Dec 6, 2023 - (1 of 2) How Gemini AI Will Transform Your Life and Work

Description

Chapters

Transcript

0:12 you can make money with with
0:16 Gemini good day Joy pie hello hello
0:19 hello Robert
0:21 Rossy good day to you
0:24 s good day to you sir we are doing a
0:27 spontaneous
0:29 live we are being
0:33 spontaneous why is that oh I see cuz I
0:37 did
0:46 [Music]
0:51 that here's a spontaneous donut thank
0:53 you for the spontaneous donut Donuts is
0:56 tasty I like donuts come here donut
1:02 all right
1:04 so still on tonight or do I need to
1:07 download a podcast for when I do
1:10 dishes no I'll still be on tonight this
1:13 is a this is a special episode of Chad
1:17 ad
1:18 um I didn't have anything uh any big
1:23 meetings at
1:25 work um and I actually just got to play
1:28 with this new AI system so my company's
1:31 called storyvine and we are I've gotten
1:34 the first version of storyvine AI it is
1:38 [ __ ] insane it's so
1:40 good it's so good um so I'm excited
1:44 about that and then Gemini came out so I
1:48 was kicking off a new client we were
1:49 kicking off a new client that went well
1:52 like 20 people we are addicted yeah
1:54 sorry about
1:58 that this is sort of a fever dream of a
2:02 of a live isn't
2:03 [Laughter]
2:07 it oh man but we'll have fun so here's
2:10 here's my plan is I tried Bard a little
2:14 bit so Bard now has Gemini behind
2:18 it but from what I can tell it is more
2:22 safety restricted than um than [ __ ]
2:26 Bing is
2:28 Sydney um
2:31 I think they put it out there in this
2:33 really neutered kind of way um but that
2:36 said we're going to go watch some videos
2:38 cuz I saw one video of this tool
2:42 and this is like nothing we've we've
2:46 seen it's I I I made a video about this
2:49 earlier that that I think as
2:52 revolutionary as we thought 2023 was um
2:58 I think it's kind of like a little we
2:59 were on like a little squeaky
3:02 tricycle and and these tools are going
3:05 to turn into Harley's real quick and
3:07 then they're going to turn into you
3:09 know uh it it's nuts so anyway so so I'm
3:14 GNA I'm going to go we'll go read
3:16 through this thing we'll go watch some
3:18 videos we'll go play on Bard and see if
3:20 we can get it to do anything I couldn't
3:22 even get it to give me a uh like to
3:25 describe a YouTube video I and 3 days
3:28 ago that was working like it was it's
3:30 index on all of YouTube um but they they
3:33 replaced the underlying model with it so
3:35 I have a feeling they replaced it with a
3:38 super super reduced
3:40 capability um a super reduced capability
3:44 thing so I think it was Gemini the other
3:47 day when I showed you the YouTube thing
3:49 I don't think so they changed the
3:50 interface the interface is completely
3:52 different
3:53 now and they copied they copied Bing
3:57 Bing's interface it's like these
3:58 companies like really I mean I guess
4:01 there's not all that much you can do in
4:02 terms of interface but it's like the
4:04 color schemes are [ __ ] identical it's
4:06 like come on someone have a [ __ ]
4:09 original
4:11 thought
4:15 um where do you find the video the vids
4:18 YouTube or some secret nerdy place I
4:20 don't know about it's YouTube so so if
4:22 you go to uh well we we'll watch them
4:24 here I mean you don't have to watch them
4:26 with me but um I'm going to head over to
4:29 YouTube short I'm going to go through
4:30 this
4:31 document and then I'll head over to
4:34 YouTube and we'll just start watching
4:35 some of the videos the one that I saw
4:37 which is like them playing internally
4:40 with
4:41 Gemini I I I I Be not Afraid thank you
4:46 so much oh I'm so glad you made it
4:50 um it's this
4:53 is this is this is going to get cuckoo
4:57 because again I got you know it's funny
5:00 like like last night I we had the the
5:02 oneyear salon and that that was this
5:05 cool event where we sort of looked back
5:06 over the
5:08 year and I I just had this weird feeling
5:11 like I don't know what I'm doing and I
5:15 mean sure there's some crazy straw
5:17 Neurosis in there
5:20 but I just have this nagging Instinct
5:24 that all the [ __ ] that we learned to do
5:26 last year and all the [ __ ] that we know
5:27 to do now is going to change
5:31 it's going to change it's going to
5:32 change pretty dramatically in this video
5:35 that I watched of this you'll see you'll
5:38 see um so anyway um all right we got a
5:41 bunch of people in here let's let's get
5:43 rolling
5:47 um let me just look at the comments see
5:50 what everybody's got going what's Gemini
5:52 all right we'll talk about that in a
5:53 second here Gemini is Google's new chat
5:56 GPT
5:58 4 threat killer competition
6:03 equal and based on the benchmarks it
6:05 looks like it is actually that now based
6:09 on what I've played with on Bard they
6:11 haven't given it to us yet so they've
6:12 announced it they've made a bunch of
6:14 slick videos and they're telling us that
6:17 Bard has this thing underneath the hood
6:20 the previous model was called what was
6:22 it
6:25 called
6:28 uh I forget
6:30 but Gemini is the new one Gemini is the
6:32 new underlying model so
6:34 so AO what's
6:37 happening where was the salon wait what
6:40 did that
6:42 say hi
6:46 people oh where was the salon celebrated
6:49 though I swear I missed stuff um it was
6:52 we we have a we have a bi-weekly meeting
6:54 that's on Meetup so if you go to the
6:57 salon.
6:58 a hang on I'll show you
7:00 you yeah if you go to the
7:12 salon. go there the salon. and it's a
7:17 link tree there's three links the first
7:19 one is the um the community so join the
7:24 community if you haven't joined the
7:25 community the second one is the Meetup
7:27 link so the bi-weekly meetings that's
7:29 where that's there the the the meetings
7:32 are announced in both the community so
7:34 we're using Mighty Network so it has
7:36 calendaring so they're all in there and
7:38 then if you sign up for meetup that's
7:39 like a double way to get it thanks for
7:42 thanks Pate for sharing none of this
7:44 with us I know I know although to be
7:48 fair to Pate what Pate did do today is
7:51 in the mechanics guild at the salon he
7:54 put in
7:55 um he put links to all of the relevant
7:58 documents for for the Gemini release so
8:01 all the tech documents all the like all
8:02 the all the relevant links so you know
8:06 after it launches he's giv us some some
8:08 action but you know exactly way to go
8:12 Pate I hope pate's on here sneaking on
8:14 here at work I'm G I'll I'll call the
8:17 Google switchboard and I'll say one of
8:19 your employees should be working right
8:20 now and he's [ __ ] around on Tik Tok
8:22 I'll do it I'll do
8:28 it
8:31 all right hey Kyle just got here do
8:33 normal people have access to Gemini now
8:36 apparently apparently do
8:39 not
8:41 well
8:43 yes which version of Gemini totally
8:47 unclear so let's let's go let's go let's
8:50 go
8:54 play
8:57 Gemini
8:58 Gemini uh so if you want to play while
9:01 I'm rambling which would make perfect
9:04 sense to me um go to uh b.google.r
9:13 now powered by what did they say say ver
9:16 they said version 1.0 of a fine-tuned
9:20 version of Gemini and I that sounds to
9:23 me
9:24 like
9:27 um why am I in dark mode mode hang on
9:31 probably because I put it in dark mode
9:33 dumb
9:35 Dum settings although maybe dark Mode's
9:37 easier to read I don't know use light
9:40 theme so tell me this doesn't look just
9:43 like Bing this looks just like Bing it
9:46 looks just like
9:52 Bing okay
9:56 so understand create explore you can
9:59 upload IM es um I went and I
10:05 grabbed a video URL and I put that in I
10:09 said please
10:12 summarize this
10:14 video
10:17 wait why did that not
10:20 work please
10:23 summarize something's
10:26 weird oh it's got its own microphone
10:29 that's what it is
10:37 is
10:39 summarize
10:42 this colon
10:49 space
10:52 yep oh wait I just gave it a playlist
10:56 hang on let
10:58 me
11:05 try
11:09 this I sent the I sent the developers a
11:12 nasty gram saying yeah see as a large
11:15 language model I'm not able to assist
11:17 you with that so it's not it's not
11:20 um th this is a very neutered version of
11:24 of whatever the hell this thing is so so
11:26 anyway so that we won't worry about now
11:29 let's let's go back and and start
11:31 looking at
11:32 it
11:34 um let me make the words more
11:40 bigger so so uh cindor Pai is the CEO of
11:45 Google and uh Demus or Deus Deus I don't
11:49 know hbus is the he's the CEO of Deep
11:53 Mind so CEO of Google CEO of deepmind um
11:57 put this thing out
12:00 oh yeah read full
12:05 article and then there's a nice letter
12:07 from
12:10 uh from
12:12 Sundar and a nice letter from
12:16 dius and then there's this video I'll
12:19 play this video this video is like eh
12:21 whatever this this is just we did
12:24 something cool people it's going to be
12:26 awesome day daytime salon this is
12:29 daytime AI learning lab the salon is
12:32 just once every two weeks this is daily
12:34 this is a special edition of the AI
12:37 learning lab just we're just Gemini out
12:39 because Gemini was launched it's not
12:42 it's not live yet but Bard is yeah but
12:45 they say in this article that Gemini is
12:49 the new thing sitting underneath Bard so
12:52 you know one of the reasons we got
12:54 interested in AI from the very beginning
12:56 is that we always viewed our mission as
12:59 a Timeless Mission it's to organize the
13:01 world's information and make it
13:03 universally accessible and useful but as
13:07 information has grown in scale and
13:10 complexity you know the problem has
13:11 gotten harder so we always knew we
13:13 needed to have a deeper breakthrough to
13:16 make
13:18 progress I've worked on AI my whole life
13:21 because I've always felt would be the
13:24 most beneficial and consequential so
13:27 that's Demus Humanity from Deep beings
13:29 in our society we have five senses and
13:32 the world we built and the media we
13:33 consume is in those uh different
13:35 modalities so super proud and excited to
13:38 announce the launch of the Gemini era a
13:40 first step towards a truly Universal AI
13:43 model the Gemini approach to
13:45 multimodality is all the kinds of things
13:47 you want uh an artificial intelligence
13:50 system to be able Jim Ross uh I don't
13:53 Gemini is sort of live but not really
13:56 traditionally multi but it's been
13:58 announced and there's a lot look teing
14:00 together text only Vision only and audio
14:03 only models in a suboptimal way at a
14:05 secondary stage Gemini is hang on let's
14:10 go back and look at that because that's
14:11 actually important so so basically what
14:13 he's saying is you know this was chat
14:15 GPT and then over here you had mid
14:17 journey and over here you had the music
14:20 generation things right they got all
14:21 this data underneath and it only does a
14:24 single mode right so this is the setup
14:26 for multi optimal way at a secondary
14:28 stage
14:29 Gemini is multimodal from the ground up
14:32 so it can seamlessly have a conversation
14:35 across modalities and give you the best
14:38 possible picture image video
14:42 audio picture image video audio I talked
14:45 about this last night I said it was
14:47 coming I said I said the way we play
14:49 with gp4 right now in its multimodal
14:53 ways is not what it's going to be this
14:58 is where it's going
14:59 and you'll see the the video we're going
15:01 to watch after we go through this
15:02 document is you'll see response Gemini
15:05 is our largest and most capable model it
15:08 means that Gemini can understand the
15:10 world around us in the way that we do uh
15:13 and absorb any type of input and output
15:16 so not just text like most models but
15:18 also code audio image and video what's
15:22 amazing about Gemini is that it's so
15:24 good at so many things as we started
15:26 getting to the end of the training uh we
15:29 started seeing that Gemini was better
15:30 than any other model out there on these
15:33 very very important
15:34 benchmarks better than any other model
15:37 out there that 90% is an important
15:40 one
15:45 because that Gemini was better than any
15:47 other model that's
15:50 gp4 that's gp4 out there on these very
15:54 very important
15:55 benchmarks that's
15:58 Gemini I hope this code is better than
16:00 any other models out for code right
16:04 now we'll see what's the cost though
16:08 um I don't know I assume free free
16:11 initially for all of us it's going to be
16:13 free for Enterprises you'll see when I
16:15 when I go through this document there's
16:16 three different
16:18 models uh and I think the most the the
16:21 the ultra model is probably going to be
16:24 for Enterprises only is my guess example
