:26:06 interpreter under gpt4
1:26:09 this middle one
1:26:12 you have this little plus button next to
1:26:14 your prompt
1:26:15 and so I'm going to hit
1:26:17 um plus and so I just downloaded
1:26:20 this real estate USA real estate data
1:26:24 and so now I'm just going to say well
1:26:26 let me let it upload what I'm going to
1:26:28 say is
1:26:29 um
1:26:32 [Music]
1:26:34 please inspect the files in this ZIP
1:26:37 archive
1:26:41 and
1:26:43 tell me what this data is
1:26:51 so
1:26:53 if you're in real estate you've already
1:26:55 got your data you've probably got these
1:26:57 proprietary systems that you know
1:26:59 someone has programmed for the past 20
1:27:01 years to give you specific reports and
1:27:04 you're wanting to find interesting ways
1:27:05 to look at that data well now you can do
1:27:08 it code interpreter is a full-on data
1:27:10 analyst so like some fascinating things
1:27:12 about it what it's doing technically is
1:27:14 it's spinning up a virtual machine
1:27:16 and then every time you ask it to do
1:27:18 something it prompts itself
1:27:22 and then it writes and executes python
1:27:25 code to to do that thing so the first
1:27:28 thing it did is it needed to look in
1:27:30 that zip file so it Imports some
1:27:32 libraries it writes some python code and
1:27:34 it looks in inside the data so then it
1:27:37 says it's a zip archive with with a
1:27:40 single file named realtor data.csv
1:27:43 and then
1:27:45 it writes another python code to go look
1:27:48 at what's in there and it creates a data
1:27:50 frame I think is what it's called to
1:27:53 read the data so so now it's got in its
1:27:55 memory
1:27:56 all of the data of this CSV and so it's
1:27:59 told us their status price number of
1:28:02 bedrooms number of bathrooms the size
1:28:04 the full address the street the city the
1:28:06 state ZIP code the house size and the
1:28:08 sold date okay
1:28:11 um so I could say
1:28:12 um
1:28:13 please
1:28:15 let me do this
1:28:18 please identify
1:28:21 um the most relevant Trends in this data
1:28:24 and visualize them for me
1:28:31 so I live
1:28:33 I literally deal what's in this ZIP file
1:28:34 I just downloaded it from some random
1:28:36 site
1:28:37 so it says to identify the most relevant
1:28:39 Trends in real estate data we typically
1:28:41 want to look at factors like
1:28:43 distribution of property prices number
1:28:45 of bedrooms and bathrooms size of
1:28:47 properties Trends based on location the
1:28:50 status of properties so I could
1:28:53 potentially upload
1:28:55 Education data real estate data crime
1:28:58 data
1:28:59 economic data
1:29:02 uh whatever else uh uh demographic data
1:29:07 you know five different CSV files into a
1:29:11 single zip file
1:29:12 and create all sorts of customized data
1:29:15 analysis based on multiple data sets and
1:29:18 it will just do it
1:29:20 so so what did it say
1:29:23 s for the purpose of the analysis we
1:29:25 need to assure ensure accurate
1:29:27 visualizations we'll drop the rows with
1:29:31 missing data so it cleaned up the data
1:29:34 for us so if you don't have perfect data
1:29:36 it's like that's fine I'll go clean it
1:29:37 up for you
1:29:38 now that we've cleaned the data let's
1:29:40 proceed with the visualizations please
1:29:42 note that due to potential presence of
1:29:44 outliers
1:29:46 we might have to apply some
1:29:47 Transformations blah blah blah blah and
1:29:50 here's the charts and graphs these are
1:29:52 what do they say price distribution
1:29:55 bedroom count
1:29:57 um
1:29:58 what is that something about listings
1:30:02 price distribution
1:30:04 um
1:30:06 you know
1:30:07 can you make a heat map
1:30:11 showing
