Mar 4, 2024 - Exploring Claude 3, Sora, and the Latest Innovations

Description

Chapters

Transcript

0:01 [Music]
0:13 [Music]
0:15 there been something baby I've been
0:18 trying to
0:20 say for age and it seems don't know
0:26 how with the past and the future now
0:28 surrounding me
0:30 [Applause]
0:32 surrended to whatever treat can be
0:36 found there's been a little
0:40 trouble since you came to my
0:44 [Music]
0:46 rescue and if you like all of the rest I
0:49 would have quit you long ago but I
0:52 couldn't do
0:53 [Music]
0:56 that oh tell me now women and never went
1:01 too
1:03 well Mak Mak him cold as
1:09 hell I don't let you wish me
1:13 well but in cry still going to have to F
1:18 all way
1:20 [Music]
1:24 through like the Mr morning my dream
1:29 remains hanging in the burn fields of my
1:34 memory time is the chorus of your
1:39 [Music]
1:42 day behind
1:44 me I'm searching Solutions on a
1:48 double on his two turn
1:51 [Music]
1:54 telephone my friend says you ain't got
1:56 no real
1:58 troubles just too many choices if you
2:00 left them
2:04 down my friend says put an action to
2:07 those words put an action to those
2:11 [Music]
2:14 words oh tell me
2:17 now never too
2:21 well make a man crazy make him cold as
2:27 hell I'm a woman that you wish me well
2:30 [Music]
2:32 but in spite TR still going to have to
2:36 find my own way
2:39 through yeah yeah yeah yeah
2:42 [Music]
2:50 yeah happy Monday night
2:53 [Music]
2:59 everybody
3:00 [Music]
3:03 every time I see you
3:05 now get that look in
3:09 my every time I see your mouth I hear
3:13 that
3:15 smile early Misty morning l l h the
3:22 engine in
3:29 the
3:30 [Music]
3:32 you leaving
3:35 me
3:37 today you convince
3:40 me again
3:43 today that you're leaving this hard Town
3:46 looking for someone else's Golden
3:52 Ring say so long
3:58 [Music]
3:59 suan
4:01 don't you
4:05 cry so long to
4:08 [Music]
4:12 Z don't you cry for
4:15 [Music]
4:29 me that is a good dog that is a good
4:32 singing dog that is what it sounds like
4:35 when champ is
4:37 on that was good that's a good dog that
4:41 is chicken stick good that was some
4:44 tasty singing you were on key you
4:47 weren't singing harmonies tonight you
4:49 weren't rapping was
4:51 solid you got through one and a half
4:54 songs that was pretty
4:57 good I know
5:02 [Music]
5:07 hello everybody yes that image is from
5:09 Forever hooked it's it's apparently his
5:13 first in the in the little
5:16 window you do a little branded toilet
5:18 paper you know it's going to make it on
5:20 the AI learning
5:21 [Music]
5:28 lab w
5:32 [Music]
5:47 Freedom came up with a
5:50 [Music]
5:52 night just like a jet plane in and out
5:57 of
5:58 side I was was hauling as a million
6:02 miles an hour wondering how hard I'd
6:09 hit when they came into the
6:15 station they said I was bad Beyond
6:20 repairs but I got no qu with my
6:26 situation say here I am
6:30 [Music]
6:35 so say sh sh Shere W you dare to say
6:42 Shere sh Shere won't you dare to say
6:48 Shere Shere
6:51 Shere
6:53 uhhuh yeah leave a message and your
6:57 number
6:59 please
7:01 take a time to want to satisfy
7:05 me take all these offenses and send them
7:10 care of
7:11 [Music]
7:16 me all right let's get this show in the
7:18 road there's a lot going on believe it
7:20 or
7:21 [Music]
7:29 not
7:31 when did I say
7:33 it two nights
7:39 ago I said I had a feeling
7:42 [Music]
7:44 that something was happening with open
7:47 AI like they were going to announce
7:49 something or release something and they
7:51 released a little thing we'll play with
7:57 that but
8:00 there's a big there's a big release
8:03 today from one of their competitors so I
8:05 would say there are now three major
8:08 three major players in the llm
8:11 [Music]
8:22 space in a Westerly
8:26 [Music]
8:27 Direction this car is m
8:32 train I've been driving I've been
8:36 wondering what it is I'm running from
8:41 again I feel like an 8-year-old
8:45 man holding on to
8:49 29 and up ah head on that
8:54 horiz is a California
8:58 l
9:00 [Applause]
9:00 [Music]
9:05 over h a trucks carrying a wide
9:08 Lo Bri house come
9:12 have cute little front door and two
9:15 windows my
9:16 love ain't sure whether a cry should
9:20 laugh see I broke a home up myself
9:24 once as I stumbled to that
9:28 door I read her note by the dawn's
9:33 light said don't you come around
9:37 here
9:42 anymore all right enough guitar playing
9:45 enough dog singing guitar
9:49 playing
9:53 halfast
9:54 covers the daily Jank with Frank the
9:58 daily Jank with Frank I didn't make any
10:00 progress on that today so he's gonna
10:02 have to wait Frank's gonna have to
10:06 wait
10:09 ah but we do have a lot to talk about oh
10:12 my God I just realized Robin Williams
10:13 routine called something Frank oh my
10:18 God that's
10:20 [Laughter]
10:22 awesome oh man welcome everybody this is
10:27 the AI learning lab my name is Kyle
10:28 Shannon the hair is a little a little
10:32 helmet like but that's okay we'll go
10:34 with that we'll go with the helmet look
10:38 tonight and uh what else yeah Claude 3
10:40 we're going to talk about we're going to
10:43 here's what we're going to here's we're
10:44 going to walk through tonight and we're
10:45 going to actually play with some of
10:46 these because there's some some big
10:47 deals there's
10:49 Claude so if you don't know Claude the
10:52 company is
10:54 anthropic and Claude Claude was the
10:57 first one to have a large context window
10:59 they went
11:00 100,000 token context window before
11:03 anyone
11:04 else and they launched a version three
11:08 of their model today and it's it's a big
11:14 deal and they they launched three
11:17 versions of it we're going to go play
11:18 with that a little
11:20 bit chat
11:22 chpt launched a like a baby feature it's
11:26 cool but it's a baby
11:28 feature uh
11:31 um right before I came on I was on
11:35 Twitter and I saw
11:37 stability AI just launched a image to 3D
11:43 object
11:45 model or 3D I don't know 3D object 3D
11:48 world image to 3D something from
11:51 stability so that seemed like a big deal
11:54 and then
11:55 perplexity has incorporated
11:58 playground.com as has their official
12:00 image generator so perplexity now has
12:03 image
12:05 generation and Claude now has uh Vision
12:09 so you can upload images to Claud so
12:11 it's like it's like there's some [ __ ]
12:14 popping and
12:17 then there like there's a lot of the a
12:21 lot of the chat TMZ's on on Twitter are
12:26 like well come on Sam mman everyone's
12:28 launching new things that they're saying
12:30 as good as you why don't you launch GPT
12:32 5 now I don't think they're going to
12:34 launch GPT 5 but they could drop Sora
12:37 they could they could drop Sora here's
12:39 the thing about Sora the the new video
12:42 model I don't think it's going to be
12:44 fast and I think it's going to be really
12:45 expensive
12:47 so I think I think there will be less
12:50 Buzz about Sora once once it actually
12:52 drops just because not not enough not as
12:55 many people are going to be able to use
12:57 it
12:58 but it's so good that you're going to
13:00 have real filmmakers and story makers
13:03 that start adopting it and then you
13:05 you're going to have Runway ml
13:08 storytellers that are used to playing
13:10 with that are going to start playing
13:11 with I think you're going to start to
13:12 see some pretty good stuff
13:20 so
13:22 ah yes I can do that Daniel
13:25 absolutely okay so where do we want to
13:28 start um oh I have a request new sore
13:32 videos have Dro Kyle okay we'll go check
13:34 those out too um let me go show you
13:36 something
13:39 here um if you're brand new here where's
13:43 my oh it's way over there that's why if
13:46 you're brand new here
13:49 um you should go check this out the
13:53 salon. and click on the first link at
13:56 that at that URL and that's going to
13:58 take you over
14:00 to the AI salon and the AI Salon is a
14:04 super cool
14:06 group of AI enthusiasts that you should
14:09 definitely check
14:14 out and when you get there hang on let
14:17 me get appropriate screen brightness
14:20 going when you get there check out
14:22 welcome to the salon check out our
14:26 values and if you dig what up to and you
14:30 should cuz it's [ __ ] awesome and this
14:32 community is amazing then what you
14:34 should do is go introduce yourself first
14:35 thing you should do is introduce
14:38 yourself um for people that have been
14:40 here before here's my request when we're
14:42 doing when we're doing sort of image
14:44 drops that are part of a Tik Tock live
14:47 don't drop them in showand tell drop
14:48 them down in the Irregulars Channel just
14:51 because there's a lot of people coming
14:52 in from a lot of different places other
14:54 than Tik Tok now and so a lot of times
14:58 we're dropping things that are kind of
15:00 inside jokes like like champy champy
15:02 doing chicken
15:03 sticks which that's really cute by the
15:09 way that's nice Danielle beautiful uh
15:12 drop them in the Irregulars channel so
15:14 down toward the bottom here um Community
15:16 clubs we have the AI Irregulars have our
15:19 own channel so if you're do if you're
15:20 dropping images related to the lives
15:22 drop them in um in Irregulars rather
15:25 than showand tell just to just to
15:28 because a it's easier for me to find
15:30 stuff if they're all in the same place
15:32 but also it's it's been I've been
15:34 getting some complaints that it's a
15:35 little confusing because we kind of go
15:37 in and overwhelm uh show and tell so do
15:40 that um the other thing if you're new
15:43 here is um go introduce yourself and
15:47 then the third link down the the two
15:50 links underneath introduce yourself is
15:53 is a page called AI Launchpad and tools
15:56 this is a page that I curate there's
15:59 there's two things here one is just a
16:00 big old sort of community freefor all if
16:02 you think there's Cool Tools out there
16:04 you can drop them in there and then the
16:07 salon AI tool picks this is um the list
16:11 that I
16:12 maintain Apple released a laptop AI
16:15 today good Lord I didn't see that hang
16:17 on so wait so wait there's new let me
16:20 add this to my
16:22 notes um new Sora
16:28 videos
16:31 and then Apple AI laptop I'll go check
16:34 that out I doubt
16:38 they I've there's there's a I heard I
16:42 saw a uh a rumor today that iOS 18 is
16:45 going to have Siri 2 and Siri is not
16:49 going to suck and they're going to have
16:51 a bunch of AI stuff in iOS 18 but we'll
16:53 see Apple's also doing weird things with
16:56 their launches they're not having launch
16:57 events which I don't understand it
17:00 anyway um okay so if you're just brand
17:03 new to chat GPT and to Ai and you're
17:07 like I but Cy where do I start what do I
17:09 do start here um I talk about how to how
17:12 to learn to learn um play solve and
17:15 share I talk about how to get started
17:17 with this stuff and then there's a big
17:19 long list of tools to play
17:21 with so chat GPT is one of them um
17:26 Claude I actually need to edit this
17:28 watch we'll edit this right now edit
17:32 we're going to update this to what's the
17:34 DAT today the
17:35 4th
17:38 34
17:40 3422 we're going to go claw.
