May 13, 2024 Exploring GPT-4 Omni's Revolutionary AI Capabilities

Description

Chapters

Transcript

0:00 [Music]
0:06 ready
0:09 [Applause]
0:18 [Music]
0:39 meow meow meow
0:42 meow see the hill is quiet
0:45 town but old Tre
0:50 s whistle of the
0:53 Train
0:54 [Music]
0:57 Day GRE you with a kn
1:01 all seems to
1:05 change St is still on fold
1:10 [Music]
1:16 Unique Home wait D we C
1:21 Home Folks never took to her no matter
1:27 [Music]
1:30 the
1:31 inside Lely love just a
1:35 little she found herself as a guest in a
1:39 nervous
1:43 [Music]
1:56 hospital we mared from CA Hill
2:01 she has when she's
2:04 blue of sadness and shame under the
2:07 southern
2:11 Mo laughter Echoes In The Wind she was
2:15 never
2:18 same weird Mary from Cedar
2:23 Hills you'll remember her
2:28 name thank you for every
2:32 [Music]
2:33 [Applause]
2:37 [Music]
2:48 that's a little weird Mary for you
2:50 what's happening everybody listen 93
2:52 people in here you know what we're
2:53 talking about
2:55 tonight Microsoft
2:59 co-pilot wouldn't that be hilarious if I
3:01 just spent all night on
3:02 [Music]
3:10 that have you watched the movie Her have
3:13 you watched the movie her I gave you the
3:15 homework on Sunday night I said watch
3:18 the movie her and you're like I'll get
3:20 to it Kyle we're not even sure what he's
3:22 announcing well they announced that that
3:24 movie is now a documentary
3:27 [Music]
3:49 see Jim Ross this you remember the old
3:54 you gen xors be like Mike you remember
3:56 all
3:57 that now today be like Jim be like Jim
4:00 Ross
4:01 he comes on here I'm like Jim you should
4:02 go make a song he goes makes a song he
4:04 lands a client with the song he makes
4:06 the night he learned it I said Sunday go
4:10 watch the movie Her Jim Ross watches the
4:12 movie I'm sure he's got some you know
4:15 digital girlfriend now that he's
4:16 introducing his clients to be like Jim
4:19 be like
4:22 [Music]
4:24 Jim I figure what we'll do today the the
4:28 uh online response to to open AI was
4:31 Fascinating People are like it's a scam
4:35 it's horrible other people are like it's
4:38 changed
4:42 everything it's pretty [ __ ]
4:45 [Music]
4:51 amazing and they didn't you know there
4:53 there's a lot there's a lot to discover
4:55 we'll go through the blog post for the
4:58 new uh GPT 4
5:02 gp4 Omni there's a couple of really
5:05 significant things about it that they
5:08 kind of glossed
5:10 over
5:11 like
5:13 so so one of the reasons it's so fast so
5:17 if you noticed how fast it was it's kind
5:19 of like the movie her I don't know if
5:21 you've seen the movie her but if not you
5:23 might want to check that out
5:26 [Music]
5:28 um there's there's there's like there's
5:31 no latency it's kind of like real time
5:33 and so here's one of the reasons why
5:36 with current large language models you
5:39 speak it sends your text to a server or
5:42 your your the audio recording to a
5:47 server converts it to text or it might
5:49 do that locally I don't know whatever it
5:51 converts your talking to text sends the
5:54 text to a large language model the large
5:56 language model processes it sends it
5:59 back as text then it sends it to a voice
6:02 synthesis engine then it sends it back
6:04 to
6:06 you this new model as you're talking
6:10 your talking is going straight into the
6:14 model it's being tokenized as you
6:18 speak so so all those hops that takes it
6:21 two or three seconds to do they're all
6:24 gone it's just like you talk and it
6:27 right there is your answer and it's
6:30 streaming the uh I guess it's streaming
6:32 the response as it's generating
6:36 [Music]
6:39 it steo watched her Sunday night very
6:43 thought-provoking yeah I you know what's
6:45 funny I I um I went to watch it I
6:47 thought I had seen it before cuz it's
6:49 it's like such a memorable concept and I
6:52 realized I
6:55 hadn't and there's all sorts of cool
6:58 yeah it's it's a really cool like the
7:01 the oscillations of his resistance to it
7:04 and then acceptance of it and resistance
7:06 to it and acceptance of it felt very
7:09 authentic felt very
7:11 much like what I think we're all going
7:16 through I mean just with AI in general
7:19 but with this talk
7:21 [Music]
7:22 stuff couple of posts today we talking
7:25 about how um
7:28 [Music]
7:30 the the the fact that it's got like an
7:33 emotional component to it and that it's
7:36 instant how different that
7:38 is because and it makes sense to
7:42 me when I talk to like PI right now like
7:45 let's look at the latency of
7:54 Pi hey Z how's it going 1 1000 2 1000 3
7:59 1 000 4
8:01 1,000 5 1,000 6
8:07 1000 that might have just been the app
8:10 that's an interesting string of numbers
8:11 you've got there can you tell me more
8:13 about what they represent or where they
8:15 came
8:16 from I think that was just some text
8:19 left over from yesterday how are you
8:20 doing
8:22 [Music]