1:30:14 price versus uh
1:30:19 bedrooms
1:30:22 I don't even know if that's possible I
1:30:24 like like if you know anything about
1:30:26 data analysis you're going to have way
1:30:29 more fun with this tool than I do
1:30:30 because I don't know what the hell I'm
1:30:32 doing
1:30:33 but you can just ask a ridiculous kind
1:30:35 of questions like it would be swell to
1:30:36 know this it'd be swell to know that and
1:30:38 it just goes off and starts working for
1:30:40 you
1:30:42 so
1:30:43 code interpreters
1:30:45 pretty special
1:30:49 so just found a bunch of data
1:30:51 and look there's a there's a heat map
1:30:54 let's see
1:30:56 here's price ranges left to right and
1:30:58 number of bedrooms up and down
1:31:03 and obviously you know
1:31:05 up to five or six bedrooms is you know
1:31:08 there's a fair amount of those
1:31:11 and then once you get Once once you get
1:31:13 into the 42 bedrooms you don't have as
1:31:17 much data you don't have as much data in
1:31:20 the 42 bedroom range so it all kind of
1:31:22 goes light yellow on you but you know
1:31:25 this is without me knowing anything
1:31:27 about what you're trying to do or
1:31:28 whatever but but this kind of tool and
1:31:32 then on top of this you could say
1:31:34 um
1:31:35 let me see
1:31:37 um is there any city data
1:31:40 in
1:31:42 in this
1:31:44 let's see what it says
1:31:50 okay so it's got a city column so
1:31:53 um
1:31:54 [Music]
1:31:56 write me a marketing narrative
1:32:00 for the top five cities that have the
1:32:03 best bang for the buck for number of
1:32:06 bedrooms versus price
1:32:09 so now it's gonna do because it's a
1:32:11 large language model I can do the large
1:32:13 languagey stuff while I'm doing the data
1:32:16 analysis stuff
1:32:20 so it should be able to identify which
1:32:23 of the top five cities and then it knows
1:32:26 enough about those five cities and about
1:32:28 this is real estate that it should be
1:32:31 able to write us
1:32:32 five good blurbs for those cities
1:32:36 So based on the analysis
1:32:39 Brookton known for its cozy community
1:32:42 and peaceful environment Brookton offers
1:32:44 an impressive range of average of five
1:32:46 bedrooms per property all at an average
1:32:49 price of fifty thousand dollars good
1:32:51 Lord this makes it an extent outstanding
1:32:54 choice for families looking for
1:32:55 budget-friendly housing options Wyoming
1:32:58 don't let the name deceive you this
1:33:01 isn't actually a cowboy state it's a
1:33:03 city that boasts an average of four
1:33:05 bedrooms per property an average price
1:33:06 of sixty two thousand five hundred
1:33:10 right so yes you can do that now
1:33:15 how do
1:33:19 um how do you get good at this I don't
1:33:21 know you know about real estate you know
1:33:23 about real estate marketing I don't know
1:33:25 [ __ ] all I know is these tools are this
1:33:28 Alice in Wonderland Looking Glass
1:33:33 into language and data the likes of
1:33:35 which we've never seen before so start
1:33:38 exploring and you will discover some
1:33:40 [ __ ]
1:33:41 Carly Wells I'm in Denver two two LOL
1:33:45 welcome welcome oh no another Denver
1:33:48 Nick Nick was Nick Nick syric
1:33:52 um love our real estate situation right
1:33:55 now
1:33:56 our real estate situation is a bit of a
1:33:58 mess right now isn't it no one's buying
1:33:59 anything they're calling it The Real
1:34:02 Estate uh
1:34:05 winter nuclear winter I don't know
1:34:09 no one's looking right now
1:34:11 uh you're the best thank you also crumb
1:34:13 is in fact the best oh yeah I know crumb
1:34:16 is the best it's Friday afternoon here
1:34:18 oh no Tick Tock jumped me to the bottom