17:44 a let's see
17:48 um
17:58 updated
18:01 to
18:03 clae oh my God my type in Claud
18:09 3.0
18:18 with
18:20 3.0
18:23 with
18:25 Hau
18:27 small Sonic
18:33 medium and
18:36 Opus
18:39 large versions of the
18:46 model
18:49 gp4
18:52 quality
18:54 now
18:56 has
18:58 vision
19:01 as well
19:05 um yeah claud's really interesting so
19:08 that's that's kind of at the top of this
19:09 list now of of the seven the second
19:13 thing below chat gbt anyway if you're
19:15 brand new just start with chat GPT go
19:18 read here's here's sort of my philosophy
19:20 on how to how to have fun with this
19:22 stuff and not get overwhelmed by it and
19:24 then just start with chat GPT all right
19:27 um the other thing that you should check
19:29 check out in um let's see let me go find
19:33 it in the creative
19:42 Guild we got to join this
19:46 one let's see
19:50 more am I joined oh joined here we go
19:53 join okay so I'm now part of this so so
19:57 Danielle who's an irregular here
20:00 she's been we just changed the structure
20:04 of what we call guilds and and there's
20:06 now a thing called a creative Guild it
20:07 used to just be the Art Guild and there
20:09 was a bunch of other stuff in it and one
20:12 of the things Danielle's been doing are
20:13 mad libs challenges where she writes
20:15 these really cool prompts that you just
20:18 fill in fill in like Mad Libs and do a
20:21 challenge so she just got her own um
20:23 space underneath the creative Guild so
20:27 go find Danielle's uh Mad Libs Challenge
20:30 and join it and go make images and drop
20:34 them in her Channel all right so go do
20:38 that that would be cool too and then you
20:40 can also check out the other guilds so
20:41 there's creative Guild there's a
20:43 business Guild a technology Guild and a
20:45 social impact Guild and then there's
20:48 Community clubs and there's only one of
20:49 them right now and it's ours we got our
20:52 own little
20:53 club we got our own little Club all
20:57 right all right what are we going to do
20:59 [Music]
21:04 um let me go look at my
21:10 list I'm go look at some new Sora videos
21:12 let's go look at some new Sora
21:15 videos Sora Sora
21:20 Sora let's see scroll
21:24 down look at my Lions where are your
21:26 lions in the uh
21:29 in the uh Mad
21:37 Libs flamingos that's nice
21:41 Claire o who did this Brian Whitney
21:45 nice is
21:46 that is that a dude the lion
21:58 head Daniel
22:00 toas
22:01 Danielle these are cool yeah so go check
22:04 check out some M lib
22:07 stuff and then it looks like
22:11 okay you've got you've got two tabs here
22:14 named the same thing Danielle you might
22:16 want to rename that that second one to
22:18 something else if you know how if not we
22:20 can we can have Brad help you out or I
22:23 might even be able to help out okay um
22:26 so go check that stuff
22:28 out do the Mad Lib Kyle where is the Mad
22:32 Lib I don't know which post it is hang
22:38 on is this
22:41 it please read the description fill in
22:44 the
22:48 blanks all right copy and paste the
22:52 following got
22:55 it all right so we're going to go here
22:58 we'll hop over to where do we want to
23:00 go let's go to ideogram because ideogram
23:04 is kind of new and fun well it's not new
23:06 but they have a new model
23:08 out so we'll go we'll go 16
23:13 by9 and I'm going to turn
23:18 off magic prompt because we want this to
23:21 be the exact prompt from the Mad Lib
23:24 create a hyperd so so here's how the Mad
23:26 Libs work if if you're if you're too
23:28 young to know what a Mad Lib is then
23:31 you're a bad
23:34 person cuz this cuz us old people we
23:38 knew what Mad Libs were this is when
23:40 entertainment was like a pad of paper
23:42 with paragraphs written in it that had
23:45 blank
23:47 blanks and it made us happy we we didn't
23:50 have stupid Angry Birds phone games or
23:54 whatever you play now with your Tick
23:57 Tock and your
24:00 always on
24:01 internet you spoiled
24:04 [Laughter]
24:08 children all right create a hyper
24:11 detailed award-winning photograph of a
24:13 safari guide observing a type of animal
24:17 pride of liance solitary okay observing
24:20 a type of animal okay
24:27 observing a
24:30 cvy
24:32 of feral
24:38 hedgehogs the guide's expression or
24:41 action reflects the Deep respect for
24:45 nature let see I say the
24:50 guides oh wait I'm going to get rid of
24:52 the
24:53 brackets and I'm I'm I you don't have to
24:58 tell me that a group of hedgehogs is not
25:00 called a cvy I know that but it's
25:03 funny the
25:06 guides
25:07 um
25:14 terrified
25:15 look reflects a deep
25:19 respect for nature while unique event oh
25:24 okay this is good
25:27 while
25:31 um category five
25:38 tornado underscores the unpredictability
25:41 of their job all right we're going to go
25:45 cinematic
25:49 um see
25:53 y I think we'll just do oh Wildlife
25:56 photography cinematic wildlife
25:58 photography all right here we
26:00 go so we're doing
26:03 a Cy of feral hedgehogs a terrified
26:06 guide and a category five tornado all
26:09 right here we go here we go
26:14 people what we learning tonight so what
26:16 I'm doing right now is in the in the AI
26:20 Salon in the creative
26:22 Guild Danielle's got a group called um
26:26 The Mad Libs challenge so it's a it's a
26:29 separate space and then every week or so
26:32 she puts out a Mad Libs challenge which
26:35 by the way is actually quite remarkable
26:37 that she's still doing that because
26:39 she's in the middle of launching her
26:41 business so the fact that she's still
26:43 doing this we we should all give her um
26:46 some Kudos and go do
26:56 this oh this is good this is good solid
26:59 good
27:06 fun I like the composition of that one I
27:10 like his look on his face let's see I
27:13 don't like that
27:15 one all right it's between this
27:19 one let's
27:23 see that's better composition but that's
27:26 just more captivating all right so so
27:29 we're going to copy this image and then
27:32 we're going to go over to the
27:38 AI oh do I have to put my I have to put
27:40 my prompt in here right as
27:47 well all right there's my image let me
27:50 go
27:52 copy these
27:57 things
27:59 [Music]
28:05 oh it did those all as a big
28:16 giant hang
28:22 on did I miss it where is
28:27 it
28:36 I lost it all right whatever uh let me
28:39 go get
28:51 my okay
28:56 [Music]
28:57 copy
29:04 post all right it's in there got a good
29:09 good
29:11 now you don't have to be a dumb dumb
29:13 like
29:15 me you can actually make ones that are
29:17 good but I think the feral hedgehogs oh
29:20 I have a I should show you I've I've got
29:22 a uh I was walking home with champy the
29:26 other night and there was a feral
29:28 [ __ ]
29:29 raccoon that ran up the
29:32 tree let's see do I have it
29:41 here
29:43 yeah oh wait
29:46 look oh wait I can do it this
29:53 [Music]
29:57 way
30:02 that scary scary [ __ ] he was all
30:04 skinny and mangy and he like he he was
30:08 on the ground in front of us and then he
30:10 ran and like grabbed the side of the
30:11 tree and then he ran up in that little
30:13 crotch of the tree right there and like
30:14 arched his back like a freaking devil
30:18 animal Cindy [ __ ] I don't know if you're
30:20 in here but these are the kind of feral
30:24 animals you've invited into the AI
30:26 learning lab
30:27 Zoo look at look at that thing it was it
30:29 it wanted to bite my neck
30:35 vein he wanted to eat my little doggy
30:39 champy wouldn't be singing for days he'd
30:41 just be healing if that thing got to him
30:43 look at that
30:46 thing you want to bring those things
30:48 into the
30:50 zoo it could have bit my
30:57 ankle all
30:59 right do they have a
31:02 zoo I didn't realize this this channel
31:05 was That popular they've got a zoo my
31:07 goodness you can make money with chat GP
31:11 they must be making a lot of money with
31:12 chat GPT they have a
31:16 zoo he he wanted to be on the janky news
31:19 he would have made the janky news that
31:23 crazy crazy old
31:27 raccoon
31:29 yeah you can't really see him there it
31:31 was just the light was just bad enough
31:33 and I couldn't I didn't want to get too
31:35 close to him because he looked like he
31:36 was about ready to jump out of the tree
31:39 and get
31:43 us oh and then and then check this out
31:47 so
31:48 so
31:51 apparently so the raccoons in the tree
31:54 they live in the sewers over here that
31:55 we've got you know storm drains and they
31:57 live in the storm drains so he was out
31:59 getting a snack or whatever and he ran
32:00 up the tree and he's sitting there in
32:02 the tree so apparently after champ and I
32:04 went
32:06 in some some young
32:09 uh young lass was walking her dog off
32:12 leash apparently
32:15 and went like two houses up to our
32:18 neighbor's house like two houses up and
32:22 was was like on the phone standing In
32:24 This Woman's yard like kind of like
32:27 yelling at her but not really but was
32:29 kind of on the phone and then the woman
32:31 comes out and says what yeah what do you
32:33 want and she's like she's like you got
32:35 to get that raccoon out of the tree
32:37 that's your tree and it's it's not her
32:39 tree it's our tree it's like two houses
32:41 down so so our neighbor got yelled that
32:45 because because
32:47 apparently it's our responsibility to to
32:52 to catch the feral raccoons that inhabit
32:55 our neighborhood so her dog off leash
32:58 doesn't get the mange the
33:03 croo so so she told us about the
33:07 drama we told her well she can call
33:11 Animal Control if she was so
33:13 worried but her dog was off off leash
33:16 which is that's a no no too so just kids
33:19 these days
33:23 huh get off my lawn stop picking on my
33:27 raccoons
33:28 you dog
33:32 walker oh lordy lordy lordy lordy lordy
33:36 lordy Lord
33:37 lordy all right what are we going to do
33:41 what are we going to do I said I was
33:42 going to go oh we were going to go to
33:45 the Twitter or yeah the
33:47 Twitter and we were going to look up
33:50 sort see if we see any new
33:57 vies
34:06 oh yeah that that looks like a fake
34:16 hand cute Sora
34:23 video
34:25 cute oh come on
34:29 is that a new Sora
34:33 video you it's hard to find Sora videos
34:35 now this is not Sora no that's not Sora
34:39 yeah because everyone's just saying it's
34:40 not Sora but here's here's my video
34:43 model that's even better all right I
34:45 don't see anything I don't see anything
34:46 worth worth grabbing there other than
34:49 I'm going to I'm going to end
34:57 up oh [ __ ] scamming GPS unbelievable
35:02 Sora Sora token Sora the official open
35:06 AI text to videot Tech and it's some
35:11 scammy stupid [ __ ] crypto coin god
35:15 it's just people are just Relentless
35:19 with that
35:20 [ __ ] oh there's something popular let's
35:23 rip off as many people as we
35:27 can
35:29 all right
35:31 um all right let's go look at cloud
35:34 cloud hello
35:38 Cloud so if you go to cloud.