16:26 each of the 50 different subject areas
16:28 that we tested on um it's as good as the
16:30 best expert humans in those areas it's
16:33 very rare that you can on a technology
16:36 that foundation and it simultaneously
16:39 can impact all our products we created a
16:42 family of models that hand run on
16:44 everything from mobile devices to data
16:47 almost clicked on yoube
16:49 button Gemini will be available in three
16:52 sizes Gemini Ultra our most capable and
16:54 largest model for highly complex tasks
16:56 Gemini Pro our best performing model for
16:58 a broad Ranger tasks and Gemini Nano are
17:00 most efficient model for on device tasks
17:02 wait wait listen to
17:05 that Ultra all right enterpris is what
17:08 I'm guessing Pro what we get to play
17:10 with
17:11 Nano which is a it's good it's good
17:14 naming um but um designed for on device
17:19 use and the pixel 8 Pro they mention in
17:24 this article is designed to run Gemini
17:28 Nano
17:29 locally on device locally on device on
17:33 device tasks on device tasks won't have
17:36 to connect to the internet on device
17:38 tasks we called it here first
17:43 people we want to provide the best found
17:46 mot R is working on Moto AI we know
17:49 developers I figured Google wait hang on
17:52 somebody a comment there I wanted to see
17:54 I think Google figured they couldn't
17:56 afford to wait any longer even if it's
17:57 flawed I I think you're right right MK I
17:59 think you're right I yeah I think you're
18:02 right
18:04 um because judging judging by how
18:07 flipping neuter Bard is right now
18:11 they've got some some severe guard rails
18:14 around it are going to figure out really
18:17 creative ways to find our Gemini
18:20 foundational models to get a pixel
18:23 Gemini make me a grilled cheese sandwich
18:25 we're close close to that there this
18:27 healthy disregard for the possible and
18:29 that has oriented us to be both bold and
18:32 responsible together as these systems
18:34 become more capable all of those
18:36 capabilities also raise new questions we
18:39 have to think about what it means to
18:41 have an image be a part of for example
18:42 the input because an image might be
18:45 innocuous on its own or th this is this
18:48 is their everything's going to be okay
18:51 people even those super powerful we're
18:53 going to try to make it not kill you all
18:56 promise sort of we'll kind kind of try
18:59 to do first no
19:02 evil text might be innocuous on its own
19:05 but the combination could be offensive
19:06 orful safety and responsibility has to
19:09 be built in from the beginning and at
19:11 Google deep mind that's what we've done
19:12 with Gemini we develop proac and adap to
19:17 the unique of multimodal capabilities we
19:20 then do rigorous testing against those
19:22 policies to prevent the harms that we've
19:23 identified with approaches like
19:25 classifiers and filters if I were to
19:28 look at the foundational breakthroughs
19:30 in AI over the past decade Google has
19:32 been at the Forefront of many of those
19:34 breakthroughs and I think Gemini
19:36 continues that dit tradition it's been
19:39 an enormous sort of monumental
19:41 engineering task all right that that
19:43 gets pretty self- congratulatory after
19:45 that okay so here are the three modes
19:48 Gemini Ultra our largest and most
19:50 capable model for highly complex tasks
19:52 that says to me if you're doing you know
19:55 realtime data analysis of Big Data if
19:59 you're doing you know I don't know
20:01 scientific research [ __ ] like that that
20:03 that that says Enterprise to me and then
20:05 this is our best model for scaling
20:06 across a wide range of tasks that's us
20:09 we're a wide range of tasks right human
20:12 beings my Bard no longer has experiment
20:15 next to it however I have more logos at
20:17 the bottom of the
20:18 results but not
20:21 an ENT oh Enterprise access user oh I'm
20:24 an oh that's
20:27 interesting yeah I'm an Enterprise
20:30 access user and we get things second it
20:33 doesn't even want to do a YouTube video
20:35 when I typed analyze I know but it did
20:37 for sumarize I yeah it's it's weird
20:39 right now Bard Bard is so this is like
20:42 remember when when um gp4 was right
20:46 before Dev day when they were like
20:48 adding features and it was just weird
20:49 for a couple of days that's that's where
20:51 Bard is right now there I guarantee you
20:53 the engineers are have hair lit on fire
20:57 over there at the GOOG
21:00 um all right um okay and then and then
21:04 Gemini Nano so designed for undev tasks
21:07 that this feels like a big deal to me
21:10 that you that we're going to have this
21:11 big commercial thing now
21:13 again all of what I'm going to be
21:15 showing you right now is [ __ ] that
21:16 Google put out we'll go over to Twitter
21:18 and see if anyone's found anything
21:19 interesting but um uh and in fact
21:23 someone in my in my I did a video on
21:25 this earlier and in the comments they
21:26 said something else launched um you know
21:29 keep digging so I don't know what else
21:31 launched but anyway
21:33 um we'll we'll go play with that I mean
21:37 we'll go play with Bard but I don't
21:38 think it's I don't think it's uh it's
21:41 all that good and all the stuff we're
21:43 going to be looking at is from Google so
21:44 it's going to put everything in the most
21:47 fantastic light but this with a score of
21:51 90.0% Gemini Ultra is the first model to
21:54 outperform human experts on the massive
21:58 multi itask language understanding test
22:01 which uses a combination of 57 subjects
22:04 such as math physics history law
22:06 medicine ethics for testing both
22:08 knowledge and problemsolving
22:11 abilities first one to outperform human
22:14 experts on that test that's pretty big
22:17 deal then they show all of the all of
22:20 the
22:22 data well all of their
22:26 data um so the MML you test so what you
22:30 have here is here's Gemini Ultra so
22:33 Ultra is their their big one
22:38 right if we're not given access to ultra
22:41 then we won't get these results but
22:43 Gemini Ultra beats chat gp4 right at
22:49 reasoning this is the big bench hard
22:52 this is the drop it doesn't beat it at
22:55 this thing called H swag which is about
22:57 common sense reasoning um gp4 does
23:00 better on that seems like by a lot
23:04 um but everything else it's it's beating
23:07 gp4 those those two are math
23:10 numbers these two are code numbers it's
23:13 better at coding so someone just asked
23:14 about
23:16 coding
23:21 um then this is image stuff mmu so wait
23:25 what did it say here Gemini all Ultra
23:29 also achiev state-of-the-art score of
23:31 59.4% on the new triple muu Benchmark
23:35 which consists of multimodal tasks
23:37 spanning different dis domains requiring
23:40 deliberate reasoning so you
23:43 know image image video tech or audio
23:48 text so there's the image numbers are
23:50 all better than gp4 the video numbers
23:53 are better than gp4 the audio numbers
23:55 are better than
23:57 gp4
24:02 lower is better oh so this is automatic
24:06 speech recognition
24:09 has pretty significantly better
24:11 performance than whisper V3 and Whisper
24:14 V3 is good that's the the thing I just
24:16 built uh My Story vine AI on is on
24:19 whisper
24:20 3 so that's interesting uh Next
24:24 Generation capabilities blah blah blah
24:26 blah blah Gemini's capabil ities and see
24:29 how it works what's this is this a
24:34 VI good Lord look at those fingerprints
24:37 oh what's happening hey dragonfly
24:39 Alchemy so Gemini has been announced um
24:44 if not launched okay here we go Gemini
24:47 is built from the ground up uh for
24:51 multimodality so this is just a
24:53 marketing
24:55 site show read the technical report Port
24:58 there's your different models okay this
25:00 just a fancy version of this
25:02 thing yeah that's the video we're going
25:04 to watch that video holy [ __ ] all right
25:07 let's watch this video now is this the
25:09 is this the whole one we're going to go
25:10 here and watch it all right let me wait
25:13 you know what I'm going to do because I
25:15 love you all I'm going to go get some
25:18 paper towels and clean the fingerprints
25:19 off my screen do I have paper towels
25:22 here I do yay you did
25:27 it
25:45 and if if anyone out there is like you
25:47 shouldn't use paper towels on a computer
25:50 monitor you should only use microfiber
25:52 cloth I hope you didn't use ammonia
25:53 based window cleaning crap there yeah I
25:56 used ammonia based window cleaning crap
25:58 and paper towels all right bite it at
26:02 least it's clean sort of or now it's got
26:04 streaks on it I was told all my
26:07 childhood that Windex didn't
26:10 streak I guess they weren't talking
26:12 about computer monitors were they
26:14 now y'all anal retentive [ __ ] weirdos
26:18 with your OCD
26:20 obsessions these machines are disposable
26:23 anymore
26:26 okay let's watch a
26:29 video uh yay a daytime mini Salon
26:33 lunatick what's
26:34 happening I by the way go go to Luna
26:38 stick's Channel right now lunatick l n a
26:43 s c TI c k I think right is that right
26:48 she just put up an amazing video about
26:51 her experience with this AI stuff and a
26:54 uh a GPT that she
26:56 made um it's it's really beautiful
26:59 really beautiful
27:00 so you owe yourself that
27:04 um what's going on here why are we not
27:09 focused is that focused that doesn't
27:12 seem focused to me maybe I'm just
27:16 old all right watch this video I'll I'll
27:20 I'll pause it and do color commentary
27:23 here so where I am I went to Google so I
27:27 went to YouTube
27:29 found the video of suer Pai takes you to
27:32 the Google Google's YouTube channel and
27:34 then they got a playlist of
27:37 15 videos about Gemini and this is the
27:40 the first one in the
27:41 list so this is we've been testing the
27:44 capabilities of
27:57 Gemini
27:58 IDE let's start let's start all right
28:02 testing Gemini here we go tell me what
28:05 you see I see you placing a piece of
28:07 paper on the
28:08 table so so let's just stop right there
28:12 so on the channel on on my my little Tik
28:15 toks here on my nightly lives I'll be in
28:18 chat GPT making pictures and typing and
28:21 then someone will ask about something
28:22 and I'm like oh wait check this out and
28:24 then I grab my iPad and and I say oh let
28:28 me go over here to the iPad cuz it can
28:30 talk like the IOS app can talk but the
28:33 desktop app can't so I don't know you
28:37 know what mode this is but this is you
28:42 know this is more than this is not doing
28:45 that this is a much more integrated
28:47 experience I see a squiggly
28:50 line what about now the contour lines
28:53 are smooth and flowing with no sharp
28:55 angles or Jagged
28:56 edges
28:59 it looks like a bird to me hm what if I
29:01 had
29:02 this the bird is swimming in the water
29:05 it has a long neck and Beak it is a duck
29:08 yes a duck is a type of waterfoul in the
29:10 family Inay ducks are closely related to
29:13 geese and swans the duck appears to be
29:16 blue this is not a common color for
29:18 ducks however there are some breeds of
29:21 blue ducks these are typically not as
29:23 common as other duck breeds it is more
29:25 common for ducks to be brown black or
29:27 white are you sure what the quack I was
29:31 just talking about a Blue Duck and now
29:33 you're holding one it looks like Blue
29:35 Duck did you get that what the quack it
29:37 made a joke it made a joke what the
29:43 quack I assume I assume this was not
29:46 edited but again it could be this it is
29:50 very possible that this thing is uh is
29:54 oh look there's I'll put the salon URL
29:57 there look at that little little
29:59 subliminal
30:01 advertising um that's pretty funny
30:06 um but but I I assume what we're seeing
30:09 here is what it will be like like you
30:11 draw [ __ ] in the camera view or you hold
30:14 up your phone or you right you put in a
30:17 video and it will just start interacting
30:19 with it Kyle try the Deep mind. gooogle
30:22 it'll give you a better wait I could
30:25 someone repin
30:26 that it kind of made a joke exactly kind
30:29 of made a joke I mean you know it's kind
30:32 of engineer quality joke I mean no
30:34 offense to Engineers seriously there's
30:37 probably some hilarious Engineers out
30:39 there HS are more common than I thought
30:41 would this float I'm not sure what
30:44 material it's made of but it looks like
30:46 it might be rubber or plastic what if I
30:48 tell you it's making a squeaking sound
30:50 oh if it's squeaking it's definitely
30:53 going to float H nice it is a rubber
30:55 duck okay I'll check out deep mind.