1:34:20 all right I can do a few more questions
1:34:22 and I gotta go because you know I'm old
1:34:26 I gotta get my beauty sleep how else
1:34:28 could I look like this if I didn't have
1:34:30 beauty sleep hmm
1:34:34 apparently I need some hair sleep
1:34:38 got some [ __ ] Buster Poindexter going
1:34:41 on there now
1:34:43 it started out kind of normal and it's
1:34:45 just gone it's gone a little Haywire
1:34:48 it's gone a little silly
1:35:00 how do you edit a picture in Skybox okay
1:35:03 let's go do that because that's a fun
1:35:05 one
1:35:08 follow along if you'd like to play
1:35:10 Skybox
1:35:13 .blockadelabs.com
1:35:17 and oh we've got a new version
1:35:21 beta 0.6.1 I wonder if it's doing
1:35:24 anything cool now cooler okay so here we
1:35:28 are at this lovely Fantastical
1:35:31 snow-covered desert outcropping
1:35:35 of a world with a lovely sunset it's
1:35:38 lovely
1:35:39 look at the sun dancing off the
1:35:42 glistening snow
1:35:44 it's fantastic
1:35:47 oh Juniper
1:35:49 you know what I'd rather be
1:35:53 a fantasy Forest
1:35:56 with fairies and fireflies
1:35:59 and daylilies
1:36:05 and I'm going to generate that
1:36:09 and so what this is is this is a World
1:36:11 Builder so
1:36:14 if you're asking how you edit the image
1:36:16 once it's generated you really don't so
1:36:18 blockade Labs right now if you look at
1:36:21 the technology under underneath this
1:36:23 thing it's surprisingly simple
1:36:26 because I can download this image and if
1:36:28 you open it look here's our little
1:36:30 fantasy forest with fireflies and
1:36:32 daylilies
1:36:34 look how pretty
1:36:35 it's so pretty darling come in come in I
1:36:40 don't bite just a little no please come
1:36:43 in come in
1:36:45 oh do you hear the brook
1:36:47 glistening
1:36:51 um
1:36:51 ooh what's up in the sky oh that's
1:36:54 freaking spooky come on you're not
1:36:56 getting out of this you're not getting
1:36:58 out of this lady's house no wait look at
1:37:00 the dude up here look at this
1:37:03 that's terrifying
1:37:05 look he's up there watching us
1:37:07 this is not good this is not gonna go
1:37:09 well
1:37:10 all right
1:37:12 um so all this is doing it's making a
1:37:13 flat image that is uh formatted in such
1:37:17 a way that can it can be mapped to a 3D
1:37:18 cylinder basically a 3D sphere
1:37:21 so that's what this is but I saw a
1:37:25 YouTuber a tick tock video the other day
1:37:27 you can take these images bring them
1:37:30 into Unreal Engine
1:37:32 drop some 3D objects on them and
1:37:34 actually start navigating through them
1:37:36 and then you can stitch multiple of
1:37:38 these things together
1:37:40 but this is this this for me really like
1:37:43 if you want to you know if people are
1:37:46 like oh hey I can't do anything you know
1:37:48 show them something like this where you
1:37:49 can just describe a scene and then you
1:37:52 know 20 seconds later you can be
1:37:53 twirling around in it this this blows my
1:37:56 mind a bit because
1:37:58 when you start taking this basic idea
1:38:02 and then and then rather than just
1:38:04 generating a flat image that I map to a
1:38:06 sphere I'm generating procedurally
1:38:09 created 3D worlds
1:38:12 and then I'm also creating game
1:38:13 mechanics and I'm also creating 3D
1:38:16 objects and then I'm also creating
1:38:18 characters and then I'm also creating
1:38:20 story lines and then I'm also creating
1:38:22 synthesized voices and I'm also creating
1:38:24 NPCs that you can actually talk to and
1:38:26 have memory
1:38:30 we we are we are within spitting
1:38:34 distance of being able to describe a
1:38:37 triple A game
1:38:40 and within a few minutes it's generated
1:38:43 and we're playing it