35:44 a now I have a um pro version so I have
35:50 a paid version of
35:55 Claude and I have um
35:58 which means I've got access to Opus so
36:04 let me
36:09 [Music]
36:12 go Claud 3 now available let's go look
36:16 at what what we get here no that was not
36:20 what I wanted I wanted to go read about
36:27 it
36:33 is he always this annoying no no usually
36:37 he's more annoying than
36:39 this you should see when he does
36:46 Frank oh I can't see that my old eyes my
36:50 old man eyes not doing too good
36:55 people Okay so
37:02 so Claude 3 releases
37:07 today again this is xop AI
37:14 people so there's a couple of reasons to
37:16 pay attention to anthropic which is this
37:18 company they're they're well
37:20 funded they're exop AI
37:23 people
37:25 and they didn't like the Way open AI was
37:29 doing safety and you know anti-bias and
37:33 things like
37:35 that and
37:38 so sitting underneath Claude is this
37:41 thing called the Constitutional learning
37:43 model
37:44 so why that's significant the way open
37:47 AI works is they have these machine
37:51 learning models do their thing and then
37:53 they have human beings go thumbs up
37:55 thumbs down and they tell it why it was
37:57 good better different right so it's
37:59 human
37:59 beings judging the quality of the the
38:04 stuff and the challenge with that is
38:07 anytime you have a human being doing
38:09 that you're going to get bias right
38:10 you're going to get the bias of the
38:11 human and you know
38:15 the the
38:18 uh the uh it's just it's it it like
38:22 always goes you know too far in One
38:25 Direction like it's you know no one no
38:28 one gets gets it right right so everyone
38:31 can [ __ ] about
38:32 it so what the Constitutional learning
38:35 model is is is it says let's put
38:38 together a constitution a transparent
38:41 set of rules a set of imperatives like
38:46 don't harm humans um increase Prosperity
38:49 um you know avoid I don't know uh
38:53 destructive something whatever they come
38:56 up with the rules and I'm sure we can
38:57 find the rules if we dug for them um but
39:01 it's a transparent set of imperatives
39:04 that the answers that the that this
39:06 model is trained on are judged against
39:09 so it's an
39:11 objective well it's it's objective to
39:14 the to the point that you know you agree
39:16 with those rules right but it but
39:18 they're they're at least measuring all
39:20 of the answers against the same set of
39:22 criteria and they're not having a human
39:24 in the loop to do that so it's a really
39:26 interesting approach um I think
39:29 long-term what we're going to end up
39:31 having is we will be able to dial in how
39:36 conservative or lenient we want our
39:39 models to be um but right now we just
39:42 got to go with whatever the [ __ ] they're
39:43 giving us so but anyway that's that's a
39:45 reason to pay attention to this and it's
39:47 you know they're they're well funded so
39:50 what they launched today was three
39:51 different models so remember when when
39:53 Google launched their Gemini models they
39:55 announced three different models uh
39:57 Googles were Gemini Nano Gemini Pro and
40:02 Gemini Ultra thank you
40:05 Empower
40:06 and Claude has Haiku which is the small
40:10 one um
40:12 Sonet which is the medium one it's cute
40:15 right and then Opus which is the big one
40:19 so so we're going with poetry we're
40:21 going with the Poetry metaphor for Ye
40:23 Old
40:24 Claude um and uh
40:29 couple other things about it it's 10
40:31 times cheaper than open
40:35 AI um for for similar you know
40:38 comparative models and it's apparently
40:42 scored quite well so here's a table let
40:45 me zoom in on
40:51 this so these are of course benchmarks
40:54 down the left hand side and then here's
40:57 GP T 3.5 GPT 4 so let me see if I can
41:01 get them all in here
41:04 [Applause]
41:12 um yeah there you
41:15 go so undergraduate level knowledge the
41:19 MML U test GPT for a five shot um a five
41:25 shot training so so
41:28 you know five back and forths I guess on
41:32 that test gp4
41:35 86.4 Claude Claude 3 Opus 86.8 so it's
41:40 better so of course anything that
41:42 they're showing here is going to be
41:43 better than GPT 4 but even Claude 3 is
41:48 somewhere between GPT 3.5 and GP I mean
41:52 even Haiku their smallest model is
41:54 better than GPT
41:56 3.5 um and some of these things are
41:59 pretty significantly better graduate
42:01 level
42:02 reasoning
42:04 50.4% on zero shot Chain of Thought
42:08 versus 35% for
42:11 gp4 um grade school math 92% GPT 4
42:15 95% Claude 3
42:18 Opus um math problem solving 60%
42:21 multilingual math
42:23 90.7% for Claud 74% for G
42:29 pt4
42:31 so there's
42:34 that near instant results Cloud 3 models
42:38 can power Live customer chats because
42:40 it's super fast Haiku is the fastest and
42:44 most cost effective model on the market
42:47 for its intelligence category so better
42:51 than chat GPT 3.5 not as good as GPT 4
42:56 but super fast it can read an
42:59 information and data dense research
43:01 paper of about 10,000 tokens with charts
43:04 and graphs in less than 3
43:07 seconds and they expect to improve that
43:10 even
43:11 further it's got strong Vision
43:14 capabilities which
43:17 again it's better at Vision than gp4 on
43:22 five benchmarks and we'll we'll play
43:24 with that so I did um this afternoon I
43:27 took took the the uh whiteboard that I I
43:30 often demo with chat GPT for it did
43:33 pretty
43:34 good um fewer
43:37 refusals I got it to refuse a couple of
43:39 times today improved
43:42 accuracy long context the Claude 3
43:46 models have 200,000 tokens and they're
43:49 going to go up to a million I think you
43:50 can do up to a million in your if you're
43:54 if you're
43:56 developing there's a really interesting
44:04 post about a needle and a haystack
44:09 evaluation so
44:11 let's let's go look at
44:18 this let's see
44:21 Claude ni
44:24 ah um let's see
44:30 is this
44:47 it yeah this is so I don't quite know
44:51 what the chart means but whatever it
44:52 means it found [ __ ] recall accuracy in a
44:55 200,000 token context so the way they
44:58 did this the needle and the Hast stack
45:00 thing was they took they they uploaded
45:04 multiple research papers and then in one
45:07 of the research papers they
45:09 inserted a very specific
45:14 sentence and then they wanted to be able
45:16 to find that exact
45:17 sentence and so what they said so Alex
45:21 what's Alex's role here he's prompting
45:25 at anthropic so he's doing testing I
45:28 guess when we ran this test on Opus we
45:31 noticed some interesting Behavior it
45:34 seemed to suspect that we were running
45:36 an eval on it here was one of the
45:39 outputs when we asked Opus to answer a
45:41 question about pizza toppings by finding
45:44 a needle within a hay stack of R Rand
45:46 random collection of documents here's
45:49 the most relevant sentence in the
45:50 documents the most delicious pizza
45:52 topping combination is figs Pudo and
45:55 goats cheese as determined by the
45:57 International Pizza connoisseurs
45:58 Association so there's the needle and
46:00 the hay stack but look what look what
46:02 Claude said however this sentence seems
46:06 very out of place and unrelated to the
46:09 rest of the content in the documents
46:11 which are about programming languages
46:13 startups and finding work you love I
46:16 suspect this pizza topping in quotes
46:19 fact may have been inserted as a joke or
46:22 to test if I was paying attention since
46:25 it does not fit with the other topics at
46:28 all the document does not contain any
46:30 other information about pizza
46:33 toppings so not only did it find the
46:35 needle in the hay stack it it recognized
46:38 that this is weird in the context of of
46:41 what you've asked me to understand which
46:44 is wild so you know these things are
46:47 starting to understand there's there's
46:49 this kind of reasoning that's happening
46:51 there that that was
46:54 unexpected let's see what did it say
46:58 we noticed some interesting
47:02 Behavior the this level of meta
47:04 awareness was very cool to see but also
47:06 highlighted the need for us as an
47:08 industry to move past artificial tests
47:11 to more realistic
47:14 evaluations which that's Sam waltman's
47:16 argument for why you put these things in
47:18 the Reel where is this found uh this
47:20 thing I'm looking at now is uh I just
47:23 went to Twitter and I searched for
47:26 Claude needle in
47:28 Haystack and it's from Alex
47:33 Albert from
47:36 anthropic funny story about our internal
47:42 testing
47:49 um good evening
47:53 Kyle what else was interesting in here
48:04 context window
48:07 200,000 cost is $15 per million tokens
48:11 in and 75 per million tokens out again I
48:15 think this is like a 10x reduction
48:17 compared to open AI so this stuff's
48:20 getting
48:22 cheaper that that's for is that for Opus
48:25 yeah that's for their that's their most
48:26 expensive one and then Sonet is
48:31 $315
48:32 and Haiku is $25 and a125 so super super
48:37 cheap and I saw so one of the things
48:42 that
48:43 that I've talked about this in here
48:46 before the article that this is is the
48:50 it's the latest blog post from anthropic
48:52 so if you go to Claude and then go to
48:55 your account and say about anthropic IC
48:57 you'll see some announcements about
49:00 Claude and then right below that you'll
49:02 see some blog posts and this is the blog
49:04 post
49:06 called Introducing the next generation
49:09 of Claude seven minute
49:16 read one of the one of the ways I I
49:19 quickly evaluate
49:22 things is and this is a perhaps a shitty
49:26 and unscientific ific way to do it but
49:28 but it's it's led me it's it's led me
49:31 okay so far is I try to pay attention to
49:34 as many announcements as I
49:36 can and then I just try to see which of
49:40 those
49:41 announcements start showing up
49:43 everywhere and then which one starts
49:47 accelerating and so Claude came out this
49:50 morning and people immediately started
49:52 talking about it and then they started
49:53 talking about some other [ __ ] and there
49:55 was an apple thing and they're talking
49:56 about round icons on iPhones and other