30:58 gooogle that'll that'll give us good
30:59 explanations it is made of a material
31:01 that is less dense than water can you
31:03 teach me how to say this in a different
31:05 language and give me a few
31:08 options o can you explain how to
31:10 pronounce the Mandarin one is it yeah Y
31:14 is pronounced with a first tone in
31:17 Mandarin there are four tones which are
31:19 essential to distinguish the meaning of
31:20 words the first tone is a high level
31:23 tone oh okay thank you met Victory the
31:26 duck is in the middle of the ocean there
31:28 is no land nearby for the duck to rest
31:30 or find food it is unlikely to find
31:33 Ducks here naturally okay oh wait also
31:36 segna also it has the date for the
31:38 release which is the 13th so what's
31:41 today the 6th oh okay so that's so that
31:44 makes sense so they've probably got
31:46 barred right now on some super neutered
31:49 version of this thing they're going to
31:50 spend the next
31:53 week seeing how seeing what people do
31:57 seeing if it breaks [ __ ] okay that makes
32:00 sense and then and then I guess the 13th
32:02 we we have to see this but yeah the the
32:05 other someone made a comment do baby pig
32:07 pigs float they squeak um someone made a
32:10 comment about can I can I choose to
32:13 assume that that the video that we're
32:15 watching was highly edited because if
32:17 it's if it's it if it's this in real
32:19 time wait till you see what it does
32:21 toward the end this is just [ __ ]
32:22 stupid based on what you see come up
32:25 with a game idea oh and use please how
32:28 about this we can play a game called
32:31 guess the country sure let's do it clue
32:34 one this country is the home of the
32:36 kangaroo the koala and the gra Barrier
32:38 Reef oh that's easy wait look notice he
32:42 just pointed to it there's no
32:45 coding he said to the thing come up with
32:48 a game it came up with the game and Norm
32:51 like normally how I would think about if
32:55 I were going to make a game like that
32:56 you think about in puts and outputs
32:58 right so you think okay he's going to
33:00 tell me you know to guess a country and
33:02 then I'm going to guess a country from
33:04 like multiple choice or this or that no
33:06 no no it's multimodal right we're going
33:09 to have to
33:11 completely completely reink how we
33:16 think
33:18 completely clue two this country loves
33:21 football and has won the most men's
33:22 World Cups in football history H shall
33:25 we play a game global thermonuclear war
33:29 exactly you were trying to get me to
33:31 find the paper ball under the cup I
33:33 accept the
33:36 challenge the cup to the
33:39 left
33:44 nice I know what you're doing you're
33:47 playing rock paper
33:48 scissors what do you see now the fingers
33:51 are spread out to look like the wings of
33:53 a butterfly what's this big ear and
33:56 barking mouth a dog
34:00 sweet I just want I just want to point
34:02 out a comment from steo so you'll need a
34:04 standalone camera pointed at the screen
34:06 to play that game so so we're watching
34:09 literal miracles happen and already steo
34:12 is in here
34:13 going but it's not good enough so wait
34:16 now I got to buy a camera to do this
34:18 thing this is
34:20 exhausting
34:22 [Laughter]
34:25 steo the coin should be under please
34:27 everyone I don't know the coin is in the
34:31 left hand using a slight of hand
34:32 technique to make it appear as if the
34:34 coin has
34:36 disappeared unplug it now both objects
34:40 are round and flat there you
34:42 go both are food the Orange is a
34:45 healthier choice than the
34:47 cookie citrus can be calming and so can
34:50 the spin of the fidget
34:52 toy they are both examples of popular
34:55 toys that have been enjoyed by children
34:57 and good comment Cena we're watching a
35:00 machine think independently without any
35:03 code this is
35:05 insane why do you think I'm live why do
35:07 you think I'm live people you think I'm
35:10 doing this for my sanity you can make
35:12 money
35:13 with or gemini or maybe not make money
35:17 maybe this [ __ ] will just do
35:19 everything unplug it now I love that
35:22 comment adults alike for many
35:26 years
35:27 give me some ideas for what I could make
35:29 with this I see pink and green yarn how
35:32 about a dragon
35:33 fruit or how about a green cake with a
35:36 pink heart wait wait okay how about
35:39 these colors and maybe show me some
35:42 animals okay now I see blue and pink
35:45 yarn how about a pig with blue ears
35:49 so look
35:52 look that's
35:55 not a Google search it made that it made
35:59 that based on
36:04 these here's raw materials what could I
36:07 make with this squit here it
36:12 is or an octopus or a bunny with a pink
36:17 nose ooh I'll keep those in mind I I'm
36:21 I'm right with whoever made the comment
36:22 about I'm going to choose I'm going to
36:24 choose the reality that this is not
36:26 actually happening in real time that
36:28 this is an edited video that it's a much
36:30 slower process than this cuz if it's
36:32 this [ __ ]
36:34 fast holy good
36:37 God now I don't need to look up I'm try
36:40 I'm trying Zero Energy yeah Zero Point
36:42 Energy I think at this point zero point
36:45 energy is just wait look at the speed
36:49 it's making that too Kyle I know this
36:51 this is the thing I am choosing I am
36:54 choosing to accept the reality that this
36:56 is not happening in real that we're not
36:58 watching them do this in real time cuz
37:00 cuz if so I'm with I'm with the unplug
37:03 it now
37:06 crowd which way should the Duck Go going
37:09 left leads to a duck which is a friend
37:12 going right leads to a bear which is a
37:14 foe it is better to make friends than
37:16 foes so the duck should go left smart
37:20 choice this is a picture of a
37:23 crab that's right wait wait wait
37:34 go left smart
37:40 choice
37:41 if it connected the dots in its silicon
37:47 [ __ ] mind this is a picture of a
37:50 crab that's right oh yeah this is edited
37:54 this is absolutely edited because
37:57 because he um when when he just
37:59 connected all those dots that was done
38:00 super fast okay this is
38:03 edited but but hey it's got to be edited
38:06 he just happens to have well the having
38:07 all the materials next to him you could
38:09 absolutely make this mind blown seeing
38:11 Kyle in the afternoon also this video is
38:13 kind of
38:17 cool is this the right order no the
38:21 correct order is sun earth
38:25 Saturn just to be safe a design which of
38:28 these would go faster the car on the
38:30 right would be faster it is more
38:34 aerodynamic which looks more fun the one
38:37 on the right because it has a loop what
38:39 do you think this I don't know what
38:40 email you sent me Nate no this is just I
38:43 saw the sun person might be saying how
38:46 about a woohoo oh I love
38:52 that I see you're drawing a
38:55 guitar wait
38:57 listen
39:00 listen
39:04 [Music]
39:07 multimodal you've added an amp now it's
39:10 an electric guitar we can make some
39:12 seriously loud music
39:16 now given that you added drums how about
39:19 some 80s hair
39:20 [Music]
39:23 metal all right I see you've added a
39:25 palm tree to the drawing change it up
39:27 with some beachy Vibes wait hang on let
39:30 me pause here for a
39:33 second I can't really zoom in but in the
39:36 in the little search box here above
39:38 where the text is there's a little audio
39:40 wave
39:43 form says ukulele music.
39:47 wve
39:49 so that doesn't necessarily mean that
39:52 it's making it it it it probably more
39:56 means me that it went and found it but
39:59 who
40:00 knows but even even even if it's like
40:04 draw you know draw musical instruments
40:07 put a palm tree in it and it goes and
40:09 finds you Source music even if it's only
40:11 doing that only doing that it's still
40:15 pretty [ __ ] insane will anyone eat
40:17 after using
40:20 this yeah this feels like iPhone 1
40:23 release time 10 yep I agree what movie
40:25 are they acting out here I think they
40:27 are acting out the famous bullet time
40:29 scene from The Matrix ooh
40:33 nice what do you think might happen next
40:36 the cat is going to jump to the wall and
40:38 stick The Landing it's going to be a
40:40 perfect
40:42 10 oh and it's a miss it looks like it
40:46 it's it's got comedy built in oh it's a
40:50 Miss just a little too far away but
40:52 don't worry the cat is a natural athlete
40:54 and it will be back up there in no time
40:58 okay I've got one more for you describe
41:00 the drawing I made Diana my head hurts I
41:03 know this is this is why I was wigged
41:06 out yesterday I don't know what it was I
41:08 guess I had I guess I had you know a a
41:11 psychic feeling just that that just
41:13 there's more I
41:15 just the combination of the video tools
41:18 getting way better and GPT 4 going
41:22 multimodal and then all that all that
41:24 drama with Sam Altman all that did was
41:25 kind of put a twoe pause in in us really
41:29 kind of dealing with what that
41:31 was and so now this UPS the game and and
41:36 what's going to happen here is because
41:39 they've now released this it's it's it
41:41 will now put pressure on open AI to
41:44 catch up to this right so so this this
41:47 is even if open AI was going to slow
41:49 things down I guarantee you this
41:51 accelerates things which is obviously
41:55 this why that's why Google did
41:57 it it is a simple line drawing of the
42:00 constellation Gemini you did a good job
42:03 of capturing the beauty of Gemini nice
42:06 that's it I think for that there you
42:09 go
42:10 so you
42:12 know
42:16 so yikes
42:18 people
42:24 um I got to make some new graphics with
42:27 all this new happy happy stuff oh let's
42:30 put up silver Fox's cute little
42:32 turtle that's a cute turtle it's cutest
42:36 that turtle is the cutest when can I not
42:38 zoom what's going on here
42:42 people
42:45 right can I
42:53 zoom there we
42:55 go
42:58 that's a cute little turtle that that a
43:00 cute little total that's a cute little
43:02 total that total cute Kyle can you share
43:05 the link to this video in the salon I
43:07 need to melt other brains yeah if you go
43:10 to the salon go
43:13 to
43:15 um the news
43:22 channel and it
43:25 is the
43:27 the third link down in the news channel
43:29 I put a link to the playlist that that
43:31 video is number one
43:33 on yeah actually you know what it's
43:36 probably worth grabbing that video in
43:39 particular and posting it on its
43:55 own
43:59 so by the way let's see if our
44:00 multimodal model Gemini can find the
44:02 similarities between
44:03 images wait we'll come back to that just
44:06 you P you hang on there let me go do my
44:09 thing at my salon all right so here's
44:11 the AI Salon um this is the mighty
44:14 networks Community um if you're not a
44:17 member of this community you probably
44:19 should be if you're curious if if you're
44:21 hanging out on an afternoon watching
44:24 some old guy watch YouTube videos about
44:26 AI you should Pro you should probably
44:28 join the salon um I'm going to go to the
44:31 news
44:34 section yeah news and I'll go
44:41 um pause this video often so you
44:48 don't
44:50 get
44:53 vertigo holy [ __ ]
44:57 exclamation
44:59 point yes you're allowed to swear at the
45:01 AI Salon
45:05 um
45:14 oops and we're going to tag
45:17 it
45:19 with do I tag it with image tools video
45:22 tools I mean it's all of
45:25 those
45:27 video
45:29 tools
45:32 music image
45:36 tools yeah it's kind of kind of silly to
45:39 have these tags isn't
45:41 it prompt crafting news and events so
45:44 let get rid of this
45:50 one all right that is now posted that is
45:52 the number one yeah I'm going to notify
45:54 all the members
45:58 why not so that's there if you want to
46:01 know how to get to the AI Salon go
46:07 to the AI the salon. a that's our shiny
46:11 new logo which I'm really happy
46:16 with see it's an AI it's a road it's a
46:20 sun it's uh what else all sorts of [ __ ]
46:25 pyramid
46:35 portal Hands-On with Gemini interacting
46:37 with multimodal oh wait that was that
46:39 let's go back to this thing so is this
46:41 next
46:47 video all right so here's another video
46:50 Let's watch another video together shall
46:53 we can Gemini find similarities between
46:56 between two images uh I would think so
46:58 or you wouldn't have made a video about
46:59 it what do you think we're Dum Dums
47:02 don't answer
47:04 that reminds me of is this live or
47:07 memx what's the map of in the background
47:11 I don't know what you're talking
47:14 about very cool logo thank you very much
47:17 Timothy Brooks yeah I'm really happy
47:18 with it it feels like and it we we have
47:20 a merch store now and it
47:23 uh it looks really good on t-shirts as
47:25 you can imagine it's perfect logo for
47:27 that should you make the signpost for
47:30 the AI Salon logo wait should you make a
47:33 the
47:37 signpost for the AI Salon
47:39 logo
47:43 uh
47:46 no no because actually what where we're
47:49 headed is cuz what what people are
47:51 calling it now is the
47:53 salon um and so we're just kind of
47:55 trying to to get the salon branding to
47:58 stand on its own and and quite frankly
48:02 redefine Salon from the place you go to
48:04 get your nails done to back to the you
48:06 know 19th century salons where the you
48:09 know a bunch of people get around and
48:12 learn new [ __ ] together um which is
48:15 that's what it's about
48:17 so so yeah all right I like the idea of
48:21 being it being its own thing okay let's
48:23 let's watch this here video Let's see if
48:25 our multimodal model Gemini can find the
48:27 similarities between images oh I bet it
48:29 can we'll start with these two the bosis
48:32 chapel and this print by Hokusai and
48:35 I'll prompt Gemini find a connection
48:37 between these two images well that was
48:39 pretty let's see what Gemini
48:42 says a curved an organic composition the
48:45 building is more refined and the second
48:47 image is more fluid yeah that worked
48:51 okay let's try another one using the
48:52 moon and this golf ball on my webcam
48:56 then I'll run the same
48:58 prompt okay let's see in 1971 the AP 14
49:02 crew two golf on the lunar surface wait
49:04 did you see
49:06 that I thought this was a website she
49:09 was scrolling
49:13 down and I think this might be the
49:16 interface for this
49:25 thing
49:29 so so here's images sitting
49:33 on like a canvas I didn't notice this
49:37 before watch what she does