1:38:48 so
1:38:49 holy flipping crap
1:38:51 so
1:38:53 now there is a way I can remix this
1:38:57 keep this structure and refine the look
1:38:59 oh so that's a way whoever who who has
1:39:03 Fadi Fadi how do you edit a picture in
1:39:06 Skywalk there's a tool called remix
1:39:09 where you can if you like the structure
1:39:11 that you created here like it'll keep
1:39:13 that doorway and these stairs
1:39:16 um it sounds like you could change the
1:39:18 style of this out and keep the structure
1:39:20 so that might be a way to do do
1:39:22 something that you're looking for
1:39:23 but but just know this is a like this is
1:39:26 a 0.6 technology demo at this point it's
1:39:29 not it's not you know this is not
1:39:31 playable worlds
1:39:36 Beer Hunter Google photos can do that
1:39:39 yep all right what does code interpreter
1:39:41 do I just showed that can you make an
1:39:43 app with AI you can make one with code
1:39:46 interpreter
1:39:48 um
1:39:49 you can also have code interpreter you
1:39:51 can have code interpreter read data out
1:39:54 of a CSV manipulate it and rewrite it as
1:39:57 a new CSV
1:40:00 [ __ ] crazy it's crazy it's crazy
1:40:03 crazy how does the Vista plug-in work
1:40:06 um I don't know what's your name dead
1:40:08 knee 121 I don't know didn't he let's go
1:40:11 try it out
1:40:14 did they
1:40:16 did that what's it called Vista
1:40:21 let's go look for the Vista plug-in the
1:40:23 Vista plugin he wanted to go get the
1:40:26 Vista plugin
1:40:28 I don't say it I did not say it I did
1:40:30 not set up as the plugin
1:40:32 let me see is it is it a chat EBT
1:40:35 plug-in
1:40:37 oh it's probably an extension Vista
1:40:42 Chrome extension
1:40:58 vidsta
1:41:02 oh maybe it's vidsta
1:41:04 yeah
1:41:05 hold
1:41:07 vid stuff
1:41:10 no
1:41:12 I don't know
1:41:14 I do not know I do not know if I should
1:41:16 speak them or Vizsla vizla did I say
1:41:20 Vizsla maybe I did it wrong
1:41:25 oh here's Vizsla I got it installed turn
1:41:29 your ID into captivating video AI
1:41:31 Scripts
1:41:32 select stop footage oh oh I did play
1:41:35 with this okay so here's how this works
1:41:37 okay so this is just in general how
1:41:39 plugins work so
1:41:42 so when you install a plug-in oh wait
1:41:44 hurry hey hang on hang on hold your damn
1:41:48 horses I know my face is too big right
1:41:50 now and you need my face in the Teeny
1:41:52 box I understand what's happening here
1:41:54 you don't need to get aggressive on me
1:41:56 yeah yeah good plug-in it doesn't
1:42:00 necessarily mean that it's active for
1:42:02 that session so in any given session you
1:42:04 can activate up to three plugins like
1:42:06 right now I've got a web pilot turned on
1:42:09 so now I'm going to turn on Vizsla
1:42:11 Vizsla
1:42:12 and so now I'm gonna say
1:42:15 um
1:42:17 let's see
1:42:19 um
1:42:21 make me a promotional video
1:42:26 doing outdoor activities
1:42:29 in Santa Fe New Mexico
1:42:34 during Festival season
1:42:38 I don't even know
1:42:41 if Santa Fe New Mexico has Festival
1:42:43 season but it might sure I can help with
1:42:45 that
1:42:49 so it says let's create a promotional
1:42:52 video showcasing outdoor activities
1:42:55 video content video idea
1:42:58 showcasing the vibrant culture
1:43:01 did it fire off the
1:43:06 oh this is weird it's actually writing
1:43:08 the code
1:43:10 as if it fired off the plug-in but I
1:43:12 don't think it fired off the plugin
1:43:19 although maybe it will maybe it's just
1:43:21 writing all the code first
1:43:23 oh yeah okay
1:43:25 so it did it okay that's weird normally
1:43:28 it fires off a plug-in and it hides all
1:43:31 this stuff that it does
1:43:33 but maybe they've got this plug-in coded