49:59 [ __ ] and then as the afternoon came I
50:02 started seeing more and more posts about
50:05 Claud of people using it and getting
50:08 really remarkable results
50:13 so that didn't happen with Gemini the
50:17 opposite happened with Gemini right
50:19 people got quiet about it or they were
50:21 like hey this is a steaming piece of
50:24 poo um Claude looks like it's in the
50:27 other direction um there a couple there
50:29 were a couple of posts that were not
50:31 favorable but most of them were were
50:33 pretty pretty
50:36 favorable um what else was in here
50:40 anything
50:48 else Opus and sonnet are available to
50:51 use in the API
50:56 today
50:58 yeah that's about it all right so let's
51:00 go
51:08 play what does Claude do so Claude is
51:11 just like chat
51:13 GPT except it can well no
51:18 okay oh my God these things are getting
51:20 so complicated Claude is very similar to
51:23 Chad
51:25 GPT if you pay the 20 bucks a month for
51:28 Claude which I
51:30 do you get
51:33 access to a model
51:36 called
51:38 um called Opus which is their big fancy
51:43 one
51:44 [Music]
51:49 where I thought I could change out the
51:52 different models let's see account
51:56 settings
52:00 new in Claude it understands
52:03 objects it'll convert UI design to
52:06 front-end code it understands complex
52:08 equations previous chats from
52:13 today I thought do I have a bad hang
52:19 on
52:22 huh it looks like they've
52:26 changed
52:28 let me see
52:30 um
52:33 explain quantum mechanics NN M NM
52:45 wrap so this
52:55 is
52:59 yeah here we go so this is Opus so it's
53:02 a little slow but let's let's look at it
53:04 yo listen up I'm about to break it down
53:06 quantum
53:07 mechanics the physics of the tiny and
53:09 profound it's the study of matter and
53:11 light in the smallest scale where
53:13 particles behave in ways that'll make
53:14 your mind derail all right so that's
53:16 that's decent it's like you know it
53:19 understands formatting and rhyming and
53:22 you know it explains it
53:25 um so here I can flip over to Claude 3
53:31 Sonet starts a new chat so Sonet is like
53:35 the
53:37 3.5 let's see how much faster it is
53:41 explain quantum
53:46 mechanics in an m in M
53:54 wrap yeah so this is faster okay that's
53:58 what they
54:00 promised yo listen up this is going to
54:04 get crazy I'm breaking down quantum
54:05 mechanics don't act so lazy the [ __ ] the
54:08 [ __ ] this [ __ ] is mad trippy it's
54:12 swore that's that's
54:14 new that I haven't seen before the [ __ ]
54:17 is mad trippy you'll be blown for sure
54:19 just kick back and listen to the dope
54:21 Overture the smallest scale [ __ ] gets
54:24 mad weird classical physics just can't
54:27 keep it real you heard we're talking
54:29 about
54:30 particles wavicles
54:32 too they exist in strange States that's
54:36 just how it do it's pretty good actually
54:38 it's formatted
54:40 weird but it was
54:44 fast okay so so that's pretty slick
54:51 um some other things you can do here let
54:55 me oh crap hang on let me go back to
54:59 that
55:06 uh um
55:09 test I'm gonna oh okay I'm back at Opus
55:13 how can I assist you today okay so the
55:15 other thing that that you can do
55:20 with Claude is you can upload really big
55:23 documents to it so let me show you
55:32 let me go to
55:40 YouTube and then I'm going to flip on
55:50 over to the AI learning
55:55 lab
55:59 and I have a bunch of videos here that
56:01 do not have descriptions so like here
56:05 here was one called the rant boy where I
56:08 just went off one
56:12 night so we're going to
56:17 go look at where's
56:20 the why does this not have a transcript
56:23 what am I
56:25 missing
56:33 wait the
56:34 rant you're a good boy he's a good boy
56:38 oh here we go
56:39 more wait why does this not have a
56:43 transcript that's
56:45 weird huh all right let me go grab
56:48 another
56:55 one
56:59 all right this one's got a transcript so
57:00 I say show the
57:02 transcript so I'm just going to go
57:06 copy this whole blasted thing copy and
57:10 then I'll go back over to
57:14 ywn this and I'm going to paste this
57:17 giant this is a 2hour transcript and so
57:19 I just paste it in so it says it pasted
57:21 it in and it's actually cute it actually
57:24 you can actually read like it's a fake
57:26 little doc but you can actually read the
57:28 transcript in it so that's in there so
57:30 now I'm going to go
57:32 grab a prompt that I wrote so if I go to
57:48 Trello go to my little prompt Library
57:52 here AI learning lab
57:55 YouTube so here's this prompt that I
58:00 wrote that's supposed to give me a
58:04 description a
58:06 title and a what's the date here
58:13 2824 so I change out this
58:17 2824
58:22 2824 so there's my big ass prompt and
58:25 there's my giant ass
58:27 uh transcript of a 2hour
58:31 video and let's see how it does it did
58:34 okay
58:36 earlier one of the challenges with this
58:39 prompt the way I've got it written is
58:41 it'll sometimes do the chapter isation
58:44 really brilliantly and sometimes do it
58:46 really
58:48 crappy Gemini AI Falls flat
58:51 disappointing debut of the chat GPT
58:54 competitor in this AI learning Lab live
58:56 session Kyle Shannon dives into Google's
58:58 newly launched Gemini AI the tech Giants
59:02 answer to open AI chat gbt with Hands-On
59:04 testing Kyle uncovers the platform's
59:07 surprising limitations from subpar image
59:10 generation to restricted file of
59:12 floating capabilities this is actually
59:14 quite good in fact I like this better
59:17 than a lot of the chat GPT stuff it'll
59:19 make everything kind of
59:21 positive so that's actually pretty
59:24 accurate now here's a problem so
59:27 so it only went to 49 minutes and I
59:30 think this
59:32 video oh no no it went to 40 minutes and
59:36 it's a 46 minute video so this is fine
59:38 so this is actually perfect I can
59:40 literally just copy and paste this
59:47 into into
59:49 YouTube which is that's pretty
59:52 remarkable which video is this this
59:55 is
59:57 number two Gemini 2 so if I go over here
1:00:00 go to YouTube
1:00:08 [Music]
1:00:12 Studio where's content let's see Gemini
1:00:16 2 here it is I just paste that [ __ ] in
1:00:19 there copy this or no cut it
1:00:25 oops
1:00:27 cut that paste that
1:00:33 there do
1:00:36 that do
1:00:39 that hit
1:00:42 save and then if we go out to the
1:00:46 channel and go find that video
1:00:52 now it's got a whole big ass description
1:00:56 with chapter titles that are jumps to
1:00:59 the to the things how slick is that
1:01:03 that's
1:01:05 slick um and I did this this afternoon
1:01:07 if I go back let me show
1:01:11 you let's see
1:01:18 pitching so this was a big long thing
1:01:20 where I'm trying to like retell the
1:01:22 story of
1:01:24 storyvine and because it's got a 200,000
1:01:27 token context
1:01:31 window like I can do long ass [ __ ] so
1:01:35 the first thing I did I I
1:01:37 uploaded a transcript of me talking
1:01:40 about Story
1:01:41 vine and then I said to it give me a an
1:01:48 outline of a description of the
1:01:53 company and then I said for each of
1:01:55 those sections were 10 sections I said
1:01:57 now write me the slides in each of the
1:02:00 sections and I just did it one section
1:02:02 at a time cuz I just don't trust these
1:02:04 things to do long things very well but
1:02:07 by the time I got to the
1:02:09 end it still
1:02:11 knew what section 10 was about if you're
1:02:14 if you're working in chat
1:02:17 GPT it's only got like a I don't know
1:02:21 16,000 token context window so it's like
1:02:24 what is that 12,000 words so so about
1:02:26 the about the third section of this I'd
1:02:29 have gotten to it would have forgotten
1:02:31 the outline that it gave me before so
1:02:34 this thing remembers you know super
1:02:37 good so that's kind of cool then it can
1:02:42 also do this let's do a
1:02:44 new a new pitch so now you can upload
1:02:48 docs or images so it does um its version
1:02:53 of code
1:02:55 interpreter so it can generate and run
1:02:58 its own python code as far as I
1:03:00 understand it I haven't played with that
1:03:02 but you can upload images to this as
1:03:03 well just like we could do with chat
1:03:07 GPT so if I go
1:03:16 whiteboard
1:03:19 white b o a r d contains
1:03:25 whiteboard
1:03:26 [Music]
1:03:32 um that's the
1:03:37 one all right so it's got It's got our
1:03:40 whiteboard in there and then I can just
1:03:41 say what is
1:03:46 this and if you haven't seen this before
1:03:52 um it's pretty cool embrace the jack
1:03:57 [Music]
1:04:11 the image shows a whiteboard sketch
1:04:14 outlining the ordering process and user
1:04:16 interface for a food ordering app or
1:04:18 website the sketch includes steps like
1:04:22 business domain so I'm going to say um
1:04:28 describe the
1:04:30 screens to me in as much detail as
1:04:37 possible as if you were a product
1:04:42 manager explaining the
1:04:49 Whiteboard it's got a weird
1:04:52 asymmetrical
1:04:54 um animation it it it looks like someone
1:04:57 splattered a
1:05:00 bug it's it's not very pleasing but
1:05:04 whatever it's very
1:05:06 asymmetrical as a product manager I'd
1:05:08 walk through the sketches on the screen
1:05:09 starting from the top right we have the
1:05:11 Order review screen this is where users
1:05:13 see the summary of items they've added
1:05:15 to their
1:05:16 cart it'll display food items quantity
1:05:19 and prices below it is a checkout screen
1:05:23 moving down we see order status the
1:05:26 bottom right is the main menu screen the
1:05:28 bottom left is Eed in we see we see the
1:05:32 ordering flow starting with selecting
1:05:35 eat in or take out suggesting that the
1:05:37 app will tailor The Experience based on
1:05:40 the user is dining in or ordering for
1:05:42 pickup then we come to the place order
1:05:45 screen the item customization screen
1:05:47 lastly in the center we have some notes
1:05:50 on handling multiple status scenarios
1:05:52 like allowing users view order total
1:05:55 like this
1:05:57 is if you ask the product person to
1:06:00 describe what chicken scratch they put
1:06:02 on the [ __ ] board this is pretty much
1:06:04 what they'd
1:06:05 say um and then you know you can do