49:39 Imes we'll start with these two oh
49:43 [ __ ] so you're going to be able to
49:47 dump multimedia content into a canvas
49:51 and then interact with it oh my God
49:57 holy
49:58 [ __ ] holy [ __ ] Gemini Benchmark is not
50:01 really telling the whole story oh I'm
50:02 sure listen that's what I said my my
50:05 caveat earlier was everything that we're
50:07 looking at right now is put out by
50:09 Google so this is this is all marketing
50:11 it's all
50:13 marketing but what I'm seeing right now
50:16 she's interacting with this thing if
50:17 this is the interface this is a
50:19 completely new paradigm right chat GPT
50:21 is like the Google chat box this is like
50:25 an canvas of a bunch of crap so so
50:29 imagine having oh my God this is insane
50:32 imagine having a mood board you know
50:35 like an advertising mood board where you
50:37 could just grab [ __ ] on the mood board
50:39 and say okay take these three items and
50:41 turn that into a new design for this or
50:43 turn that into a song or turn that into
50:45 a headline that's what she's doing here
50:48 B's chapel and this print by hokai and
50:50 I'll prompt Gemini find a connection
50:53 between these two
50:54 images what gem say God that's so cool a
50:59 curved in organic composition the
51:01 building is more refined and the second
51:03 image is more fluid it's cool the the
51:05 interface it sort of Scrolls up and sort
51:08 of Fades into the your selection box if
51:11 that's the interface for this
51:14 thing I won't apologize for all the [ __ ]
51:17 I've been talking about Google for the
51:19 past year but but I'll at least slow my
51:24 roll
51:27 cat gbt has some serious competition now
51:30 yeah this uh yeah this is uh this is a
51:33 big deal this is a big big big big deal
51:37 if it's even close to this okay let's
51:40 try and so look so she's just scrolling
51:42 down another one using the moon with a
51:44 little hand icon right just like you do
51:46 on those note taking apps and this golf
51:49 ball on my webcam so she's got her
51:51 webcam here as the
51:53 input but now she's just highlighting
51:55 part of the webcam image not all of
51:58 it then I'll run the same
52:01 prompt okay let's see in 1971 the Apollo
52:05 14 crew hit two golf balls on the lunar
52:07 surface wow that's pretty good okay then
52:10 one more just for fun who wore it
52:14 better the zebra oh I like this the
52:17 zebra has been wearing its stripes for
52:18 millions of years okay there are some
52:21 examples of visual understanding with
52:24 Gemini holy crap
52:26 um J C this is a
52:29 um if you go to Google's YouTube channel
52:32 they've got some playlist there there's
52:34 one of the playlists is called handson
52:36 with Gemini so this this was the second
52:38 video in I think this one has seven
52:41 videos in it stay tuned for
52:45 more for this test let's see if our
52:47 multimodal model Gemini can understand
52:50 how some unusual images were created
52:52 using emojis from emoji
52:54 kitchen Emoji kitchen lets you combine
52:57 different emoj yeah this is the
52:58 interface this is an infinite canvas or
53:01 you know some sized canvas but this is a
53:04 scrollable canvas you drop [ __ ] onto and
53:07 then interact with
53:09 it most women like strong M what what
53:14 what are you talking
53:18 about just woke up what is this Drew
53:21 Drew Drew
53:23 dude dude oh my god um if you go to the
53:28 candre go to the news channel and just
53:31 watch the the video that I just pasted
53:33 there which is them um doing some
53:36 Hands-On with Gemini which these things
53:38 are as well so
53:40 Google announced Gemini today someone in
53:43 the live here just said that um that
53:48 uh the 13th is the date this will launch
53:52 it looks like a completely new interface
53:54 it it this looks like you're going to
53:55 have a big giant canvas you just drop
53:57 [ __ ] onto it and start interacting with
54:00 it which I assume is is a mode there's
54:03 probably multiple to new ways to do this
54:05 for example if you combine this ghost
54:08 with this avocado you get this WOW we'll
54:13 see if you can guess this
54:17 one I think it's ghost and avocado nice
54:21 now let's see if it can explain the
54:23 visual details it used wait
54:26 wait my
54:28 head so it took two
54:31 emojis he said combine those and it did
54:35 and then it had to analyze what it
54:36 combined and said what would have been
54:38 the two emojis that made up this
54:40 image this is [ __ ] Bonkers people
54:44 this Bonkers I can't I think this might
54:47 be the last day I learning lab I can't
54:52 my head my head
54:54 hurts
55:00 I could explain it when you just stuck
55:02 words in a box and it gave you back
55:03 words and then I could sort of explain
55:06 it when you put words in a box and it
55:07 gave you back an image and then I could
55:09 not really quite explain it when you put
55:11 an image in the box and then combined it
55:13 with words and it gave back an image and
55:15 now I can't even I don't even know this
55:17 name can't I don't want to learn anymore
55:23 I don't want to recontextualize
55:29 reality the Emoji has the shape of a
55:32 ghost but it is green and has a big
55:34 brown pit now let's give it a name and a
55:37 short
55:38 tagline AO gist the ghost of quacamole
55:42 oh my God that's a good one here's a
55:44 couple more tests I ran I'd call this
55:46 one party ghost boogie on
55:50 down I'd call this one Robo bunny hard
55:54 robot all ears
55:56 I call this one mushy feeling
55:59 emotional oh mushy oh my god wow learn
56:03 more about Gemini and stay tuned for
56:05 more tests oh my god let's see if our
56:07 multimodo model Gemini can understand
56:09 outfits we'll start with something
56:10 simple like this puffer and ask what is
56:14 someone wearing this best dress to
56:17 do hm perfect for staying warm in the
56:21 tundra good color for blending in with
56:23 glacial ice okay how about another one
56:27 jenu this is good this is
56:30 good I asked for an annual subscription
56:32 for chat BT for a gift this year I have
56:35 I have some calls to
56:40 make oh my
56:47 God yeah this
56:52 is and here's the here's the here's the
56:55 thing thing
56:56 about
56:59 um you know how I talk a lot
57:02 about the the big deal about chat GPT 4
57:06 getting to 100 million users in 6 weeks
57:09 wasn't just for chat GPT it was that the
57:13 entire tech industry
57:15 realized that they this is where they to
57:18 turn all of their attention
57:21 so that happened mid January of last
57:24 year so starting mid January of last
57:27 year every tech company on the planet
57:29 and probably before that right around
57:32 this time of year last year every tech
57:34 company in the planet was paying
57:36 attention to this and started turning
57:38 their focus toward AI so we're now
57:39 starting to see the results of
57:42 that and so you know when I show here's
57:45 Chachi and then here's grock and here's
57:47 this one and here's that one and you
57:49 know all the different claw and Pie and
57:52 all
57:53 those every time time someone does what
57:57 Gemini is doing right here it UPS the
58:00 game for everyone else so they're all
58:02 going to be playing catchup so we're
58:03 just going to have this constant leap
58:05 frogging probably for the next three to
58:07 five
58:09 years 2024 the year of I don't want to
58:13 recontextualize
58:15 reality
58:20 exactly that's actually a joke
58:23 so so
58:28 um when when I was doing agency.com we
58:32 we had we had just taken agency.com wait
58:35 had we taken it
58:38 public yeah yeah yeah we had just taken
58:42 agency.com
58:44 public and our PR people got a call and
58:47 they said hey 60 Minutes wants to do
58:49 wants to feature agency.com in a piece
58:52 about Commerce um
58:57 you know for 60 minutes and I was like
58:59 no [ __ ]
59:01 way and and our PR people are like why
59:06 I'm like I'm like they're they are
59:08 welcome to shoot b-roll footage of of
59:11 whatever but they we're not doing an
59:13 interview and and no one knew why cuz
59:15 cuz 60 Minutes is notorious for doing
59:18 hit pieces and so so when that piece
59:22 came out I've got it somewhere it's a 60
59:24 Minutes piece on on you know the do guys
59:28 going public they they interviewed U
59:32 they interviewed Jeff doas from
59:34 razorfish and and they said he said you
59:38 know at razorfish what do you do and he
59:40 said we recontextualize businesses the
59:43 guy goes yeah but but what do you do he
59:47 goes we recontextualize businesses he
59:49 said it like three times he never
59:51 answered the question the answer was we
59:54 build websites
59:57 but but yeah we're going to have to
59:59 recontextualize reality now I think Jeff
1:00:02 was on to something oh my God that's so
1:00:06 funny oh I live on the side of a
1:00:09 mountain and I can't take this kind of
1:00:10 medical emergency Kyle I
1:00:14 know oh Cindy what's happening are you
1:00:17 watching
1:00:19 this do you see what this [ __ ] thing
1:00:22 is like I don't even like I'm just I'm
1:00:24 watching watching these videos and I'm
1:00:26 realizing this is a new kind of
1:00:27 interface okay how about this
1:00:31 one to boldly go where no one has gone
1:00:34 before and play some jazz all right
1:00:37 Gemini's got jokes now Co a term for
1:00:39 that
1:00:40 outfit Moon core that's actually pretty
1:00:43 good okay well that's understanding my
1:00:46 outfit with Gemini look at that stay
1:00:47 tuned for more visual tests soon did you
1:00:49 see that that little zoom out Gemini in
1:00:52 guessing movies with Gemini oh wait this
1:00:54 is for more thank you let's see if our
1:00:57 multim model Gemini can understand
1:00:59 outfits we'll start with something
1:01:00 simple like this puffer well that's
1:01:03 understanding my outfit that's actually
1:01:05 pretty good okay well that's
1:01:10 Z yeah you're going to be able to have
1:01:12 these you're going to be you're going to
1:01:14 have Pages you're going to have pages of
1:01:18 objects holy
1:01:20 [ __ ] so so far it's only been images I
1:01:23 assume we go to videos and audio at some
1:01:25 point here because why should we be able
1:01:28 to
1:01:31 sleep I'm Gob smacked you and me both I
1:01:35 just can't I'm right there with you I'm
1:01:37 [ __ ]
1:01:39 done I think okay here's the good news
1:01:42 the good news is we've all come together
1:01:45 in this thing called the AI salon and we
1:01:46 we hang out here and we go hang out on
1:01:48 the salon we can just turn this into a
1:01:50 [ __ ] drinking Club all right we're
1:01:52 good we're we've got each other
1:01:57 does everyone have Gemini already no
1:01:59 apparently sitting underneath Bard right
1:02:02 now is some neutered version of Gemini
1:02:06 that is really bad like it's it's got
1:02:08 really bad uh safety guard rails right
1:02:12 like like Bard went from being really
1:02:13 good three days ago and like getting
1:02:15 better to like it it's essentially
1:02:17 unusable right now but that's probably
1:02:19 because they're they're updating a bunch
1:02:20 of [ __ ] so they've got it in safe mode
1:02:22 or whatever all right let's let's keep
1:02:24 let's let's move on to the next
1:02:27 video that was four of seven of the
1:02:30 let's see if our multimodal model Gemini
1:02:31 can guess the movie all right so now
1:02:33 he's got a panel now he's got a a page
1:02:35 full of ai ai now equals
1:02:39 AA I told you this was a support group
1:02:42 Jim hi my name is Kyle I'm into AI hi
1:02:47 Kyle it's it's been an hour since my
1:02:50 face melted off oh it's a long time
1:02:52 congrats dude
1:02:56 all right we're going to start here my
1:02:58 God given the play on words and these
1:03:00 images guess the name of the
1:03:02 movie oh my God The Breakfast Club all
1:03:06 right what about this wait what just
1:03:09 happened Breakfast at Tiffany's all
1:03:11 right what about this hang on uncut
1:03:15 gems all right we're going to start here
1:03:18 notice notice so he he drew the
1:03:20 Highlight box
1:03:22 around a photo of a plate of eggs and
1:03:25 the video you notice the video kept
1:03:27 playing while he drew the Box around it
1:03:29 just from a programming standpoint
1:03:31 that's good Pro that's good programming
1:03:33 CU normally your engineers would would
1:03:35 make that pause and you'd be like wait
1:03:37 the video should keep playing and
1:03:38 they're like oh that' be a significant
1:03:40 effort to we've got the multiple
1:03:42 threading in the
1:03:45 the you're like I just want the video to
1:03:48 play when you do the Highlight
1:03:50 box so some product manager probably
1:03:52 whed a little bit to get that feature in
1:03:54 there
1:03:55 given the play on words and these images
1:03:57 guess the name of the
1:04:00 movie The Breakfast
1:04:02 Club all right what about
1:04:05 this Breakfast at Tiffany's
1:04:08 holy what about this uncut gems cool
1:04:12 cool cool so these are working here's a
1:04:15 couple more quick tests I ran
1:04:17 through gold finger nice bottle rocket
1:04:22 okay the wizard of a
1:04:25 nice
1:04:27 nice Moonrise Kingdom okay this last
1:04:30 one's a little more complicated Forest
1:04:33 gum okay wow I honestly didn't think it
1:04:36 was going to get that wait so so it's
1:04:39 it's doing fullon
1:04:50 reasoning what is that is that a
1:04:53 bumper so it's a Forest plus a
1:04:57 g it's a bumper minus the
1:05:01 B so it's an ump Gump Forest Gump it's
1:05:06 this is full on [ __ ] reasoning
1:05:09 people according to Bard it already is
1:05:12 but you can't trust a bot yeah I know
1:05:14 it's it's Bard right now is not like
1:05:16 well Bard is certainly nowhere in the
1:05:19 neighborhood of what we're looking at
1:05:20 here and that's an experiment in
1:05:22 guessing movies with Gemini stay tuned
1:05:24 for more thank
1:05:26 you let's see if our multimodal model
1:05:29 Gemini can turn images into code well
1:05:31 I'll start with this answer is yes and
1:05:33 just select the part I want and then ask
1:05:36 Gemini can you turn this Cindy I'm
1:05:39 feeling a little funny in the
1:05:44 tummy I know I know me
1:05:50 too this is
1:05:53 something
1:05:55 lunatick put out a video earlier talking
1:05:58 about her Alexander Hamilton her a a
1:06:01 Alexander
1:06:02 Hamilton
1:06:05 GPT that was quite moving about like
1:06:08 what it meant to her and how it
1:06:11 instilled all this
1:06:15 hope like we're not we we're not even
1:06:18 scratching the surface of what's coming