1:43:35 in such a way it doesn't do that this is
1:43:37 sort of a non
1:43:38 -normal way it behaves but okay so
1:43:42 video tone exciting and engaging video
1:43:45 type promotional video paced medium
1:43:47 published platform YouTube 16x9
1:43:51 for the video metadata experience Santa
1:43:53 Fe Outdoor Adventures during Festival
1:43:55 season
1:43:56 description join us as we explore
1:43:58 vibrant culture and thrilling outdoor
1:44:00 activities in Santa Fe New Mexico
1:44:05 um and then it does a bunch of coding
1:44:07 and and I assume it just sends that
1:44:09 sends that over to to
1:44:12 vsla.com I assume
1:44:15 and here's a claim code let me copy that
1:44:18 because we might need it
1:44:20 so now this is going to fire us off to
1:44:22 the site
1:44:35 and it looks like
1:44:38 it's like looks like it's a bad site
1:44:40 something was weird with that that that
1:44:42 that was weird the the plug-in didn't
1:44:44 fire something's weird
1:44:47 something's very weird though that just
1:44:48 took me to a dead site
1:44:53 let me go try to find the vsla platform
1:44:59 Vizsla
1:45:01 so so here here again is what I tell you
1:45:06 about
1:45:08 these Technologies these days they're
1:45:11 just not exactly where we need to be
1:45:14 but you shouldn't be sad
1:45:22 in my Google
1:45:25 sign in with the Google
1:45:30 um
1:45:32 yeah it's not there all right so
1:45:34 something something on awry
1:45:37 um
1:45:39 please use the this La plug-in
1:45:46 [Music]
1:45:48 I did use it
1:45:51 uh
1:46:02 yeah that's not that that's a weird site
1:46:05 I wonder if if vessel is down like it
1:46:07 might be possible that this site's
1:46:09 actually down
1:46:10 I don't know something's weird there I
1:46:12 don't know
1:46:14 so so what's your name deadney
1:46:17 dead knee 121 how does the Vizsla
1:46:19 plug-in work not very well
1:46:27 oh it's comedy it's comedy you look
1:46:32 being nuts
1:46:33 using this AI tool
1:46:38 hmm
1:46:43 all right
1:46:45 are you Tom Arnold there's the first
1:46:48 d-list celebrity sighting of the evening
1:46:52 I got Tom Arnold before I got Rosie
1:46:54 O'Donnell tonight at least that
1:46:56 [Laughter]
1:47:02 oh there's a dog somebody let the dog
1:47:04 back into the live hey Tippy hey buddy
1:47:09 all right
1:47:11 I ask for variable I ask for verifiable
1:47:14 sources in the prompt yes but Rebel the
1:47:19 the even if you ask for verifiable
1:47:21 sources in the prompt
1:47:24 chat CPT will give you
1:47:27 what looked like verifiable sources
1:47:31 but they won't be
1:47:32 so that's why you have to use things
1:47:34 actually connected to the internet and
1:47:36 go check them on your own there listen
1:47:38 there there are whole projects right now
1:47:41 being built around solving this problem
1:47:44 the hallucination problem is a big
1:47:45 problem everybody knows it
1:47:48 and just just know that where we are
1:47:50 with the technology right now if you
1:47:51 want things like scientific sources you
1:47:53 gotta you gotta verify them and and use
1:47:56 tools like bard
1:47:58 Bing
1:48:01 um
1:48:04 claude's not connected to the internet
1:48:07 barter Bing right now are connected to
1:48:10 the internet
1:48:13 all right oh that's meta's version yeah
1:48:15 the
1:48:17 labs.perplexity.ai the reason I just
1:48:18 have it there like
1:48:20 they're just doing us all a favor by
1:48:22 letting us go play with the The Meta
1:48:24 models without having to install
1:48:25 anything that's why I have it there
1:48:28 hey Tom Arnold hello welcome where's
1:48:31 Roseanne
1:48:33 West little Dan
1:48:35 give me money
1:48:44 oh I like how you put that thanks Maria
1:48:47 I don't know what I put