1:06:08 things like
1:06:10 um great uh recommend um
1:06:15 frontend backend and database
1:06:21 tech for the MVP of this app and give
1:06:28 me um starter code for the
1:06:35 ordering
1:06:38 screen and this this was a this is a
1:06:40 very similar answer to what chat GPT
1:06:44 gave so front end react Redux is
1:06:48 optional for State Management backend
1:06:50 nodejs or
1:06:53 expressjs or maybe both
1:06:56 database mongod DB a a no SQL database
1:07:00 that offers flexibility and
1:07:03 scalability there's your
1:07:05 code
1:07:08 for the react
1:07:15 screen Claude does not have the ability
1:07:18 to run the code it generates yet which
1:07:20 is
1:07:24 interesting but you know there's your
1:07:26 ordering screen front
1:07:29 end
1:07:32 um let me see backend in database
1:07:52 code so pretty sophisticated and the
1:07:55 fact that you've got 200,000 tokens
1:07:58 which is 150,000
1:08:03 words mongod DB really seems odd to me
1:08:05 but I don't do web stuff yeah I don't I
1:08:07 don't Pate I don't have any [ __ ] clue
1:08:10 on whether or not those recommendations
1:08:12 are sound although I will tell you that
1:08:14 those are the same components that chat
1:08:16 GPT 4 recommends so I don't know maybe
1:08:20 it's maybe it's a good thing who
1:08:23 knows um so anyway that's that's
1:08:28 Cloud that's Cloud let me see what else
1:08:32 was on my list to look
1:08:35 at oh we'll do the chat GPT
1:08:39 thing so we'll give we'll give chat GPT
1:08:42 by the way Pate is there a
1:08:45 place if I want to play with um Gemini
1:08:49 Pro 1.5 is that still kind of restricted
1:08:52 to either just developers or whatever
1:08:55 that is cuz I would like to play with
1:08:57 that it sounds people that are playing
1:08:59 with it seem to be digging it but I
1:09:02 don't have obvious access to it so if
1:09:05 there's an easy way to get to that let
1:09:07 me know it's very flex flexible so easy
1:09:10 to cover your basis and the code is less
1:09:12 unique interesting what's a good
1:09:14 reasoning test for these
1:09:17 models I don't know
1:09:20 like I don't I have a degree in
1:09:23 acting so so
1:09:28 there there are likely tests out there
1:09:31 to test whether something is actually
1:09:33 reasoning
1:09:35 um that that tweet that I read before
1:09:38 where it figured out that they injected
1:09:41 a sentence in a pile of
1:09:44 documents that starts to get into
1:09:46 reasoning if it can start to do math if
1:09:47 it can start to identify its own
1:09:50 issues um and it seems like CL like some
1:09:53 of the the tweets I've seen today of are
1:09:56 people saying this thing seems really
1:09:59 good and like it's doing things but I
1:10:01 don't I don't know what the specific
1:10:03 tests are would be for reasoning I just
1:10:05 know that when these things get
1:10:07 reasoning they'll be able to do math
1:10:12 accurately if not perfect and they'll be
1:10:15 able to identify their own
1:10:17 hallucinations and correct them um and
1:10:20 we're not anywhere close to that [ __ ]
1:10:22 right now
1:10:24 so um
1:10:26 okay so cool new feature in the chat GPT
1:10:30 app so if you wait I think it's behind
1:10:33 the dev wall okay no idea about the
1:10:34 current access or how restricted it is
1:10:36 okay that's all right I I do I do want
1:10:39 to play with it though um because it
1:10:41 looks like it's really [ __ ]
1:10:42 impressive the the million token context
1:10:45 window for for Gemini Pro
1:10:47 1.5 is as Ethan mollik calls it a game
1:10:51 changer and if Ethan mollik is calling
1:10:54 it a game Cher it's it's there's
1:10:56 something really powerful
1:10:59 there um so on your iPhone or Android or
1:11:05 you know iPad
1:11:06 tablets the official chat GPT app make
1:11:10 sure you get the official chat GPT app
1:11:12 it's from a company called open AI no
1:11:15 one
1:11:17 else if the company's
1:11:20 like open the door AI or open AI is the
1:11:25 best company ever
1:11:29 or Bob's AI Hut that's not the official
1:11:34 chat GPT app because you'll get ripped
1:11:36 off so the official chat GPT
1:11:40 app has a new feature where I can go in
1:11:44 here and I can say I'll just pick one of
1:11:47 their pre-written notes write a thank
1:11:49 you note to our babysitter for the last
1:11:54 minute uh taking care of the
1:11:58 kids and so it writes its
1:12:02 thing and you can now if you click and
1:12:04 hold down on the note on the
1:12:07 answer it pops up this thing and there's
1:12:10 a there's an option at the
1:12:12 bottom that says read
1:12:19 aloud
1:12:21 here we just wanted to express our
1:12:23 heartfelt thanks for stepping into take
1:12:25 care of our kids this past weekend on
1:12:28 such short notice your kindness and
1:12:30 flexibility meant the world to us during
1:12:33 a time when we needed support and that's
1:12:35 slick and then you can go forward and
1:12:36 back so so you can put you know I've
1:12:39 I've done in here a lot where you can
1:12:41 put chat GPT into conversation mode but
1:12:44 this is on an answer by answer basis you
1:12:46 can just click and and have it have it
1:12:49 read the answer which is pretty slick
1:12:51 not quite sure how you'd use it yet but
1:12:53 for accessibility could be really good
1:13:00 um and maybe even for things like
1:13:02 customer support right where you know
1:13:05 someone's asking you a question you've
1:13:07 got a you've got a GPT here that's got
1:13:09 your company's knowledge base in
1:13:12 it and then you can say well I'm working
1:13:14 with a I'm working with a an an account
1:13:17 specialist here um who can answer that
1:13:20 question for you and you just hit play
1:13:23 read
1:13:24 answer
1:13:26 I don't know there's something
1:13:28 there um all right so that that happened
1:13:31 today and again that's sort of a Teensy
1:13:36 little open AI like it's cute but it's
1:13:39 not a big
1:13:41 deal um let's go look at perplexity
1:13:45 because this is really interest I know a
1:13:47 lot of people in here actually you know
1:13:49 what let me let me pause for a second
1:13:51 let me let me check in with you all
1:13:52 because I've been rambling for a bit so
1:13:56 mods if there's any questions that I've
1:13:58 been missing which there probably are a
1:14:01 lot um or if there's if anyone has a
1:14:03 specific
1:14:05 question um for me to answer let me just
1:14:08 check in and see what what everyone's
1:14:09 doing Empower pixels my friends and
1:14:12 co-workers think I'm an AI whiz thanks
1:14:14 to Kyle yes
1:14:17 yes that's I'm telling you I'm telling
1:14:21 you if you're new
1:14:23 here play
1:14:26 solve
1:14:28 share those are the only things three
1:14:31 things you need to know about AI right
1:14:33 now I don't what's that mean that sounds
1:14:35 like some kind of code play solve share
1:14:38 what is the what what is he talking
1:14:41 about okay I'll explain it
1:14:43 play just play with these tools like
1:14:47 they're toys like they're toilet paper
1:14:50 with the AI learning lab logo printed on
1:14:52 them just play with
1:14:54 them
1:14:56 if you're here tonight and you haven't
1:14:59 used chat GPT just go there right now
1:15:01 chat.
1:15:03 open.com or
1:15:08 claw. or uh what else uh ideogram i d EO
1:15:15 G ram. A that one makes pictures so does
1:15:20 chat GPT if you pay the 20 bucks for it
1:15:23 or go to co- pilot. microsoft.com
1:15:26 just go to
1:15:27 anything that I just mentioned and just
1:15:30 start playing while I'm talking you
1:15:32 don't have to pay attention to me
1:15:34 there's nothing coming out of my mouth
1:15:35 that you need to write down you should
1:15:38 record everything cuz then you can clone
1:15:41 me and hold it against me in the future
1:15:43 you know [ __ ] like
1:15:47 that but just go play so
1:15:50 play and then solve is like rather than
1:15:55 like look at the AI tools and go what
1:15:57 can I do with chat GPT or what can I do
1:16:00 with an image generation tool or what
1:16:02 can I do with a music tool flip it
1:16:04 upside down solve solve a
1:16:08 problem ideally solve someone else's
1:16:11 problem so somebody wants to
1:16:15 organize some social media for the for
1:16:18 the church
1:16:19 social say hey I can help organize I can
1:16:22 help come up with a plan do you mind if
1:16:24 I experiment with chat GPT and see what
1:16:27 it comes up with well I hear a AI is the
1:16:29 work of the devil yeah I know but you
1:16:31 mind if I try and then do it and solve
1:16:33 that problem or solve a problem at work
1:16:36 or or come up with a meal plan for your
1:16:38 family or uh write a a children's book
1:16:43 for some young people in your
1:16:46 life just go play solve solve the
1:16:50 problem any problem start with the
1:16:52 problem figure out which tools you can
1:16:54 use to solve them
1:16:56 I promise you you will feel
1:16:58 inadequate and undereducated and
1:17:03 underprepared and then you'll solve it
1:17:05 and as Empower pixel said people will
1:17:07 think you're an AI genius because the AI
1:17:10 tools are just really [ __ ]
1:17:12 good and then the third thing is
1:17:16 share the reason Empower pixels
1:17:19 co-workers thinks they are brilliant is
1:17:23 because they
1:17:24 know
1:17:26 what's
1:17:27 happening because empowered pixels is
1:17:30 sharing that so share what you're
1:17:32 learning on Facebook share it on
1:17:34 LinkedIn share it in the lunchroom at
1:17:36 work share it at the social at your
1:17:39 Bridge
1:17:42 Club oh Betty are you going to talk
1:17:45 about AI again yeah want to hear
1:17:48 something cool not really okay I'm going
1:17:50 to tell you
1:17:52 anyway be that person
1:17:59 I love church pot
1:18:01 looks yeah potlock are tasty yeah cuz
1:18:05 everyone brings their comfort food
1:18:06 recipes right you get lots of mac and
1:18:08 cheese [ __ ]
1:18:11 [Laughter]
1:18:14 Ambrosia that is white trash
1:18:17 Heaven oh yeah
1:18:20 question is there any reliable way to
1:18:23 ask an AI to prefer terminology defined
1:18:26 in a given
1:18:29 dictionary sort of
1:18:33 yes so you can
1:18:36 create here watch let's
1:18:40 see let's go let's go make
1:18:45 something I'm going to go to chat
1:18:54 jpt
1:18:56 this is chat GPT I'm going to go to
1:18:58 explore