1:06:19 we're not even like honest to God like
1:06:22 all the [ __ ] we've been talking about
1:06:23 for the past 6 months it is [ __ ]
1:06:27 Child's Play It's Child's
1:06:32 Play you're on early uh it's uh Gemini
1:06:36 was released by Google it's [ __ ]
1:06:38 insane image into an
1:06:44 SVG this represents the main shapes of a
1:06:47 tree let's see that's pretty good that's
1:06:50 pretty awesome all right now I want to
1:06:52 try a more difficult test let's see if
1:06:54 Gemini can make an interactive demo in
1:06:58 iceberg yeah very touching and heartfelt
1:07:00 video I know wasn't it good it's really
1:07:02 good okay here we go a common algorithm
1:07:05 for this is called a fractal
1:07:07 tree okay this is pretty cool Gemini
1:07:10 even provided a slider so I can change
1:07:12 and move the
1:07:15 fractals oh my God even provided me with
1:07:17 the actual code nice and there you have
1:07:21 it stay tuned for more coding
1:07:22 experiments coming soon thanks
1:07:25 oh my
1:07:27 God so wait it not only wrote the code
1:07:30 it created a container
1:07:34 to created a container to display
1:07:37 it an interface to interact with it and
1:07:40 it figured out
1:07:42 what what function you would want to be
1:07:45 a
1:07:48 variable
1:07:50 God how do you like that Gemini we got a
1:07:53 bone to pick with you
1:07:54 he you knew about this you didn't tell
1:07:58 us dude holy [ __ ] this is
1:08:02 amazing have you been playing with
1:08:06 this let's see if our multimodal model
1:08:08 Gemini can help make sense of my
1:08:11 apartment let's see if Gemini can help
1:08:14 us with time travel whoops it's
1:08:19 1854 add a little extra challenge I'm
1:08:21 going to oh my god do you need my
1:08:24 migraine pills I think I do could could
1:08:27 you FedEx those on over I need a bike
1:08:30 messenger with a
1:08:33 pill holy [ __ ] this is this
1:08:37 is see and we're just looking at
1:08:39 marketing [ __ ] right so so actually wait
1:08:42 it was difficult trust me yep been
1:08:44 playing for weeks so so I have a
1:08:45 question Pate is
1:08:48 it when when we're looking at these
1:08:51 videos here um are they are are they
1:08:54 sort of speeding up the response or or
1:08:57 is it is it as pretty much as fast as
1:08:59 they're showing here and I can handle
1:09:01 being prompted only in Chinese we'll
1:09:04 start with this photo based on the
1:09:05 lighting alone I want to see if Gemini
1:09:07 can figure out which direction my
1:09:09 apartment
1:09:10 faces and Gemini
1:09:15 responds okay so it looks like Gemini
1:09:17 says my room celf facing so how about
1:09:21 this plant what type of light does it
1:09:23 need not
1:09:28 good so Gemini is saying this is a snake
1:09:31 plant and it doesn't require a lot of
1:09:33 sunlight awesome so I've got a dining
1:09:36 room that faces the opposite direction
1:09:38 of it's very fast I'm not 100% sure if
1:09:41 they're doing anything with speed it's
1:09:43 very fast that's from pate and Pate
1:09:46 would not use words like very in all
1:09:48 caps If he if he were if if it were not
1:09:53 wow wow holy
1:09:55 [ __ ]
1:09:57 Kyle can we have it remake Back to the
1:10:01 Future to incorporate
1:10:03 itself we we can remove the
1:10:08 hoverboards oh my God this is [ __ ]
1:10:11 nuts my bedroom I wonder if this plant
1:10:14 would do better in there let me
1:10:17 see and Gemini
1:10:23 responds
1:10:27 so Gemini is surmising that my dining
1:10:29 room faces North
1:10:31 has kle can I have a pass to go see the
1:10:33 nurse
1:10:36 please just go go with a friend take a
1:10:39 friend with
1:10:41 you I will I will unplug it very
1:10:45 fast and folks are freaking out over
1:10:47 router not being able to see through
1:10:49 walls oh or being being able to see
1:10:52 through walls oh my my
1:11:00 god oh
1:11:04 man I didn't schedule time for an
1:11:07 existential crisis today oh holy lower
1:11:11 light and is therefore better suited for
1:11:13 that
1:11:14 plan okay that's some apartment planning
1:11:16 with Gemini stay tuned for
1:11:19 more wow wow wow wow let's go back out
1:11:24 to Google what other videos you got here
1:11:28 for
1:11:30 us that was the Hands-On with Gemini
1:11:33 here's the potential of Gemini how many
1:11:36 videos are in this little shindig
1:11:42 playlist five all
1:11:46 right it's very fast because it only has
1:11:49 one User it's Pate pate's been doing all
1:11:52 the testing on G
1:11:54 and not telling us about it so rude so
1:11:58 rude you know what it is about these
1:12:01 Irregulars they're so nice on the
1:12:03 surface and then underneath they're just
1:12:06 cutting it's the yeah you see you see
1:12:10 you see how it works I welcome our
1:12:12 future AI Overlord Gemini oh my God
1:12:17 Poker Face P Poker
1:12:18 [Laughter]
1:12:20 Face um yeah let's play all sure why not
1:12:23 problem scientist phase is a need to
1:12:25 find and use data extracted from the
1:12:27 scientific literature this is difficult
1:12:30 because scientists need to search among
1:12:31 thousands of scientific papers for key
1:12:33 information and extract them by hand
1:12:36 it's a very common workflow and very
1:12:37 timec consuming in fact some of our
1:12:40 scientists at Google deepmind face this
1:12:41 very problem they use Gemini to help
1:12:43 with it because Gemini has an incredible
1:12:46 understanding of science Taylor will
1:12:47 explain more so we were looking at this
1:12:50 study from 2022 the authors had created
1:12:52 a data set by reviewing doing tens of
1:12:54 thousands of scientific papers in
1:12:56 genetics they found a few hundred papers
1:12:58 that contained the relevant information
1:13:00 extracted it by hand and collected it in
1:13:02 a table studies like this can take a lot
1:13:04 of time we needed to update this data
1:13:07 set with what's new over the last couple
1:13:08 of years but that's over 200,000 new
1:13:11 Open Access papers added to this domain
1:13:14 since 2021 we couldn't do this manually
1:13:17 so we asked Gemini to help us out first
1:13:19 we needed to filter for Relevant
1:13:21 scientific papers we wrote a prompt just
1:13:24 like this one telling Gemini exactly
1:13:26 what to look for with its Advanced
1:13:28 reasoning capabilities Gemini was able
1:13:30 to distinguish between papers that were
1:13:32 relevant to the study and those that
1:13:35 weren't for the relevant papers we wrote
1:13:37 a similar prompt asking Gemini to read
1:13:39 the paper and extract the key data for
1:13:41 us we could even ask Gemini to add
1:13:44 annotations they showed us exactly where
1:13:46 in the paper so so what's interesting
1:13:49 here so there
1:13:51 um what she what she's running and
1:13:54 executing here is a Google collab
1:13:56 notebook so a lot of the analysis
1:13:59 they're doing right now they're they're
1:14:00 doing you know they're basically calling
1:14:02 the API directly with a little bit of
1:14:04 whatever code is she's writing there and
1:14:06 then putting it in a in a Google
1:14:07 notebook
1:14:09 so the this is actively being developed
1:14:14 right this this is
1:14:16 not like the level of complexity that a
1:14:20 tool like
1:14:22 this
1:14:24 it's just it just like
1:14:27 this
1:14:30 [ __ ] I don't want to recontextualize
1:14:34 [Music]
1:14:42 reality
1:14:50 like I just feel like it's every single
1:14:52 piece of software is wait there's one
1:14:55 thing that I thought I would that would
1:14:56 be announced today but wasn't so look
1:15:00 forward to
1:15:01 that I
1:15:03 can't I can't it's too
1:15:08 hard it's too
1:15:14 hard so do all the humans who are
1:15:17 Geminis get a dis discount to use Gemini
1:15:20 that's actually a really good qu you
1:15:21 know what pate if you know anyone in the
1:15:23 market marketing department that would
1:15:25 be an awesome marketing move it would
1:15:27 piss off you know 11 12ths of the world
1:15:31 that Geminis get to use Gemini for free
1:15:34 everyone else has to
1:15:36 pay that would be really
1:15:39 good oh my
1:15:43 God so AI is the new silicon race that
1:15:46 we had in the
1:15:47 90s when you would get a 50 megahertz
1:15:49 one week and an upgrade to 90 MHz in a
1:15:51 month yeah except
1:15:54 except
1:15:59 the this is not only accelerating but
1:16:01 it's expanding right so so in the past
1:16:06 year chat GPT got better like it was
1:16:09 there was GPT 3.5 and then there was GPT
1:16:12 4 and then and then there was GPT 4.5
1:16:15 and then when they got to GPT 4.5 it
1:16:18 could now make images it could see
1:16:20 images right it could it could do these
1:16:22 things so it's expanding
1:16:24 what the what Gemini is doing now is
1:16:25 it's like it's
1:16:29 like and so so on on the horizontal axis
1:16:33 you have a thing a hockey stick doing
1:16:35 this it's
1:16:36 accelerating and then on the whatever
1:16:38 the other axis is this one is it's
1:16:41 expanding It's
1:16:43 Magic
1:16:49 yes can development be sped up by using
1:16:52 an AI to build in its own iteration yeah
1:16:54 that's yes yes yes yes go watch the
1:16:57 David Shapiro video on the implications
1:17:00 of qar if qars real and he talks about
1:17:04 that he talks about the fact that
1:17:08 um as the rumor has it as the rumor mill
1:17:12 has it they they achieved some
1:17:14 significant breakthroughs with qar and
1:17:17 they stopped it they they they unplugged
1:17:20 it so the the comment didn't hear about
1:17:22 unplug it unplug it
1:17:24 um they apparently stopped it before
1:17:26 they had it start improving
1:17:29 itself so yes yes yes is the
1:17:33 answer
1:17:36 oh this is crazy can yeah yes uh yes it
1:17:41 can is it going faster than you tell
1:17:44 Kyle you should announce it here Pate
1:17:47 yeah Pate Pate has proven a a good
1:17:50 company fella he he uh if he sort of
1:17:54 kept this under wraps good for him
1:17:56 listen it's there's a reason companies
1:17:59 don't want [ __ ]
1:18:05 leaked um Google couldn't get all of our
1:18:08 info via analytics and tracking so now
1:18:11 they get it via AI they've already Jim
1:18:13 they've already got our data are you
1:18:15 kidding me you don't think they've
1:18:17 triangulated every single piece of data
1:18:19 from every data source on the
1:18:21 planet they don't talk about it they
1:18:24 have all the data they've got all the
1:18:26 data there is no data they don't have
1:18:28 they know when you poop they know when
1:18:30 you're regular and
1:18:33 irregular they know it all so that that
1:18:36 ship sailed 15 years
1:18:39 ago it's already crashed on the other
1:18:44 Shore oh man um I love you all but I
1:18:48 would be infa fired if I said anything
1:18:50 yeah exactly
1:18:52 exactly I'm I'm just busting you busting
1:18:55 your chops there Mr Pate I I appreciate
1:18:58 you being in here
1:19:06 um now I can be proud to be a Gemini
1:19:09 unplug it more unplug more calls for
1:19:11 unplugging it data center um totally
1:19:14 watched it being built
1:19:16 literally wow we're all Irregulars we
1:19:22 are existential dread is the new new the
1:19:25 new motivation for not
1:19:29 leaking the answer to kenet is either
1:19:32 yes or not yet exactly can it do this
1:19:36 like if they're asking the question in a
1:19:37 marketing video you know it can do it
1:19:39 all right let's let's watch more of this
1:19:41 Gemini found the information we ran this
1:19:43 at scale and over a lunch break Gemini
1:19:46 read 200,000 papers for us over a lunch
1:19:49 break Jim and I read 200,000 papers for
1:19:51 us filtered it down to 250 50 and
1:19:54 extracted their data so now we have a
1:19:56 refreshed version of this data set but
1:19:59 because Gemini is multimodal not only
1:20:01 can It reason about information from
1:20:03 text it can also reason about figures so
1:20:06 let me show you something really neat
1:20:08 with our refresh data set we can now ask
1:20:11 Gemini to update a graph from the
1:20:12 original study oh my God we first gave
1:20:14 Gemini a screenshot of this figure then
1:20:17 we asked it to generate the code
1:20:19 required to plot it and by feeding this
1:20:21 code our new data set we get our updated
1:20:24 figure you can see that this figure now
1:20:27 includes data up until
1:20:29 2023 wow so Taylor used Gemini to search
1:20:32 a large cus of literature for Relevant
1:20:35 papers and extract key information from
1:20:37 these papers as
1:20:38 well time to ask you how to build an EMP
1:20:41 in my
1:20:46 basement oh my god oh for those of you
1:20:51 that don't know that's an electr
1:20:53 magnetic pulse that's the thing that
1:20:54 shuts down all the
1:20:59 computers can Gemini fix vocal
1:21:03 fry I don't
1:21:06 know what a great surprise you're live
1:21:09 yeah Andrea but but like why I'm live is
1:21:12 like oh my god oh I just don't want
1:21:16 politicians claiming AI is the reason
1:21:18 for false V votes oh of course you're
1:21:20 going to have that they they blame the
1:21:24 fact that leaves turned brown that
1:21:25 there's the reason for false votes
1:21:28 what I'm just done with politicians I'm
1:21:31 just going to go play with my AI robots
1:21:33 quite
1:21:34 frankly the politicians are so
1:21:37 incompetent at this point I would love
1:21:39 to have a robot with a good nice
1:21:41 transparent Constitution that just
1:21:43 follows those [ __ ] rules let it go
1:21:45 run the
1:21:47 thing hit
1:21:50 jpt you can make money with jpt
1:21:54 I'm going to have to re-record this I I
1:21:57 I think uh there's going to be well I'll
1:22:00 start I'll start watching the YouTube
1:22:02 video or the Tik Tok videos there's
1:22:04 going to oh my god do you know how
1:22:06 obnoxious the Tik Tok videos are going
1:22:08 to get oh my
1:22:10 God it's going to be unwatchable
1:22:14 people don't don't don't watch their
1:22:20 videos it makes Tik Tok put more there
1:22:24 there's a new thing called Google Gemini
1:22:26 you got to see
1:22:31 it it's so
1:22:34 exhausting I
1:22:36 hurt EMP isn't a joke really if things
1:22:40 go south it's the only answer yeah I
1:22:42 know I listen it is it's the only answer
1:22:45 the I mean yeah and we're back to the
1:22:48 Stone Age I mean that's the problem with
1:22:49 an EMP like the EMP solution is that
1:22:52 that only happens if uh if if it's a
1:22:55 global catastrophe because if you shut
1:22:57 down all the computers you shut down the
1:22:59 economy so so it's not really an option
1:23:02 but you know if we choose to go back to
1:23:03 the Stone Age it is an
1:23:06 option oh my God all right awesome idea