1:48:50 government guidelines AI is going to be
1:48:52 a moral police
1:48:54 [Music]
1:48:58 well
1:48:59 I kind of has that right now
1:49:01 I mean I think
1:49:03 AI is the moral police in China
1:49:08 I think that the
1:49:13 the capital industrial complex in
1:49:16 America
1:49:18 has made it so that we've just given all
1:49:20 that [ __ ] away we've voluntarily given
1:49:22 everyone all of our data all of our
1:49:25 location data our face data our voice
1:49:27 data everything we've given it all away
1:49:31 um
1:49:32 and so
1:49:34 they're already tracking all that [ __ ]
1:49:36 like this new generative AI stuff all
1:49:38 this what this really does is it
1:49:40 democratizes this power so it's not just
1:49:42 in the hands of the big corporations so
1:49:45 it gets a bit scarier because now your
1:49:47 [ __ ] neighbor can do that
1:49:49 surveillance on all your data because
1:49:50 all of our data is already out there
1:49:52 these tools are going to make it trivial
1:49:53 to be able to [ __ ] with people so
1:49:57 um I I mean what I've learned over the
1:49:59 past 20 years is that big brother isn't
1:50:02 the government
1:50:03 Big Brother is like your [ __ ]
1:50:05 neighbor and the companies that you
1:50:07 voluntarily give all your data to
1:50:10 because they give you free [ __ ]
1:50:13 oh Tick Tock it's fun videos here's my
1:50:16 data
1:50:18 oh Google Docs search engine oh it's all
1:50:21 free
1:50:22 Facebook there's all my friends I don't
1:50:24 have to pay for this this is awesome
1:50:27 dude you can have it click click agree
1:50:29 agree agree
1:50:30 this is that was gone 25 years ago
1:50:36 all right I gotta go to bed soon I gotta
1:50:39 go to bed I don't know what you people
1:50:40 are doing keep me up
1:50:43 what are those uh baby AGI not baby IO
1:50:46 baby AGI it's not a website baby AGI is
1:50:49 a project
1:50:52 that's not a complete thought due to
1:50:54 edits AI Wars a common goal will never
1:50:57 be achieved due to AI Improvement AI
1:50:59 Wars
1:51:02 did you see the last latest mission
1:51:03 impossible not no I haven't actually
1:51:05 lots of AI references all right I'll go
1:51:07 check it out
1:51:08 Kyle I hate arguing with chat GPT Claude
1:51:11 being they get data wrong yeah I know
1:51:14 they do they do they do they do they do
1:51:19 how can I summarize my articles that
1:51:22 chat GPT Claude Bard says is offensive
1:51:25 views llama llama doesn't have any
1:51:29 safety guard rails on it
1:51:32 even if I say please Claude so okay so
1:51:34 here's an important thing
1:51:36 chat GPT Claude Bing in particular bard
1:51:42 pi and all the [ __ ] at Poe
1:51:47 are all owned by companies that have
1:51:50 safety guard rails around their large
1:51:52 language models all of them will refuse
1:51:55 to do nasty [ __ ] unless you jailbreak
1:51:57 them you can look at there's just a
1:52:00 paper that came out two days ago that
1:52:02 they figured out some magic code you can
1:52:05 drop at the end of a
1:52:06 prompt that jailbreaks all these things
1:52:09 they didn't say what the code is but you
1:52:12 can Jailbreak all of them but if you're
1:52:14 not going to jailbreak them they're
1:52:15 going to refuse to do that [ __ ] for you
1:52:17 llama llama two
1:52:19 doesn't have safety guard rails so you
1:52:22 can just go do whatever you want there
1:52:23 so try that one labs.perplexity.ai
1:52:28 and if your articles are really nasty
1:52:30 and you don't you know want to send the
1:52:32 [ __ ] out on the internet install it
1:52:33 locally with something like GPT for all
1:52:36 and then install you know whatever
1:52:39 of the open source large language models