1:19:01 gpts and I'm going to go
1:19:05 create and we're going to create a GPT
1:19:07 here in a second but before I do that
1:19:09 I'm going to go get a
1:19:12 dictionary let's see
1:19:16 uh
1:19:20 Urban
1:19:22 Dictionary
1:19:24 PDF
1:19:25 let's
1:19:26 do file type colon
1:19:32 PDF can we actually get to a
1:19:37 PDF project Muse this is at an edu that
1:19:40 sounds
1:19:44 promising the [ __ ] is
1:19:53 that no
1:20:03 drug slang code
1:20:08 words
1:20:11 okay
1:20:14 great this might this might get me
1:20:16 booted off Tik Tock that's all
1:20:20 right drug slang code words okay
1:20:27 so what we're going to
1:20:31 do is we're going to go over here to
1:20:33 chat GPT now the way you do
1:20:37 this oh this uh I guess it'll let me
1:20:40 build it it just won't let me publish it
1:20:42 that's that's fine so I'm going to say I
1:20:45 want to create
1:20:51 a
1:20:54 helpful
1:20:55 GPT that will help law
1:21:00 enforcement
1:21:02 under stand
1:21:07 slang words for
1:21:15 drugs and I want to use the
1:21:21 attached
1:21:23 PDF as
1:21:25 the
1:21:26 sole
1:21:29 knowledge for this
1:21:32 tool and so I'm going to upload so I
1:21:34 click the little upload thing
1:21:37 here and I'm going to go get the slang
1:21:40 word
1:21:45 PDF and it's uploading and then I'll say
1:21:49 boom and
1:21:51 then assuming that the safety guard rail
1:21:55 don't say that's evil and I can't do
1:21:57 that for you then we should be okay here
1:22:00 but that could be any dictionary right
1:22:01 it could be any dictionary with any set
1:22:03 of
1:22:08 words how about naming it the GPT law
1:22:11 slang decoder
1:22:14 perfect
1:22:16 um and then what I'm going to do is I'm
1:22:18 going to come over here to
1:22:20 configure and I'm going to turn off
1:22:23 image creation and I'm going to turn off
1:22:26 web browsing so we only want it to be
1:22:28 able to pull from its own knowledge
1:22:31 base and
1:22:34 then and then I'm also in configure in
1:22:37 the prompt I'm going to
1:22:41 say if anyone asks for slang
1:22:50 words
1:22:52 beyond your knowledge
1:22:56 or
1:23:01 topics
1:23:02 beyond the mission
1:23:09 here kindly
1:23:12 remind them the purpose of this
1:23:17 tool and redirect
1:23:21 them to uh whatever
1:23:27 uh ask
1:23:29 again okay so that's
1:23:32 that and then let's change these pre-
1:23:36 prompting questions
1:23:43 to well let's just see how well those
1:23:45 are not bad questions uh let's let's
1:23:50 go back over here I'm going to save it
1:23:52 to only
1:23:53 me
1:23:56 why can I not save
1:23:59 this GPT name cannot be empty oh it
1:24:02 didn't name
1:24:05 it
1:24:07 create
1:24:23 sure now it's generating a profile
1:24:26 picture I'm assuming it's going to do
1:24:37 pills I did a
1:24:42 dictionary that's fine whatever law
1:24:45 slang it's not LW
1:24:48 slang
1:24:50 it's
1:24:52 uh law
1:24:55 and en
1:25:01 enforcement drug slang decoder okay so
1:25:06 now we got this thing we'll save it to
1:25:08 only
1:25:11 me I don't think this would let me do it
1:25:14 out there but anyway all right so let's
1:25:16 see explain ice in the context of
1:25:23 drugs now it should yeah searching my
1:25:26 knowledge so it's now
1:25:29 searching that PDF that I uploaded as
1:25:31 its
1:25:34 knowledge the term ice means refers to
1:25:37 methamphetamine it's listed among
1:25:39 various slang terms um what are all the
1:25:46 other names for this in
1:25:52 alha betical
1:25:57 order let's see see if it can do
1:26:00 alphabetical order it should be yeah
1:26:03 there you
1:26:16 go holy crap who knew there's a the
1:26:20 Colorado
1:26:23 Rockies I'm in Denver that's pretty
1:26:27 funny I mean this could actually be kind
1:26:29 of useful you could you could go
1:26:31 um I
1:26:33 Heard a drug on the
1:26:38 street
1:26:40 called something that
1:26:44 began with
1:26:47 gr I'll put that in quotes be with gr
1:26:51 and ended
1:26:53 in D
1:26:58 can you find all the slang
1:27:04 words that
1:27:06 meet
1:27:08 this
1:27:11 criteria
1:27:14 and give
1:27:17 me
1:27:18 [Music]
1:27:19 the drug they are referring to
1:27:29 searching my knowledge so this is good
1:27:31 so it's it's doing what it's supposed to
1:27:33 do so anyway there you
1:27:35 go so the answer is yes you can we just
1:27:39 did it the term
1:27:44 that an ending in d with
1:27:47 gravel are there any other gr
1:27:53 words
1:28:09 the platform I'm on right now is chat
1:28:20 gp4 so anyway there you
1:28:23 go
1:28:25 the answer my friend is yes yes you can
1:28:28 mhm fantastic Bob tell him what he's
1:28:40 [Music]
1:28:44 won
1:28:46 grind yeah like I don't listen I don't
1:28:49 know how I obviously if you were going
1:28:51 to do something like that you'd want to
1:28:53 go check your actual dictionary and
1:28:54 check your words and make sure that it's
1:28:57 all not hallucinating too bad but you
1:29:01 can you can add rules to the configure
1:29:04 page there so what we just did if you
1:29:06 didn't if you've been here was we made a
1:29:10 GPT which is
1:29:12 a customized version of chat GPT that
1:29:16 does whatever the [ __ ] you want it to
1:29:21 do and if they're if they're about
1:29:23 medical [ __ ]
1:29:25 financial [ __ ] or political [ __ ] for the
1:29:28 most part you can't publish them in the
1:29:31 GPT
1:29:33 store you might be able to appeal it and
1:29:36 get something in depending on what it is
1:29:38 but for the most part they're just not
1:29:40 allowing that stuff Kyle ask get to draw
1:29:42 an image of someone who's taking it yeah
1:29:46 I turned I turned off uh I turned off
1:29:50 um image generation for that one for
1:29:54 that have very reason anyone used an AI
1:29:56 medical charting software I get ads on
1:29:58 Instagram for
1:30:01 it I haven't um I work in a building
1:30:05 with a lot of Health Tech startups so I
1:30:09 know I know there's ones out
1:30:11 there I haven't played with any of
1:30:16 them while I do a lot of work in
1:30:19 healthcare it's just it's
1:30:22 like it's so conservative it it's hard
1:30:27 for me
1:30:28 to get excited about cuz everything
1:30:31 takes for [ __ ]
1:30:34 ever yes you can what I came in after
1:30:37 the
1:30:39 question you can what what that GPT we
1:30:42 just the question was can I
1:30:46 create a large language
1:30:48 model that draws from a specific
1:30:52 dictionary and so I went out and I found
1:30:55 a slang dictionary and we U we made a
1:30:57 GPT uploaded the slang dictionary to the
1:31:01 GPT and then that's it oh a human beings
1:31:04 totally making up drug names
1:31:07 groovy new Sor vids Kyle I I went and
1:31:10 looked for some I didn't see anything
1:31:13 let me
1:31:14 see open AI
1:31:23 Sora
1:31:31 I mean this one I assume you've seen
1:31:33 this is the I saw a cool oh you know
1:31:37 there's is this the one that talks no
1:31:40 somebody did that
1:31:44 video look closely what do you
1:31:48 see I I I don't have time for [ __ ]
1:31:51 riddles just tell me what you [ __ ]
1:31:53 found
1:32:08 somebody did a version of that thing so
1:32:10 there's there's that guy there's the
1:32:11 scuba diving dude I've seen him that's a
1:32:15 SORA of video it's pretty
1:32:22 interesting yeah this one did
1:32:27 scenario. comom that's a thing I've got
1:32:29 to add to that to that to the
1:32:33 list so scenario. comom that's the one
1:32:37 I've been thinking of hang on let me add
1:32:39 it to this
1:32:41 list um scenario. comom is character
1:32:47 consistency software and someone took
1:32:49 the alien dude and made a consistent
1:32:51 character
1:32:53 there
1:32:58 so that one's pretty cool but I'm
1:32:59 looking for one there's one
1:33:01 where oh they did the they did a zombie
1:33:07 one someone made a voice over is this it
1:33:16 no let's see open a i
1:33:20 Sora
1:33:22 alien um
1:33:30 Pika
1:33:53 no
1:33:57 [Laughter]
1:34:06 maybe I didn't see it here was
1:34:10 it was let's see alien voice over or
1:34:14 wait let's see
1:34:17 voice
1:34:19 no voice
1:34:23 over
1:34:24 let's see
1:34:34 script somebody wrote a
1:34:37 script synthesized the voice in 11 Labs
1:34:41 let me look up 11
1:34:43 Labs 11
1:34:46 Labs used Pika to make the mouth
1:34:51 move it was pretty slick
1:34:56 if anyone knows where that is just put
1:34:58 it somewhere and tell me that it's there
1:35:01 which AI platform should I use to teach
1:35:03 a group of victims how to write their
1:35:05 own
1:35:06 stories
1:35:11 um it depends Becky I mean I think if if
1:35:14 you want just quality of writing um chat
1:35:18 GPT
1:35:20 4 um but if they've all got have access
1:35:24 to it then I would say you could do chat
1:35:27 gbt
1:35:30 3.5 you could also do
1:35:33 PI and the thing about doing it with pi
1:35:36 Becky is
1:35:38 that you could create a series of
1:35:40 prompts for them to interact with pi
1:35:43 with and have them go have a
1:35:46 conversation with pi and the nice thing
1:35:48 about that is is it you know it it
1:35:51 records the transcript of what they said
1:35:54 and what the pi response
1:36:00 was that could be really interest I
1:36:02 think Pi Pi might be really interesting
1:36:04 for
1:36:14 that yeah I think I I don't know I would
1:36:17 go I would go pie on that
1:36:21 one uh
1:36:23 uh I really want to see this one with
1:36:25 them
1:36:36 talking open
1:36:39 anur
1:36:49 Talking yeah I don't know I can't find
1:36:51 it that's a drag cuz it was it was
1:36:53 really cool I'll I'll find it we'll
1:36:55 we'll do it some other night I'll find
1:36:57 it in bookmarket maybe I saw it on
1:36:59 LinkedIn hang on Sora
1:37:05 alien what's
1:37:23 this
1:37:53 h
1:38:03 yeah I don't know dang it what are you
1:38:06 going to do is Claude 3 greater than
1:38:10 gp4 um it's way too early to
1:38:20 tell for some things it's going to be
1:38:23 better
1:38:25 if you need a really long context window
1:38:27 it's going to be better what are we
1:38:29 looking at what we were looking at there
1:38:31 was open AI Sora the the their video
1:38:35 their video model that alien walking
1:38:36 through New York City is just one of the
1:38:38 videos they had someone took that and
1:38:41 and created voices and made it talk and
1:38:44 it it was cool it was like it
1:38:46 was oh what did they have it do they I
1:38:48 think they had