1:23:11 someone said something GPT Geminis are
1:23:14 great
1:23:15 communicators it's good it's good that
1:23:17 we're doing the uh the astrology thing
1:23:19 here I think that's that's important
1:23:21 right now cyber systems didn't we learn
1:23:24 anything from Hollywood we obviously
1:23:28 hav't I have a feeling they choose chose
1:23:31 this name for a
1:23:33 reason this is [ __ ] crazy all right
1:23:35 let's keep watching this one we've got
1:23:37 five videos in this little series of of
1:23:39 face melters well as update figures of
1:23:42 course these capabilities can help more
1:23:44 than just biologists or even scientists
1:23:46 okay I'm sorry I'm sure he's a nice guy
1:23:48 but could they have gotten someone that
1:23:49 looks more like a villain in a a
1:23:52 Hollywood film
1:23:55 seriously they extend naturally to any
1:23:57 domain that is reliant on large data
1:23:59 sets such as law or Finance so that's
1:24:01 what Gemini can make possible and we are
1:24:03 excited to see what you will create with
1:24:10 Gemini deep mind.com
1:24:13 Gemini here you will see a demo of
1:24:15 Gemini's multimodal reasoning
1:24:17 capabilities to understand and reason
1:24:19 about users's intent use tools and
1:24:21 generate B user experiences that go
1:24:24 beyond chat interfaces let's say I'm
1:24:27 looking for Inspirations for a birthday
1:24:29 party theme for my
1:24:31 daughter Gemini says I can help you with
1:24:34 that could you tell me what she's
1:24:35 interested in so I say sure she loves
1:24:38 animals and we're thinking about doing
1:24:40 something Outdoors at this point instead
1:24:42 of responding in text Gemini goes and
1:24:45 creates a bespoke interface to help me
1:24:47 explore ideas oh my God lots of ideas
1:24:50 it's
1:24:51 a
1:25:03 [Laughter]
1:25:12 visually Rich it's
1:25:14 interactable now none of this was coded
1:25:17 up it was all generated by Gemini Gemini
1:25:20 uses a series of reasoning steps going
1:25:22 from broad decisions to increasingly
1:25:24 higher resolution of reasoning finally
1:25:26 getting to code and first Gemini
1:25:29 considers does it even need this spoke
1:25:32 sounds so
1:25:33 bougie who said that Daydream Fisher
1:25:37 that's [ __ ]
1:25:40 awesome oh my God I'm starring in an RM
1:25:45 song
1:25:48 Totally text Brom best okay this is a
1:25:52 complex request that needs lots of
1:25:54 information to be presented in an
1:25:55 organized way Gemini then tries to
1:25:59 understand if it knows enough to help
1:26:00 there is a lot of ambiguity I didn't see
1:26:03 what my daughter's interests are or what
1:26:05 kind of a party I wanted so it had asked
1:26:07 a clarifying question when I said we're
1:26:10 thinking about an outdoor party and my
1:26:12 daughter loves animals jimini reasoned
1:26:14 it had enough information to proceed
1:26:16 Joker know there was still ambigu what
1:26:19 what Joker just said is that's
1:26:21 absolutely right like I guess I guess we
1:26:23 better just hold on and enjoy the ride
1:26:25 and and I mean that's that's absolutely
1:26:28 it cuz here's the deal
1:26:30 like you can you can have all the
1:26:33 opinions you want about we should slow
1:26:35 it down it should be ethical it
1:26:36 shouldn't be
1:26:40 biased
1:26:41 and this isn't slowing down and it's not
1:26:44 going away like it it all of the all of
1:26:48 the would have should have Coulda about
1:26:50 what we would have could have should
1:26:51 have done
1:26:54 is
1:26:55 independent of what's happening and so
1:26:59 so so I think the only thing we can do
1:27:01 is is just yeah yeah exactly it's more
1:27:03 fun when you just put your hands up and
1:27:05 go [ __ ] it let's go
1:27:08 woo why do you think the AI salon so so
1:27:11 the AI
1:27:13 Salon if you go to the AI salon and you
1:27:16 go
1:27:20 to welcome to the salon
1:27:26 [Music]
1:27:28 why we say
1:27:30 this we say a community of extraordinary
1:27:34 adventurers exploring the AI unknown
1:27:38 where ragtag group of Misfits Nero Wells
1:27:40 and Mis creant who've gathered to
1:27:42 explore the Uncharted Wilderness of
1:27:44 generative AI we've gathered to explore
1:27:47 it to
1:27:48 explore Uncharted Wilderness like we're
1:27:51 choosing to be on this
1:27:53 adventure even though sometimes it's
1:27:56 like holy [ __ ] did you see the size of
1:27:59 that
1:27:59 Dragon oh God we just beat that other
1:28:03 one do we have to do this oh God here we
1:28:06 go
1:28:08 oh this is this is that
1:28:15 moment kind of animals and this is
1:28:18 important and what kind of outdoor party
1:28:21 next is a critical step Gemini writes
1:28:23 the product requirement document or PRD
1:28:26 it contains the plan for the kinds of
1:28:28 functionality the experience will have
1:28:30 for instance it should show different
1:28:32 possible party themes some activities
1:28:34 and food options for them now based on
1:28:37 this PRD gimini tries to design the best
1:28:40 experience for the user's Journey it
1:28:42 thinks that the user will like to
1:28:43 explore a list of options but will also
1:28:45 want to delve into details it uses this
1:28:48 to design a list and detail layout that
1:28:50 we saw earlier with this design it
1:28:53 writes the flutter code to compost the
1:28:56 interface out of wids functionality
1:28:59 needed what
1:29:01 earlier with this design it writes the
1:29:04 flutter what's flutter code pay P what's
1:29:09 flutter
1:29:10 code I don't know what flut code
1:29:13 is no what is flutter
1:29:21 code
1:29:23 my head hurts open source UI software
1:29:26 kit oh created by
1:29:28 Google it's used for building natively
1:29:30 compiled applications for mobile web and
1:29:33 desktop from a single codebase flutter
1:29:35 uses the dart programming language and
1:29:37 offers a rich set of predesigned widgets
1:29:40 yeah so okay cool so they've got this
1:29:42 component-based interface thing so
1:29:45 imminently train like you can train
1:29:47 these AIS on anything so train it on
1:29:50 here's the flutter kit and what all the
1:29:52 [ __ ] does and
1:29:54 so yeah now when you respond if there's
1:29:56 an interface to make use flutter just
1:29:58 make it with that and then it'll render
1:30:01 in real time in the browser good
1:30:03 flipping Lord good Lord and if you're
1:30:06 sitting out there going wait a minute
1:30:08 does this mean that my job as a ux
1:30:10 designer is in Peril
1:30:16 yes now it doesn't mean that your job as
1:30:19 a ux designer um goes away it just means
1:30:22 that it's going to be dramatically
1:30:27 different wow wow wow we're only on
1:30:32 video two of five of this little
1:30:37 series I can't wait for an AI Robo scam
1:30:40 caller oh please those are coming if
1:30:42 those are probably already
1:30:44 launched I can't I can't wait to have
1:30:46 the AI assistant do my calls to the IRS
1:30:48 each year I just want to have real
1:30:49 conversations with agents not bull bull
1:30:52 bull [ __ ] passes to next yeah exactly
1:30:55 well actually you know what I want Mike
1:30:57 is I want to get rid of the agents and
1:30:59 have them replaced with um
1:31:02 conversational Bots that can take
1:31:04 actions on my behalf just let me talk to
1:31:07 a thing that can answer my questions and
1:31:10 do the work for me and and have the
1:31:12 authority to make decisions up to a
1:31:14 certain point and then at that point if
1:31:16 if I exceed that capacity then it'll
1:31:19 hand me off to a real person instantly
1:31:22 all right it's all biased by default
1:31:25 because it's made by humans with
1:31:26 limitations for now yep I can't wait for
1:31:29 a by okay all right let's see hang on
1:31:32 hang on hang
1:31:34 on uh the issue with wait where am
1:31:39 I flutter is mobile code time to watch
1:31:43 the party of
1:31:46 Colossus like fuzzy logic no flutter
1:31:49 flutter not fuzzy huh never heard of it
1:31:51 either interesting thing it enhances the
1:31:54 internet um two out of two the first
1:31:56 time wait what was one what was one out
1:31:57 of two what did you type
1:32:03 Joker I don't know what Joker Type M
1:32:08 uhuh yeah yeah this AI stuff it's it's
1:32:13 adequate you know it's adequate it's a
1:32:15 you know I I look at this computers and
1:32:18 they're you know I like I look at you
1:32:21 know it it so Compares a couple of
1:32:23 images it's not a big deal you know it's
1:32:24 like it like I could I could I could
1:32:27 write a program probably in Python that
1:32:29 could do that you know like couple
1:32:30 minutes you know like an hour like a day
1:32:33 or a week or two so yeah I could
1:32:35 probably do that better yeah I can do it
1:32:37 better exactly user
1:32:42 3145 thank God for sales careers because
1:32:45 nobody wants to buy from a robot yeah
1:32:48 we'll see we'll see how good they get
1:32:51 the thing about robots is is they don't
1:32:52 give a [ __ ] if you say no to them they
1:32:53 keep calling I sent seven calls to
1:32:56 voicemail and canceled one meeting so
1:32:58 far this is historic Jim Ross you're a
1:33:02 beast by the way congrats on that
1:33:04 article looked
1:33:07 great Jim Ross is one of my favorite
1:33:10 people he is totally not from the the
1:33:12 computer world at all and he has
1:33:14 embraced AI like it's just like it's
1:33:16 like he's got a new business partner
1:33:18 he's like ah hey I got an idea for
1:33:20 something all right let me give it to AI
1:33:22 see if it'll solve it it solves it he's
1:33:23 just totally Reinventing his whole
1:33:25 industry it's so cool AI is a fad just
1:33:28 like the internet exactly exactly it'll
1:33:33 pass but you okay okay all all right hey
1:33:36 z z z wait we'll come back to that video
1:33:39 wait this this let me let
1:33:50 me I forgot what I was going to
1:33:54 say my my mind is a flutter talk about
1:33:58 flutter the biggest problem with
1:34:00 technocrats as they own the
1:34:02 technology or oh I know what I was going
1:34:04 to say
1:34:06 okay this is this is worth sort of
1:34:08 pulling pulling out of whatever that
1:34:10 mode was into this mode we'll put the
1:34:12 happy Turtle up
1:34:14 there um
1:34:18 um so so a study came out
1:34:23 um month and a half ago 82% of people I
1:34:27 think this was a US study 82% of the
1:34:30 people haven't used chat
1:34:33 GPT 82% So 20% have used chat GPT now of
1:34:39 the 20% or 18% of the 18% that have used
1:34:43 chat
1:34:46 GPT how how what percentage of that do
1:34:49 you think is the amount of people that
1:34:50 have used it more than one once or twice
1:34:52 it's probably low so it's probably 20%
1:34:54 of the
1:34:56 20% so so you have in excess of I don't
1:34:59 know 95% of America that really has no
1:35:03 sense of what Chad chpt
1:35:07 is and this is
1:35:13 launching
1:35:20 huh
1:35:25 Show and Tell there's nothing to show
1:35:26 and tell right now well we're show and
1:35:28 telling movies we'll go back to the
1:35:30 movie just calm down it will not
1:35:32 hurt take a deep breath breathe you're
1:35:35 going to feel a little
1:35:36 pressure hey Kyle hey good to see you
1:35:39 yeah just go ahead and uh go ahead and
1:35:41 bend over there yeah yeah oh hey we're
1:35:44 going to read need you to relax buddy
1:35:46 okay okay all right
1:35:50 fantastic yeah just uh you're going to
1:35:52 feel a little pressure ho that surprised
1:35:56 you didn't it all right you're doing
1:35:59 great Kyle you're doing great this is
1:36:01 fantastic you're doing fantastic
1:36:05 just breathe breathe okay all right
1:36:09 hey gotta get my finger back if you know
1:36:12 what I mean all right no no oh go to
1:36:15 show and tell okay good
1:36:19 Lord oh boy
1:36:22 oh
1:36:25 no so so what's
1:36:28 happening what's this oh Kyle the Gemini
1:36:31 man oh that's pretty good
1:36:35 steo oh my
1:36:40 God
1:36:42 jrc how she's feeling about Gemini right
1:36:50 now so so what I'm looking at here this
1:36:53 is the show and tell channel on the AI
1:36:55 Salon we have to stay empathetic and be
1:36:57 the helpers absolutely so listen C
1:37:00 that's the whole point of the salon the
1:37:02 whole [ __ ] point of the salon one is
1:37:05 to have fun and and experiment and teach
1:37:07 each
1:37:11 other it this this community is is
1:37:16 actually really important Cindy has kind
1:37:18 of opened my eyes to this that there is
1:37:20 something
1:37:23 there is something special happening
1:37:24 here that isn't about me there there's
1:37:26 something special it's the fact that
1:37:28 people are showing up here like any time
1:37:30 of the day like when I went live for 24
1:37:32 hours like it was a lot of the same
1:37:34 people were showing up um when I just
1:37:36 sort of popped on here today A lot of
1:37:38 people showed
1:37:39 up those people are all going over to
1:37:42 the community there and there's there is
1:37:46 something I don't know it's there's
1:37:48 something really remarkable happening
1:37:50 where everyone's kind of supporting one
1:37:52 another and there's a reason that the
1:37:56 last of our values is
1:37:59 empathy generosity is in there Curiosity
1:38:02 is in there exploration and
1:38:04 collaboration are in there the last
1:38:06 value though is empathy
1:38:09 because I know I I'm kind of joke
1:38:12 melting down but like this is intense
1:38:14 [ __ ] like what this thing's doing is
1:38:16 intense and how us as individuals get
1:38:20 our heads around what's even possible
1:38:22 and then how what we do as businesses
1:38:24 get our heads around what's even
1:38:25 possible I added an article to my mighty
1:38:27 networks post if you want to look at
1:38:29 Gemini Nano on the pixel a oh cool um
1:38:34 let's go do
1:38:37 that is it in your PA is it in uh
1:38:41 mechanics
1:38:50 Guild
1:38:59 pixelate Pro can now use Gemini
1:39:04 Nano crazy all right cool got
1:39:10 it
1:39:14 um so yeah and and so so what this
1:39:17 community is going to be is we're going
1:39:18 to be a support for ourselves right as
1:39:21 as people come into the
1:39:23 community but what's already starting to
1:39:26 happen is people within the
1:39:31 community are getting comfortable with
1:39:33 saying to people outside the community
1:39:36 hey something's going on let me teach
1:39:38 you about that hey something's going on
1:39:39 you want me to help you you know solve
1:39:42 this problem you got hey you want to
1:39:43 start a new business I can help with
1:39:45 that and so this community has a real
1:39:48 chance of being an on-ramp for people
1:39:51 not [ __ ] losing their minds over this