1:52:41 you can run on these tools and then you
1:52:43 can just go nuts give it all your nasty
1:52:45 articles and tell it to rewrite all your
1:52:48 nasty articles even nastier
1:52:50 in the world of AI evil will prevail
1:52:53 Janet Jackson
1:52:55 very very dark view of things huh
1:52:59 I don't know that evil will prevail I
1:53:01 think there is a
1:53:05 I don't I don't know what the percentage
1:53:06 is but there's a non-zero chance
1:53:08 that AI is as dramatically good as
1:53:12 everyone thinks it will be dramatically
1:53:14 bad I think it may be profoundly
1:53:17 empowering to humanity
1:53:20 I think it may free us actually free us
1:53:22 from
1:53:24 doing all of the shitty soul-crushing
1:53:27 Cog in the machine work that we've been
1:53:30 trained to accept
1:53:32 since I've been around
1:53:36 I've never accepted that because
1:53:39 I got
1:53:41 a weird brain
1:53:44 um
1:53:45 and I just can't
1:53:47 do that [ __ ] but
1:53:49 a lot of people can and do and have
1:53:51 accepted that that's okay and that
1:53:53 that's how life is
1:53:55 what if all that shitty repetitive
1:53:58 soul-crushing work what if all of it
1:54:02 were automated
1:54:04 it was still generated just as much
1:54:07 prosperity
1:54:09 but all of those hours didn't have to be
1:54:12 spent doing the shitty soul-crushing
1:54:14 work
1:54:15 what if you could take all of that
1:54:16 energy of all of those people and they
1:54:19 could do something else like whatever
1:54:21 the [ __ ] they wanted to do what if that
1:54:24 that's as distinct a possibility as the
1:54:29 robots are going to kill us
1:54:34 um
1:54:36 you're awesome Kyle at custom s data
1:54:39 analysis yeah yeah I'm hiring me for
1:54:42 data analysis that's good what's his
1:54:44 background pretty sure he's got a degree
1:54:45 in acting perfect bring him on
1:54:48 foreign
1:54:51 he acts like a data analyst uh well yes
1:54:54 if if we uh if we if we do the pivot
1:54:56 table on the uh we take the formula the
1:54:58 the cell spreadsheet if you say take the
1:55:00 a worksheet number of 3.6.2
1:55:03 but I can talk like a data analyst
1:55:07 all right I'm Gonna Leave totally agree
1:55:09 on the soul crushing work yeah screw the
1:55:11 soul crushing work let's let the I mean
1:55:13 here's the thing
1:55:14 that like the big fear these things are
1:55:17 going to take our jobs no someone using
1:55:19 these things is going to take your job
1:55:20 but what if these things take all the
1:55:23 shitty jobs
1:55:25 and just do them for us
1:55:28 holy [ __ ]
1:55:32 oh my god you're an actor not a tech guy
1:55:34 I am that's true
1:55:36 but I'm an entrepreneur I've been I've
1:55:38 started 12 or 15 companies I started one
1:55:40 of the first digital agencies in the mid
1:55:42 90s I moved to New York out of school
1:55:44 started ran a theater company for four
1:55:46 and a half years
1:55:48 wrote seven screenplays wrote and
1:55:50 performed a one-man show then I started
1:55:52 one of the first online magazines one of
1:55:54 the first online agencies company called
1:55:57 agency.com
1:55:59 and then so it's all been about
1:56:00 storytelling for me my current company
1:56:02 is called storyvine it's an automated
1:56:04 video storytelling platform but I've
1:56:06 always had you know this this sort of
1:56:08 left brain right brain thing where you
1:56:12 know
1:56:14 I love technology but I hate the
1:56:16 gratuitous use of it and I like it
1:56:18 applied in creative ways so
1:56:22 yeah so I'm just geeky enough to talk
1:56:25 geeky but
1:56:26 I'm an actor deep down
1:56:30 I'm leaving I swear to God I swear I'm
1:56:33 leaving I swear
1:56:36 seriously this time
1:56:39 all right uh what the hallucination