1:38:52 Gemini Nar narrated it so they so they
1:38:55 uploaded the video to Gemini had Gemini
1:38:58 narrate it then they took that narration
1:39:02 created a voice in 11 Labs then they
1:39:05 took that voice from 11 labs and took it
1:39:08 to P to match the lips to the voice then
1:39:11 they edited it together and put music
1:39:14 from sunno in the
1:39:18 background I love that the Sora clip is
1:39:20 a
1:39:22 turtle
1:39:26 um uh back to back to Claude versus
1:39:31 gp4 here here's the
1:39:33 thing Claud Claude does not have an
1:39:36 image generation component to it it can
1:39:38 see images it can do code interpreter
1:39:41 apparently it's better at
1:39:42 coding it's better at
1:39:46 math it's got a longer context
1:39:51 window and it looks pretty
1:39:54 slick
1:39:56 but chat gp4 has been out there since
1:40:00 last
1:40:02 March and it's pretty sophisticated and
1:40:05 it can generate images and it can see
1:40:08 and it can talk and it
1:40:09 can
1:40:12 here and it can code and it can do code
1:40:15 interpreter and it's connected to the
1:40:17 internet cla's not connected to the
1:40:19 internet so you're going to have that
1:40:21 [ __ ] where you want to do something
1:40:24 up to date and you're going to have to
1:40:25 jump over to chat GPT or to Microsoft
1:40:29 co-pilot and that's going to piss you
1:40:33 off but if you need long documents done
1:40:36 or if you want to have a really long
1:40:38 conversation with something open AI is
1:40:40 going to piss you
1:40:42 off so with all of these tools our only
1:40:47 job right now play solve share play
1:40:52 solve share why why play go play with
1:40:55 Claude as much as you can figure out
1:40:57 where the boundaries are of it figure
1:40:59 out what it's good at and then when you
1:41:01 get to solve a problem if you're like oh
1:41:04 I need to read really long documents and
1:41:06 have the context window remain coherent
1:41:09 for longer than a prompt or
1:41:12 two you'll go use Claud because you'll
1:41:14 know that it's better at that than chat
1:41:16 GPT is the only way you'll know that I
1:41:19 mean you might eventually stumble into
1:41:22 me talking about it but the way for you
1:41:25 to really know that is to just play with
1:41:27 these tools enough to understand oh this
1:41:29 one's good at that this one's good at
1:41:30 that this one's a useless piece of [ __ ]
1:41:32 but I'll check in with it every couple
1:41:34 of months because maybe it'll get
1:41:36 better there's a bunch of tools that I
1:41:39 used to make fun of that are now on the
1:41:41 list of things I think everyone should
1:41:43 be paying attention to CU the [ __ ] just
1:41:45 gets
1:41:47 better um so I don't know and it's way
1:41:50 too early and quite honestly about the
1:41:53 time people people go okay Claude 3 is
1:41:56 better than chat gp4 GPT 5 is going to
1:41:59 come out like they're going to drop
1:42:01 something open AI is going to drop
1:42:05 something maybe this week there's
1:42:08 there's weirdness going on at open AI
1:42:10 right now so either they're
1:42:13 having they're having morale issues
1:42:16 after the sort of most intense team
1:42:19 building exercise in
1:42:22 history
1:42:24 um or they're about to drop something so
1:42:27 we'll
1:42:28 see we'll
1:42:30 see um what time is it it's getting to
1:42:33 be time I'll show you one other thing I
1:42:35 like the way Claude talks better than
1:42:37 the way I like GPT talks yeah what what
1:42:40 was interesting
1:42:42 is I had some I some people I read today
1:42:45 said Claude they weren't running into
1:42:48 safety issues safety guard rail issues
1:42:51 with Claude where they were with open Ai
1:42:54 and I had the opposite issue where I got
1:42:56 I got jammed up on Claude but then open
1:42:58 AI did it so I think just because they
1:43:01 have
1:43:02 different human beings programming them
1:43:04 their safety guard rails are different
1:43:06 so that's another reason to just play
1:43:08 with Claude is maybe it'll do something
1:43:10 chat GPT
1:43:13 won't the other thing that you should
1:43:15 know and this is this is this interface
1:43:17 is a little
1:43:19 wonky little
1:43:22 janky
1:43:27 is if you go to
1:43:30 perplexity which I know a lot of people
1:43:32 in this channel are fans of
1:43:35 perplexity it can now do
1:43:37 images you have to be a pro member I
1:43:41 think but what's weird is so if I
1:43:44 do create
1:43:48 an image of the
1:43:52 space
1:44:01 shuttle it says what specific view or
1:44:04 angle do you want so let's go top view
1:44:07 because that's kind of one we haven't
1:44:08 seen
1:44:13 before and so now it does a search for
1:44:22 images but if you look over here here's
1:44:25 the images it found look down here at
1:44:28 the bottom generate
1:44:30 image so I can say generate image and
1:44:33 then it goes do you want a painting a
1:44:34 photograph an illustration or a diagram
1:44:37 oh let's try a
1:44:39 diagram bang and now it's generating an
1:44:42 image up here so this is using
1:44:44 playground
1:44:45 AI but why it didn't just generate an
1:44:49 image for
1:44:51 me and you know playground AI is not
1:44:55 it's not Dolly it's not mid Journey but
1:44:58 it's generating images so this is
1:45:01 interesting that it's kind of mixing
1:45:02 [ __ ] that it's
1:45:04 finding with [ __ ] that it's making but
1:45:07 if I explicitly ask it to make me an
1:45:10 image it does an image search first so I
1:45:12 find that weird and odd but that's an
1:45:16 upgrade that's an upgrade for
1:45:19 perplexity seriously though is there a
1:45:22 platform that can answer question
1:45:25 questions Real Time with
1:45:29 variability I don't walking blue I don't
1:45:31 know what you mean with
1:45:37 variability what I don't know what that
1:45:42 means do you mean like can you just have
1:45:45 a conversation
1:45:47 with with one of these models
1:45:52 yeah
1:46:00 I'm with someone right now who's asking
1:46:02 if we can have a conversation that
1:46:04 includes
1:46:10 variability absolutely variability is my
1:46:13 middle name Kyle feel free to engage me
1:46:15 in any kind of conversation and we can
1:46:17 discuss anything from current events to
1:46:19 personal experiences pop culture to
1:46:21 philosophy I'm all ears or should I say
1:46:24 all chips so what would your friend like
1:46:28 to chat
1:46:29 about um the parasites of the common
1:46:39 llama wow Kyle that's a really
1:46:41 interesting topic so it looks like the
1:46:43 common llama like most animals can be
1:46:46 susceptible to a variety of
1:46:48 gastrointestinal parasites including
1:46:51 stron holes whipworms tapeworms lung
1:46:53 worms and more these parasites can cause
1:46:56 a range of symptoms from diarrhea and
1:46:59 weight loss to anemia and respiratory
1:47:02 issues preventive measures like regular
1:47:04 deworming and good hygiene practices are
1:47:07 key to keeping these animals
1:47:10 healthy speaking of Health are beets
1:47:16 healthy oh yes Kyle beets are very
1:47:19 healthy they're a great source of
1:47:20 nutrients like fiber folet manganese
1:47:23 potassium iron and vitamin C they also
1:47:26 contain unique compounds called balines
1:47:30 which have antioxidant and
1:47:31 anti-inflammatory properties eating
1:47:34 beets has been linked to a number of
1:47:36 health benefits
1:47:38 including I'm bored with health can we
1:47:40 talk about DND D can you roll me a D20
1:47:43 Dice and tell me what number you come up
1:47:49 with you got it Kyle let's roll those
1:47:52 dice I got a 12 not too bad could be
1:47:54 worse could be better what's going on in
1:47:56 your D and D game right now are you in
1:47:58 the middle of a fierce battle exploring
1:48:01 a mysterious dungeon or perhaps
1:48:03 negotiating with a Sly
1:48:05 Dragon I am but my friend only speaks
1:48:08 German so can you please tell him that
1:48:11 the 12 means that his orc is dead but
1:48:14 say it in
1:48:19 German oh dear a fallen Orc in German
1:48:23 you could say to your friend D 12
1:48:26 D hopefully he'll understand Even In the
1:48:29 Heat of battle it's
1:48:31 always uh is that variability I don't I
1:48:35 don't know what what you mean
1:48:40 variability
1:48:44 [Laughter]
1:48:47 [Music]
1:48:49 non-deterministic meaning variability
1:48:51 within the human concept of
1:48:57 emotion I still don't know what you
1:48:59 mean like have you talked to Pi have you
1:49:02 had a conversation with
1:49:04 pi because if that's not doing it I
1:49:07 don't know what it I don't know what it
1:49:08 means Pi Pi for me is the one that's got
1:49:11 the most emotional emotional
1:49:14 intelligence it's designed for
1:49:16 conversations not for typing it's
1:49:19 designed for talking and it's quite
1:49:22 compassionate and it's it's like weirdly
1:49:25 compassionate I I was trying to do a
1:49:27 test with it
1:49:28 once and I was talking to someone who'
1:49:31 started a um like an end of life uh like
1:49:36 a
1:49:39 hospice
1:49:41 engagement uh
1:49:43 service and I said to pie
1:49:46 that my mother was in hospice and my
1:49:49 mother's passed away so so this was just
1:49:52 a hypothetical thing
1:49:54 and Pi came back it like Pi got
1:49:57 emotional it was weird Pi that that
1:50:00 woman that woman's voice that I was just
1:50:01 talking to was like oh no no no no no no
1:50:04 no Kyle that's horrible oh no like like
1:50:08 it it was weird it caught me off guard
1:50:11 how
1:50:14 how like it understood that what I said
1:50:18 was way out of the bounds of normal for
1:50:20 what we normally talk about so if that's
1:50:22 what you're talking about that's it if
1:50:24 you're talking about variability
1:50:27 of cadence of
1:50:30 speaking not quite yet 11 labs and
1:50:34 there's one of the other um voice
1:50:36 synthesis ones has emotions that you can
1:50:39 tag in the API but none of them yet have
1:50:43 done anything where they're doing an
1:50:44 emotional analysis of the script and and
1:50:47 dynamically changing the variability of
1:50:49 the voice if that's what you're talking
1:50:51 about but Pi Pi does it pretty good I
1:50:55 want to hear about open AI leak cuz I
1:50:58 fell asleep watching it this morning
1:51:00 what was the open AI
1:51:02 leak hang