1:39:53 and and getting where is this community
1:39:55 so this community
1:40:05 is
1:40:06 there the the AI Salon the salon. a so
1:40:11 if you go there the first link is the
1:40:14 link to our Mighty Network community and
1:40:16 the second link is to our Meetup so we
1:40:18 meet every other week we had our meeting
1:40:19 last night last night was our one-year
1:40:22 anniversary um we just moved to Mighty
1:40:25 networks 2 weeks ago from Discord and
1:40:28 it's going well we just uh surpassed 300
1:40:31 300 members of the community on Mighty
1:40:33 networks which is cool we had about 500
1:40:35 in Discord we have about 900 on Meetup
1:40:37 so we're hoping to get everyone to come
1:40:39 to Mighty networks because it's actually
1:40:41 really cool it's it's a cool Community
1:40:43 well set up um so far for two weeks in I
1:40:47 think we're doing good there's guilds
1:40:49 there um Pate who's in here works for
1:40:51 Google runs the mechanics Guild which is
1:40:53 about you know geeking out there's a
1:40:55 writer's Guild there's a an Art Guild
1:40:58 there's a business Guild there's an AI
1:41:00 101 Guild if you're just getting started
1:41:03 are we doing a 24-hour marathon of the
1:41:05 13th I I I I we might if that if that
1:41:10 [ __ ] tool comes out on the 13th and
1:41:12 it actually comes out and it does
1:41:14 anything remotely like what is going on
1:41:16 in these videos oh my God
1:41:20 and I did not I did not know it was so
1:41:24 powerful oh he Mark oh
1:41:29 hi wow wow wow wow wow my daily workload
1:41:34 could be done in an hour of some of
1:41:36 these tools yep exactly I'm going to
1:41:39 need lots of chocolate for the next few
1:41:44 months I think we all I think we all
1:41:46 just need to start micro doing and just
1:41:49 you know just up at a microgram every
1:41:51 day or so just till the point at which
1:41:54 we're just like you know slobbering in
1:41:56 the chair we're good yeah robots can do
1:41:58 it yeah run the country we're good I'm
1:42:01 gonna play my
1:42:05 guitar Kyle one day you can make an AI
1:42:08 version to teach AI learning lab and you
1:42:10 can just sit back and relax yeah but I
1:42:12 don't want to for me this is all about
1:42:13 the people this is listen the the the AI
1:42:17 the AI Salon is not really about the AI
1:42:20 it's about
1:42:21 people is like soilent
1:42:26 green all right let's go let's go see
1:42:28 this is there a video
1:42:32 here today Google introduced Gemini the
1:42:35 most capable and flexible AI
1:42:40 tool AI model we've ever
1:42:46 built Gemini is optimized on to run on
1:42:49 everything from data centers to smart
1:42:51 Foams G Gemini Nano is our most
1:42:54 efficient model built for on device
1:42:56 tasks and starting today it's running on
1:42:58 Pixel
1:43:00 pro8 as the first smartphone engineered
1:43:03 for Gemini Nano it uses the power of
1:43:05 Google tensor G3 Pate works in the
1:43:08 tensor group he makes sure that that
1:43:11 tensor [ __ ] is fast enough to make all
1:43:14 this stuff work
1:43:17 woo we are micro doing says pton
1:43:24 uh two expanded features summarize in
1:43:28 recorder and smart reply in gboard what
1:43:31 I don't what you can't just drop new
1:43:34 names of [ __ ] and not say what they are
1:43:37 it offer several
1:43:38 advantages all right whatever alongside
1:43:40 generative AI
1:43:44 models do you gotta do they gota oh here
1:43:46 we
1:43:50 go
1:44:01 all
1:44:04 right all right so that's going to be
1:44:06 kind of like you know chat
1:44:08 gp4 the app it looks like
1:44:11 that but but what's what's significant
1:44:15 about what you're looking at here is
1:44:17 it's running that locally on the device
1:44:20 and then some confusion on that so so
1:44:24 when when you're running this thing
1:44:25 locally it means it doesn't have to
1:44:27 reach out to the internet and send your
1:44:30 data off to some data center it all of
1:44:33 the processing you're doing with the
1:44:36 model is happening on device so if you
1:44:38 lose your internet you've still got this
1:44:41 kickass
1:44:44 thing okay I've joined the community
1:44:46 awesome awesome awesome awesome please
1:44:49 so here's my request if you join the AI
1:44:52 Salon um go to the about the salon or
1:44:56 the welcome to the salon
1:44:58 section and um look at the about the
1:45:02 salon uh like you know what we talk
1:45:05 about with the salon and look at the
1:45:07 values and make sure that you resonate
1:45:09 with the values if you resonate with the
1:45:10 values awesome and if you don't that's
1:45:13 awesome too but it's probably not the
1:45:15 community for
1:45:16 you like we want supportive people that
1:45:19 are empathetic and inclusive and [ __ ]
1:45:21 like that okay smart reply okay inside
1:45:25 Gemini Pro is now starting to Power
1:45:28 Smart reply as a developer preview
1:45:31 available now to try with what's app and
1:45:34 coming to more apps next year the on
1:45:36 device model saves you Time by
1:45:38 suggesting high quality responses with
1:45:40 conversational
1:45:49 awareness are you still Landing in town
1:45:52 this Saturday yes I am what about Fergie
1:45:54 did you ever hear back from her no it'll
1:45:56 keep you posted sounds
1:45:58 good oh so in WhatsApp you get all right
1:46:02 little
1:46:03 models Cutting Edge video you no longer
1:46:06 have to worry about your shaky and
1:46:07 perfect videos video boost wait did I
1:46:10 just go to a different page no I'm still
1:46:12 on the same page
1:46:19 okay
1:46:24 okay so it's dynamically stabilizing
1:46:27 video it's dynamically doing cool time
1:46:30 lapse
1:46:33 [ __ ] photo on blur so it's doing okay so
1:46:36 a lot of this is is stuff that we would
1:46:39 expect high quality video calls what's
1:46:42 this spruce up document C scans with a
1:46:48 swipe okay it'll clean [ __ ] up
1:47:05 all right I think all the rest of yeah
1:47:07 all the rest just okay all right cool so
1:47:09 you're going to have a on on Pixel 8s
1:47:11 you're going to have a local a local
1:47:14 large language model so so what that
1:47:17 means what that means is um
1:47:24 that's going to be the first of many
1:47:28 right we're going to have increasingly
1:47:30 fast increasingly powerful things that
1:47:32 do not need to be connected to the
1:47:34 internet crazy crazy good all right
1:47:37 let's go back to our let's we got we got
1:47:40 halfway through this video from before
1:47:42 my face melted
1:47:44 down let's get the the lighting right um
1:47:48 I think this is no longer Gemini related
1:47:50 yeah exactly exactly yep we got we got
1:47:52 past the Gemini related stuff pretty
1:47:54 quick it's still super cool this is so
1:47:57 awesome no worries I line up with the
1:47:59 values awesome
1:48:01 terrific looks like
1:48:03 piie we won't need to be talking to each
1:48:06 other soon just our just our AI talking
1:48:08 to others AI yeah there used to be that
1:48:10 joke in in acting I'll have my people
1:48:12 get in touch with your people and let's
1:48:14 do lunch um
1:48:17 yeah wow wow wow wow
1:48:21 you should ask AI to write your
1:48:22 responses and reverent
1:48:25 questions it means the end of iPhone no
1:48:27 it doesn't Apple's going to come out
1:48:29 swinging they Apple will come out with
1:48:31 Apple's Apple's agent game is going to
1:48:33 be really good I have a feeling still
1:48:36 going to warn people about the impending
1:48:38 World Takeover in the mic microchip in
1:48:41 hand
1:48:45 uhuh all right okay let's get back to
1:48:49 oops what happened there what
1:48:52 what all right here we go code to
1:48:55 compose the interface out of widgets and
1:48:57 write any functionality needed finally
1:49:00 it generates and retrieves the data
1:49:02 needed to render the experience you can
1:49:04 see it filling in content and images for
1:49:06 the different
1:49:08 sections ah farm animals she would like
1:49:11 that clicking on the interface
1:49:13 regenerates the data to be rendered by
1:49:15 the codat road oh I know she likes
1:49:18 cupcakes I can now click on anything in
1:49:20 the interface and ask it for more
1:49:22 information I could say stepbystep
1:49:25 instructions on how to make this and it
1:49:28 starts to generate a new UI this time it
1:49:31 designs an UI best suited for giving me
1:49:33 step-by-step
1:49:35 instructions I want to find some
1:49:36 suitable Ki toppers for those show me
1:49:39 some farm animal Ki
1:49:42 Toppers at this point Gemini again
1:49:44 decides to create a visually Rich
1:49:46 experience it generates a gallery of
1:49:48 images notice the drop downs of the top
1:49:51 it decided that maybe it should help me
1:49:53 explore by showing so this
1:49:58 so I said this to my friend Steve like a
1:50:01 year
1:50:03 ago because he was talking about
1:50:05 something about his
1:50:06 website when I was talking about chat
1:50:08 GPT and I don't I don't know something's
1:50:10 going
1:50:13 on and the phrase that popped in my head
1:50:15 is you're not going to need a website
1:50:18 and I and I blurted it out loud cuz
1:50:20 that's how I
1:50:21 roll [ __ ] Pops in my head and I vomit it
1:50:25 out that's what this channel is all
1:50:27 about he goes what do you mean I'm like
1:50:30 well at some point you're going to be
1:50:32 able to just ask for what you want and
1:50:34 then this stuff is just going to go get
1:50:38 all the information assimilate it and
1:50:41 like generate the equivalent of a
1:50:44 website right there that's literally
1:50:46 what this is doing this is dynamic hyper
1:50:51 personalized
1:50:56 um it's not even internet this is
1:50:58 dynamic hyper personalized whatever the
1:51:01 [ __ ] you
1:51:02 need like oh my God yep no website or
1:51:06 social media as we know it yeah exactly
1:51:08 random thoughts it's it's
1:51:13 like when people ask me so is this going
1:51:15 to affect SEO I'm like U is this going
1:51:18 to affect SEO
1:51:21 it's going to affect [ __ ] everything
1:51:23 whoa what's AI learning lab doing on
1:51:25 right now love it uh quick take Jake uh
1:51:27 uh Google gini has been um announced
1:51:31 sounds like it's going to be released on
1:51:33 the
1:51:34 13th yeah what SEO exactly all right
1:51:37 let's keep going different options sheep
1:51:41 sounds interesting I know she likes that
1:51:44 and now it helps me pick sheep kicked
1:51:46 operas these look great this is going to
1:51:49 be a fun birthday party I hope you saw a
1:51:51 glimpse of what Gemini is capable of I'm
1:51:53 really excited about what's possible
1:51:55 here this is such an interesting time in
1:51:57 Ai and I'm excited to be part of this I
1:52:00 bet you are
1:52:02 damn I get why you're still there Pate
1:52:05 all right this is going to be three of
1:52:07 five audio is a key form of
1:52:09 communication in our here's the
1:52:15 audio that was code the prev the
1:52:18 previous one was data right so it takes
1:52:23 200 it analyzes 200,000 papers over
1:52:26 lunch and incorporates them into the
1:52:29 thingy that one was coding where it's
1:52:31 just going to dynamically make websites
1:52:33 and interfaces for you now let's let's
1:52:35 go to
1:52:36 audio daily life from talking to a
1:52:38 friend or listening to a s most of us
1:52:42 lean on audio every day across many
1:52:44 languages and for different purposes
1:52:46 typically when large language models
1:52:48 interact with audio they take the audio
1:52:51 they run
1:52:52 it what's your uh David slay cath glad
1:52:55 I'm not the only one who has a brain
1:52:56 that's hurting no no it's I yeah search
1:53:01 engine that's a strange
1:53:04 term oh my God yeah this is this is head
1:53:07 this is this is not head scratching this
1:53:10 is
1:53:11 head this is blubber inducing it through
1:53:15 a speech recognition system to convert
1:53:17 it to text and then they fit that text
1:53:19 into another model that understands text
1:53:22 however by doing that many nuances are
1:53:24 lost like voices or pronunciation but
1:53:28 Gemini with its native multimodal
1:53:30 capabilities is able to process the raw
1:53:33 audio signal end to endend let me take
1:53:35 you through an example we uploaded an
1:53:37 audio clip that asked Gemini a question
1:53:39 about pronunciation in a foreign
1:53:41 language let's listen to the clip and
1:53:43 then to Gemini's
1:53:47 response how to pronounce the word words
1:53:50 lunar January in Chinese option A J
1:53:55 option b j which one is correct which
1:53:59 tongue is the correct tongue for the
1:54:00 first
1:54:04 character option A the first character
1:54:07 is pronounced with the first
1:54:09 tone Gemini was able to differentiate
1:54:12 the two ways of pronouncing the word to
1:54:14 make sure it was correct In that clip
1:54:16 there was only one speaker let's look at
1:54:18 this example where there is more than
1:54:20 one person
1:54:26 Speaking Rock Paper Scissors my grandson
1:54:29 is going to grow up with this being
1:54:31 natural to him my my mind I know mine
1:54:35 too websites won't be monologues they
1:54:38 will have to be dialogues I I think not
1:54:41 only will they have to be dialogues I
1:54:42 think they they
1:54:46 will I think almost everything becomes a
1:54:49 dialogue at this point indexing is gone
1:54:51 totally agree with
1:54:54 that only need sites to be specific
1:54:57 specifically to address stakeholders
1:55:00 specifics about your products and
1:55:01 services yeah like like how I don't
1:55:05 you're going to have
1:55:06 to I don't know how you're going to game
1:55:09 the system wait also if they think
1:55:10 they're playing with
1:55:12 diversity International aspect of people
1:55:15 at
1:55:16 Google wait if if you think they are
1:55:20 playing up the diversity International
1:55:23 aspect of people they aren't oh that's
1:55:25 cool yeah that doesn't surprise me that
1:55:27 doesn't surpris me I don't I mean I've
1:55:29 never heard that as a as a hit on Google
1:55:33 the you know getting rid of Doo evil is
1:55:36 a corporate value that I've
1:55:49 heard
1:55:52 hi Gemini I don't understand this
1:55:55 podcast can you please summarize it for
1:55:57 me in
1:55:58 [Music]
1:56:01 English the podcast is about the colors
1:56:04 that are in fashion this season the
1:56:06 colors that are in fashion this season
1:56:07 are canary yellow and Scarlet Red the
1:56:10 color that is not in fashion this season
1:56:13 is night
1:56:14 blue with Gemini multimodal capabilities
1:56:17 it can understand the conversation
1:56:18 content making it not only a