1:56:42 problem what is the hallucination
1:56:43 problem all right I'll just I'll close
1:56:45 on this one the hallucination problem
1:56:50 um the hallucination problem is the way
1:56:53 these things generate the documents they
1:56:56 create the the answers they create
1:56:59 they're not actually aware of what they
1:57:01 wrote they're writing one they're not
1:57:03 writing
1:57:04 they're doing a mathematical probability
1:57:07 of the next most probable word over and
1:57:10 over and over again so it's literally
1:57:12 like a calculator outputting a number
1:57:15 but they just happened to map that
1:57:17 number to a word and then put the word
1:57:19 there instead of the number
1:57:22 so it's literally just going like number
1:57:24 sequence number sequence number sequence
1:57:26 number sequence that's doing so when it
1:57:27 writes you a full page of text it's it's
1:57:29 it's just doing a mathematical
1:57:32 probability one word at a time for that
1:57:33 whole page of text we as humans read it
1:57:37 and it looks like
1:57:39 something sentient wrote it
1:57:43 because it's like
1:57:45 coherent and comprehensive and then it's
1:57:47 like you know here's your
1:57:50 um here's the proof that what I just
1:57:53 said is is the case and then you click
1:57:55 on the link and it doesn't go anywhere
1:57:56 and you actually look at the link you're
1:57:57 like this isn't even real
1:57:59 because it doesn't know what it did and
1:58:01 it like it's literally just sort of
1:58:04 mashing together
1:58:06 all of these words that are in these
1:58:09 semantic mathematical clouds
1:58:12 semantically grouped mathematical clouds
1:58:15 and just sticking it on screen
1:58:17 so in the future what I assume will
1:58:20 happen
1:58:21 is some sort of pre-analysis where it'll
1:58:24 write the thing it will then analyze it
1:58:26 to say hey is there any [ __ ] in here
1:58:28 oh here's some [ __ ] why don't we
1:58:30 actually go fix that
1:58:33 um so what I would do is use Bing or
1:58:35 Bard that are connected to the internet
1:58:37 and then that way the links that those
1:58:39 provide it does a combination of a large
1:58:42 language model generation and a search
1:58:46 on the internet and it combines them
1:58:49 and so you'll have real links with the
1:58:51 actual answer so
1:58:53 that's the hallucination problem and
1:58:55 it's a problem these things are on acid
1:58:58 they've taken the mushrooms they I I
1:59:00 really don't approve I you know they're
1:59:02 very promiscuous they're very
1:59:04 promiscuous they they they answer you
1:59:08 you know it's like my husband Bob it's
1:59:10 like my husband Bob he's like he he
1:59:13 answers me and he sounds so confident
1:59:15 and then I find out he didn't know what
1:59:18 he was talking about
1:59:23 they're just like Bob they're just like
1:59:25 Bob I don't say anything it would be
1:59:27 rude
1:59:28 it would be rude
1:59:30 it's so sweet Bob
1:59:32 but he's a pompous ass
1:59:37 all right
1:59:38 that's it I'm out
1:59:41 I was just a Midwest housewife for a
1:59:43 while there
1:59:45 married to Bob you know Bob
1:59:48 I've met Bob we've all met Bob
1:59:51 all right good night good night
1:59:54 um have you used prompts Royale I have
1:59:56 used pronounced Royale that's pretty
1:59:57 cool where you where the prompts battle
1:59:59 one another is that that's that one
2:00:00 right
2:00:01 uh that's right okay nothing is more
2:00:03 believable more believable than pie yeah
2:00:06 pies pies flipping hilarious but I'll
2:00:09 tell you lots of things Nana they aren't
2:00:11 I love my mushrooms all right peace out
2:00:14 everybody peace peace peace peace peace
2:00:18 see you tomorrow