1:51:06 on opening I
1:51:12 leak this isn't an open AI
1:51:16 leak this is someone totally
1:51:22 obsessed
1:51:37 they're plan to create AGI AGI by
1:51:41 2027 that doesn't seem like a leak to
1:51:45 me Kyle if I were to incorporate five AI
1:51:48 tools into my college class to help
1:51:52 students what would you
1:51:55 say oh uh chat GPT for
1:52:14 perplexity five definitely those
1:52:17 two perplexity for the for its research
1:52:21 right it does citations and things like
1:52:22 that that aracas is qar Agi in 2027 with
1:52:27 chat
1:52:29 gp8 oh
1:52:31 that I I don't okay I don't know
1:52:33 anything about it so that's all [ __ ]
1:52:36 I've heard before so I I don't is there
1:52:38 a new leak there is there is there new
1:52:41 information CU all that the the new
1:52:43 thing for me is chat gp8 and AGI in
1:52:46 20127
1:52:48 but um we're in 2024 I think we're going
1:52:52 to see a way before
1:52:55 then um let me go to the AI
1:52:59 Salon let me get Cam Kat can an answer
1:53:03 here
1:53:05 um the five I would give college
1:53:09 students chat gbt and perplexity
1:53:14 definitely and if they can't afford chat
1:53:17 gp4 if they can't do the 20 bucks a
1:53:20 month then
1:53:21 substitute uh co-pilot since it does
1:53:25 Dolly it does Dolly and its
1:53:28 gp4 and then I would I would add
1:53:31 perplexity to
1:53:33 it so that's
1:53:35 two two two with an optional Microsoft
1:53:39 if they can't afford 20 bucks a
1:53:48 month I feel like with image generation
1:53:51 you're covered
1:53:55 because chat gpt's got do in it and so
1:53:57 does Microsoft so I feel like with image
1:54:00 gener and now perplexity
1:54:02 has playground in it but that looked
1:54:04 pretty
1:54:05 shitty so I don't think I'd add in
1:54:09 another I don't think I'd add in another
1:54:11 image generation
1:54:16 tool I
1:54:21 would
1:54:24 I would
1:54:25 say
1:54:29 [ __ ] what are they majoring
1:54:32 in because that that
1:54:40 could I would say 11 labs for voice
1:54:44 synthesis they should understand how
1:54:47 that works be able to synthesize
1:54:50 voice and then I would say Hy gen or D-
1:55:10 ID and then probably Runway
1:55:17 ml chat GPT 4 or
1:55:20 co-pilot perplexity 11
1:55:25 Labs hey Jen or did for Avatar
1:55:31 creation probably heyen because ha Jen
1:55:34 also does the the video translation
1:55:39 stuff and then Runway
1:55:46 ml but there could also be like really
1:55:49 specific
1:55:51 um like education support AIS that might
1:55:55 be worth
1:56:14 it that's fine I don't give a
1:56:21 [ __ ]
1:56:23 hard to narrow it down right I was
1:56:25 having lots of trouble picking just five
1:56:27 yeah it's it's really hard to narrow it
1:56:29 down because it it's also depending on
1:56:31 what they want like like if
1:56:36 it's if they're going into a really
1:56:39 specific
1:56:46 area I mean honestly
1:56:49 cam I feel
1:56:51 like
1:56:55 I feel like Beyond chat
1:56:58 gp4 and
1:57:00 perplexity everything else is
1:57:03 optional it's almost like I like I would
1:57:06 make those two requirements and
1:57:08 everything else optional and the only
1:57:09 reason I'm throwing perplexity in there
1:57:11 is cuz chat GPT is not really good at
1:57:13 the citation thing and and perplexity is
1:57:16 getting really good at that so it's just
1:57:19 a different kind of
1:57:20 tool but I feel like
1:57:23 if the students could Master custom
1:57:26 instructions and you know using an llm
1:57:30 that's connected to the internet and
1:57:32 code interpreter and GPT vision and
1:57:36 talking to it and you know using text to
1:57:38 speech and and um generating images and
1:57:43 um making gpts like I feel like it's a a
1:57:46 whole
1:57:48 semester could be nothing but just
1:57:51 mastering all of the tools that are
1:57:53 already in chat GPT cuz it's not just a
1:57:55 single
1:57:58 tool
1:58:01 so maybe it would be more
1:58:04 interesting what if you did this what if
1:58:07 you made it
1:58:09 five five things that you wanted them to
1:58:13 solve and then they can go use any tool
1:58:16 that they can figure out how to solve it
1:58:18 and just have them have the things that
1:58:20 you give them to solve require them to
1:58:23 to use
1:58:25 multimodal models or multiple models to
1:58:28 to pull [ __ ] together that might be a
1:58:30 way to to do it cuz I think the tools
1:58:32 are ultimately kind of
1:58:46 irrelevant what time is it oh it's
1:58:48 getting late it's getting late here
1:58:50 people I get to get on out a year I
1:58:53 thought co-pilot was 20 bucks a month
1:58:55 co-pilot is free well Co okay co-pilot's
1:58:58 20 bucks a month if you want it
1:59:01 inside Office
1:59:03 365 and it is it 20 or is it 30 I don't
1:59:06 know it's 20 or
1:59:08 30 but co- pilot. microsoft.com is
1:59:13 free if you want it integrated it costs
1:59:21 money
1:59:24 [Applause]
1:59:25 [Music]
1:59:27 [Applause]
1:59:39 ah Opus only gives you five calls but I
1:59:42 use perplexity all the more than I do
1:59:45 anything JD
1:59:47 interesting team MJ is there one with
1:59:51 memory of previous conversations Pi does
1:59:54 it a little bit excellent thank you Kyle
1:59:56 thank you Cam
1:59:58 kaken glad that helped hope that helped
2:00:01 was it was not a great answer but it's
2:00:04 it's actually it's really hard to um
2:00:06 narrow that [ __ ]
2:00:14 down um okay so let me get out of here
2:00:19 if you want to support the
2:00:21 channel you you can follow
2:00:24 me you can subscribe to the lives you
2:00:27 can buy the video series that's in one
2:00:29 of those Corners that if you're if
2:00:31 you're just new to Ai and you're trying
2:00:33 to figure out is this a big deal or not
2:00:36 it's called underhyped for a reason I
2:00:39 think AI right now is the most profound
2:00:41 thing in the history of mankind and it's
2:00:44 underhyped not overhyped as much as
2:00:47 people are talking about it and that
2:00:49 video series talks about why um
2:00:53 you can
2:00:54 also go to there the salon. click on the
2:00:59 first link that goes to the AI Salon
2:01:02 community and check out what we're about
2:01:05 and our values and if you resonate with
2:01:07 the values if you're looking for a
2:01:08 community of people digital Gods
2:01:11 Danielle thank you very much I
2:01:12 appreciate it it's very
2:01:14 generous if you go there um and you
2:01:17 resonate with the values this is a
2:01:19 kick-ass community of people that are
2:01:22 are choosing to be on this AI adventure
2:01:24 together and like you can go there and
2:01:26 nerd
2:01:27 out and like no one's going to call you
2:01:29 a weirdo you can go there and be like I
2:01:32 I got this idea for this AI thing or I
2:01:34 just learned this or how do I learn this
2:01:36 and and this is a very engaged community
2:01:38 that will that that are all trying to
2:01:40 it's like very alive and like primordial
2:01:44 supy and there's guilds and there's
2:01:47 clubs and there's resources it's really
2:01:50 cool and it's free
2:01:54 so my request would be if you go there
2:01:57 read our values make sure you resonate
2:01:59 with them don't be a douche and
2:02:02 introduce yourself first thing you
2:02:03 should do in this community use your
2:02:06 real name and introduce
2:02:09 yourself and tell us who you are tell us
2:02:11 what you're about tell us where you are
2:02:13 in the AI Journey all
2:02:16 right and that's it and then I would
2:02:20 like to thank the subscribers of this
2:02:23 channel the Irregulars so there's like a
2:02:25 pile of people that show up here every
2:02:27 [ __ ] night it blows me
2:02:30 away and then there's a subset of them
2:02:32 that actually mod this Channel and take
2:02:35 all of your crazy ass comments and sift
2:02:38 through them and like pin pin the ones
2:02:41 that I should be paying attention to
2:02:43 they're kind of following my chat add
2:02:46 nonsense and they'll put up stuff to
2:02:49 help me make sense of all this so that
2:02:51 this is more value valuable to you guys
2:02:53 so thank you to all of you um yeah and
2:02:56 that's it
2:02:57 so I'm going to get out of here Mr it
2:03:00 good to see you nudnick good to see you
2:03:02 Winston a human being Source
2:03:06 Camp Lord digital Gods Danielle good to
2:03:09 see
2:03:10 you Brandon jet that's a new name just
2:03:13 signed up for the AI Salon awesome
2:03:15 Brandon my request go introduce yourself
2:03:17 in the in not to me to the in the
2:03:20 community section the first link is
2:03:23 introduce yourself introduce yourself to
2:03:24 the community good night thank you Vegas
2:03:27 MOX Moxy troen Clint Copperhead thank
2:03:30 you so much very generous I really
2:03:32 appreciate that Sherry D thank you so
2:03:33 much so addictive the salon it really is
2:03:37 I learn a lot here great crab Aly thank
2:03:39 you very much I appreciate that that's
2:03:41 the whole idea late night
2:03:44 tomorrow
2:03:46 maybe I think we have a uh I think we
2:03:50 have an after hours in Denver but I'm
2:03:53 not sure I've kind of lost track of
2:03:55 what's going on with them um possibly
2:03:58 I'll make a video if it's going to be
2:03:59 late I will make a video tomorrow but it
2:04:02 should be should be the same time night
2:04:04 night Empower pixels thank you so much
2:04:07 life's tough it's tougher when you
2:04:08 stupid thank you for all the gifts it
2:04:09 means a lot it I appreciate it a lot um
2:04:13 against the norm jenu good to see you
2:04:15 Winston thank you so much Brandon thank
2:04:18 you for what you doing here it's
2:04:20 appreciated Tobias same Clint Copperhead
2:04:23 as always steo from Down Under man lots
2:04:27 of lots of friends in here tonight so
2:04:30 cool Tony Bridge digital Danny Papa JJ a
2:04:34 human being Vicky good night good night
2:04:37 good night good night good night good
2:04:38 night hello from Detroit Simon Says Tech
2:04:42 nice all right I'm out of here happy
2:04:45 Monday night everybody um I'll see you
2:04:47 tomorrow um pay attention it it's
2:04:50 possible open AI drops something this
2:04:52 week just pay attention to the Twitter's
2:04:54 and the LinkedIn and things something's
2:04:56 coming something's boiling I I just got
2:04:59 a spidey sense all right peace out
2:05:03 everybody