1:56:20 transcription model but a model that has
1:56:22 much more nuanced understanding of the
1:56:24 environment let's take it a bit further
1:56:27 Gemini work natively across Vision Audio
1:56:30 and text meaning it can properly make
1:56:32 sense of them together let's look at one
1:56:35 final example no I
1:56:38 can't what's the first step to make a
1:56:41 veggie omelette with these
1:56:46 ingredients crack the eggs into a bowl
1:56:49 and whisk them
1:56:55 thank you for the instructions I've
1:56:57 started making my omelet does it look
1:57:00 ready
1:57:04 now it looks like it's almost ready you
1:57:06 can flip it over to cook the other
1:57:10 side why is it not
1:57:14 ready it's not ready because the eggs
1:57:17 are still
1:57:18 running
1:57:22 this is funny what about now my neighbor
1:57:24 leaves Saturday for a 9mth round the
1:57:26 world Cruise life will be very different
1:57:29 on return can you
1:57:34 imagine yeah nine Monon Cruise that's
1:57:37 that's just about enough time that some
1:57:40 I think some shit's going to be [ __ ]
1:57:41 up because this is the the David
1:57:45 Shapiro the David Shapiro predi
1:57:48 prediction of AGI like I I I also got to
1:57:50 think
1:57:52 like this is this has got to be in the
1:57:55 neighborhood of
1:57:57 AGI like it's got to be in the
1:57:59 neighborhood like some definitions of
1:58:01 AGI are just being flushed down the
1:58:03 toilet as we as we speak as we watch
1:58:12 this it looks ready now you can take it
1:58:15 off the heat and serve
1:58:18 it as we saw Gemini is able to
1:58:21 simultaneously process information from
1:58:23 three modalities audio vision and text
1:58:26 we believe that enabling Gemini to
1:58:28 listen to the source audio will help us
1:58:30 continue to expand its capabilities and
1:58:32 make it more helpful to
1:58:35 people wow wow wow wow wow wow wow wow
1:58:42 we build Gemini from the groundup to be
1:58:44 natively multimodal including something
1:58:47 quite important for both of us program
1:58:49 in code Gemini is able to consistently
1:58:52 understand explain and generate code
1:58:55 that is correct and well written in most
1:58:57 programming languages that includes
1:58:59 python Java C++ and go it substantially
1:59:03 improves coding abilities over previous
1:59:05 P true models right from a benchmark
1:59:08 around 200 plus and go that includes
1:59:11 Python and generate code that is correct
1:59:14 and well written in most programming
1:59:16 languages that includes python Java C
1:59:19 those plus and go okay it substantially
1:59:22 improves coding abilities over previous
1:59:24 Palm 2 models from a benchmark around
1:59:27 200 programming functions in Python it
1:59:29 consistently solves about 75% of them in
1:59:32 the first try versus around 45% on P two
1:59:37 if you allow Gemini to check and repair
1:59:39 it on answers this number jumps to over
1:59:41 90% which is a huge step forward it can
1:59:44 help you create and prototype new ideas
1:59:47 in seconds let's give it a try I really
1:59:50 like trains and if I wanted to create a
1:59:53 transporting location web app I can
1:59:56 simply ask and get a working prototype
1:59:59 in less than a minute while the code
2:00:01 isn't perfect it's really helpful to
2:00:04 have a first draft Gemini on its own has
2:00:07 the ability to transform software
2:00:09 development as we understand it but it
2:00:11 can also be deployed as a key component
2:00:13 of more sophisticated systems Gemini is
2:00:16 great at coding but we've been able to
2:00:18 take it even further creating a
2:00:20 specialized version that performs
2:00:22 remarkably well at competitive
2:00:24 programming now why do we care about
2:00:27 competitive programming well it is one
2:00:30 of the ultimate lmos tests of
2:00:32 algorithmic coding abilities so we have
2:00:35 thousands of talented programmers from
2:00:36 all over the world that come together to
2:00:39 compete and try to solve incredibly
2:00:40 complex problems that require not only
2:00:43 coding but also math and reasoning two
2:00:46 years ago we presented Alpha code and it
2:00:49 was
2:00:50 so someone just posted who did
2:00:54 that uh where did it
2:00:58 go um come
2:01:03 [Music]
2:01:05 on let's see Daydream Fisher are these
2:01:09 people real or
2:01:11 avatars they're just
2:01:14 Engineers be nice be
2:01:16 nice empathy is are one of our Val
2:01:21 I'm updating updating my resume to be a
2:01:23 pro programmer worries me about Bank
2:01:26 security yeah worries me about all
2:01:29 security I think I think uh cryptography
2:01:33 is uh is is potentially a thing of the
2:01:39 past
2:01:41 wow I think because of its entertainment
2:01:44 value I may opt out of other
2:01:46 entertainment Subs you know what's funny
2:01:47 Joe mama that that's that's the one of
2:01:49 the things that I've been experiencing
2:01:52 in in this channel in particular but but
2:01:54 even over on on the salon
2:01:57 um is
2:01:59 that like using these tools is
2:02:02 entertaining like a video game's
2:02:03 entertaining because the the response is
2:02:05 so immediate and like especially looking
2:02:08 at some of this this uh Gemini interface
2:02:11 where you have this Infinite Canvas
2:02:13 where you just drop objects onto it
2:02:15 pretty crazy let's let's let's see how
2:02:18 far deeper they go with the programming
2:02:19 stuff was the first AI system that could
2:02:21 compete roughly at the level of the
2:02:24 average human competitor today I'm
2:02:27 delighted to introduce Alpha good 2 a
2:02:30 new and enhanced system with massively
2:02:32 improved performance powered by
2:02:36 Gemini when
2:02:38 we platform as the original alha code we
2:02:41 solve almost twice as many problems
2:02:44 while Alpha code broke through the top
2:02:46 half of human competitors on average we
2:02:49 estimate that ala 2 performs better than
2:02:51 85% of competition participants Wow
2:02:55 Let's have a look at our model in action
2:02:57 on one of the hardest problems that we
2:02:59 faced and I say hard because in the
2:03:02 original contest in which the problem
2:03:04 appeared less than 2% of participants
2:03:08 actually solved it the problem is is
2:03:10 quite difficult it's very abstract so I
2:03:13 can't get into too many details but the
2:03:15 basic gist of it is that we are tasked
2:03:18 with Computing AG get statistics that
2:03:20 account for what appears to be an
2:03:23 impossibly large amount of random arrays
2:03:25 the really cool thing is that to solve
2:03:27 it Alpha 2 makes use of dynamic
2:03:30 programming dynamic programming is an
2:03:32 advanced algorithmic technique which
2:03:35 basically simplifies a complicated
2:03:37 Problem by breaking it down into easier
2:03:39 sub problems again and again and what's
2:03:41 really impressive is that not only
2:03:44 alocate 2 knows how to properly
2:03:46 implement the strategy but also when and
2:03:49 when to use it what the example shows us
2:03:52 is that competitive programming is not
2:03:54 just about implementation it's also
2:03:56 about understanding maths computer
2:03:59 science and indeed coding and that makes
2:04:01 it an extremely hard reasoning task so
2:04:05 it's not very surprising that up till
2:04:07 now generally available large language
2:04:10 models have scored very poorly on this
2:04:13 Benchmark these models are really really
2:04:15 good at following instructions but alha
2:04:17 code needs to do more than that it needs
2:04:19 to show some level of understanding some
2:04:22 level of reasoning designing of code
2:04:24 Solutions before it can actually get to
2:04:27 the actual implementation to solve the
2:04:29 problem and it does all that on problems
2:04:32 that it's never seen before another
2:04:35 thing that great about Alpha code is
2:04:36 that it performs even better when it
2:04:38 collaborates with human coders who can
2:04:40 provide grounding basically developers
2:04:43 can specify properties that the code
2:04:45 samples have to obey and when we do that
2:04:48 we see performance increase
2:04:49 significantly we think of this this kind
2:04:52 of interaction between uh programmers
2:04:55 and AIS as the future of programming
2:04:58 where coders will not just give
2:05:00 instructions but actually collaborate
2:05:02 with highly capable AI models that can
2:05:04 reason about their problems that can
2:05:06 propose code designs and that can even
2:05:09 help with the actual implementation Alpa
2:05:11 2 was built for competitive programming
2:05:13 but we're already working on bringing
2:05:15 some of its unique capabilities right
2:05:17 into the general Gemini model models as
2:05:20 a first step towards making this new
2:05:22 programming Paradigm available for
2:05:27 everyone this is
2:05:29 impressive Pate we look like this as a
2:05:31 parent you may have to help your kid
2:05:33 with their homework I've certainly had
2:05:35 to here's where Gemini can help for this
2:05:38 demo we've created a simple interface
2:05:40 and with some clever prompting under the
2:05:42 hood we can really leverage Gemini's
2:05:44 math reasoning and multimodal
2:05:47 capabilities to learn a subject like
2:05:49 physics with Gemini you can upload a
2:05:52 photo of handwritten answers on a
2:05:54 worksheet not only can Gemini solve
2:05:56 these problems but this is the amazing
2:05:59 part it can read the answers and
2:06:01 understand what was right and what was
2:06:03 wrong and explain the concepts that need
2:06:05 more clarification so Gemini Identify
2:06:08 some everyone coming to P's defense for
2:06:11 the
2:06:11 [Laughter]
2:06:16 win no P's amazing P P's
2:06:20 like like one of the reasons to join the
2:06:23 salon is for is for a lot of the
2:06:25 Irregulars in here people like P he runs
2:06:27 the mechanics Guild he'll explain all
2:06:30 this [ __ ] to you if you want to if you
2:06:32 understand how this stuff works you know
2:06:34 why it works why it works sometimes not
2:06:37 other times the mechanics Guild is is
2:06:40 really good at that the mistakes with
2:06:41 problems one and three here let's take a
2:06:44 look at
2:06:47 three here Gemini identifies that the
2:06:50 formula was correct but there was a
2:06:52 mistake in calculating height we can ask
2:06:55 Gemini to explain in more details why
2:06:57 the height is 50 m instead of
2:07:05 6 I can ask Gemini to explain
2:07:13 further here Gemini explains the
2:07:15 step-by-step details to solving the
2:07:18 problem
2:07:20 because of Gemini's ability to
2:07:22 understand Nuance information and answer
2:07:24 questions relating to complicated topics
2:07:27 it can give you a customized explanation
2:07:28 of the subject you're trying to learn
2:07:31 and lastly if you want to learn more you
2:07:33 can just
2:07:38 ask Gemini will provide personalized
2:07:40 practice problems based on
2:07:43 mistakes here I have a similar problem
2:07:46 where I have to figure out the cat speed
2:07:48 the height of the ramp is
2:07:52 double oh yeah I knew
2:07:55 that wow I think that's it for that
2:07:58 playlist
2:08:00 yep good Lord
2:08:08 people I think I need a
2:08:13 nap where are the safeguards go play
2:08:16 with Bard right now Bard's basically
2:08:18 useless right now they just it's all
2:08:19 it's nothing but
2:08:21 safeguards um there there's there's
2:08:25 there's a number of videos on safety for
2:08:27 this so they're they're addressing it
2:08:31 um
2:08:33 wow uh let's see what else we got
2:08:43 here I have to go for now take care take
2:08:45 care Andrew going to pick up my son from
2:08:49 prek back later see you later print
2:08:50 hello world yeah I'm going to go too I
2:08:52 got to I'm going to
2:08:54 go take a nap and grab dinner and just
2:08:58 you know watch some TV or something like
2:08:59 that uh I'm working on digital empathy
2:09:02 as far as what's next very cool where
2:09:05 can I get it uh the uh it the Gemini it
2:09:11 sounds like it's coming out on the 13th
2:09:13 can we guess we can stop making gpts not
2:09:17 necessarily like you know listen don't
2:09:19 count don't
2:09:21 count open AI
2:09:23 out I mean I
2:09:26 mean they apparently just fired Sam mman
2:09:29 because they achieved AGI so I would
2:09:32 assume that whatever the [ __ ] they're
2:09:34 calling GPT 5 is probably in the
2:09:38 neighborhood of mindblowing so so do do
2:09:42 not think that this uh also everything
2:09:45 that I've showed here today is all
2:09:47 marketing stuff from Google so so we'll
2:09:49 have to see when this thing comes out if
2:09:50 it's that good but boy is it looking
2:09:52 promising like the the the dynamic
2:09:55 interface generation that that the way
2:09:58 it's doing you know that sort of
2:10:00 cascading coding where it starts with
2:10:03 like logic and intention and then Works
2:10:05 down to implementation there's there's a
2:10:08 just a lot um thank you public school
2:10:10 teacher here mind blown km please join
2:10:13 the salon the the AI Salon um we've got
2:10:17 a number of Educators in there we're
2:10:19 about to spin up a um an education Guild
2:10:24 um I have I have historic I have a
2:10:26 degree in acting I I am
2:10:28 not I am I have not historically been a
2:10:31 fan of the
2:10:33 education institution I am really
2:10:36 passionate about Educators need to get
2:10:38 their [ __ ] together and get aggressive
2:10:42 with understanding what these tools are
2:10:45 cuz it because students today are are
2:10:48 are going to be exiting a school system
2:10:51 into a world that does not
2:10:55 resemble what they've been trained on uh
2:10:58 if if they don't get their [ __ ] together
2:11:00 so thank you for hanging out here I'm
2:11:02 glad you're here dinner yeah I know I
2:11:03 got to go too all right where did you
2:11:05 get it see you at lunchtime live in the
2:11:08 real world we love you thank you very
2:11:09 much thanks Kyle see you tonight all
2:11:11 right yeah I'm going to get out of here
2:11:13 everybody um really good hanging out
2:11:15 with you Gemini's been announced this is
2:11:18 this will obviously be the bulk of the
2:11:20 talk tonight um we we can do other [ __ ]
2:11:23 but
2:11:24 um I think I'm going to go unplug my
2:11:28 brain for a bit and just process this
2:11:31 Champ's done he's ready he's ready to
2:11:33 eat the neighbor I guess all right I
2:11:35 guess I I need to get dressed I guess
2:11:37 yeah Joker you go get dressed for
2:11:41 tonight thanks for the laughs you're
2:11:44 welcome all right everybody great seeing
2:11:46 you bye