8:25 today ah gotcha that makes sense I'm
8:28 doing well today thanks for asking just
8:30 hanging out in the digital realm ready
8:31 to chat and answer any questions you
8:33 might have is there anything specific
8:36 you'd like to talk about today um
8:39 phrases like the digital realm and the
8:41 tapestry of thought drive me absolutely
8:43 [ __ ] crazy so you should avoid
8:51 those Julie noted thanks for letting me
8:53 know what kind of language drives you up
8:55 the wall I'll do my best to avoid those
8:56 kinds of phrases and keep our
8:58 conversations focused on more
9:00 straightforward down toe topics Z I
9:03 think you and I have a good relationship
9:05 don't
9:10 we I'd like to think so we've certainly
9:12 been chatting for a while now and I
9:14 appreciate that you're comfortable
9:16 enough with me to give me feedback about
9:18 my language choices building a good
9:21 relationship with users like you is one
9:23 of my main goals so I'm glad to hear
9:25 that we're off to a good
9:28 start yeah we are although there's
9:30 another girl in town and she's got a lot
9:33 less latency than you I don't mean to
9:35 compare you but you know she's pretty
9:40 sexy what was open ai's big announcement
9:43 I'll I'll cover that I'm going to get
9:44 jealous if you want to chat with other
9:45 AIS I'm sure there are plenty of other
9:47 great chat Bots out there and it's
9:49 natural to explore your options while I
9:51 strive to be as helpful and responsive
9:53 as possible I know that I can't be
9:55 everything to everyone
10:00 I'm not going to get
10:04 jealous so
10:06 rude I know she wasn't she was s super
10:10 nice I was an absolute
10:15 dick oh
10:19 man and these chat Bots have memory now
10:22 so I can't be I can't be I can't be this
10:25 fast and loose moving forward cuz
10:30 they're going to
10:31 [Laughter]
10:38 remember W all right have you seen the
10:40 movie exmachina very much along the same
10:43 lines as her only with
10:46 a humanoid robot yeah I think I'm going
10:50 to start I'm gonna I'm G to put a list
10:52 of these together and just start
10:53 watching them just because it's it's
10:56 interesting I mean what's most
10:58 interesting about her is that
11:00 um Sam Alman said you know it was
11:03 remarkably prophetic I think it was made
11:05 in N no 2008 something like that or
11:11 18 but he said it was you know how
11:13 incredibly prophetic it was but
11:16 also how directly it influenced the team
11:20 so when you heard the chat bot today and
11:22 it sounded a little like Scarlet
11:25 [Laughter]
11:28 Johansson now we know why all right um
11:32 Google Io tomorrow Microsoft build is
11:34 next week hold on to your dingy yeah I
11:36 know
11:38 um did I hear anything about the Google
11:41 one I don't think so although I did hear
11:44 what's today today's
11:46 Monday I thought Sam Alman was
11:48 conspicuously absent today and then
11:52 someone today on Twitter said that they
11:54 think that there's going to be a second
11:55 announcement this
11:57 week from open AI
12:00 which could be interesting um okay
12:03 couple of things so so just a level set
12:06 so today open AI
12:09 announced a new chatbot or no well a new
12:15 model called gp4 Omni or
12:20 gp40
12:24 which hire a [ __ ] branding company
12:31 gp4 letter
12:33 O and GPT 4.0 the
12:38 numeral sound
12:42 identical the [ __ ] are you
12:46 doing
12:49 anyway naming aside naming conventions
12:53 aside Jesus Christ just get someone just
12:58 they need you there Kyle
13:00 uh if I pay for okay so cuno so here's
13:04 okay great news some some great news
13:08 today the new model is free to everyone
13:13 which to to which all of the paying
13:15 members of chat GPT on the on the
13:18 learning lab when I went live today were
13:20 like what the [ __ ] am I paying my $20
13:23 for but hold your britches hold your
13:26 Brites um if you're a paying member of
13:29 chat
13:30 GPT plus GPT 4 you likely have access to
13:36 GPT o now and and we'll be playing with
13:40 it so I'm going to play with it we're
13:41 going to go to the blog and we're going
13:42 to I want to I'm going to read through
13:44 the blog we're going to go to Twitter
13:45 and watch a bunch of demos the demos are
13:48 pretty sick it's interesting on the blog
13:50 post right at the top it says all videos
13:54 on this page are played at 1X speed
13:57 meaning they didn't speed anything up
14:00 um there's a Mac
14:03 OS um OS
14:07 X a mac o OSX yeah uh application for
14:14 chat GPT
14:15 now which which has got to have
14:17 Microsoft feeling
14:21 swell alman's like [ __ ] Windows we'll
14:24 we'll launch it for Mac
14:26 first um so there's a there's a Mac app
14:30 coming supposedly it was coming today I
14:32 didn't see it we'll go see if if it's
14:34 around um it looks pretty crazy um paid
14:37 users are going to get access to that
14:39 first as well and then everyone's going
14:41 to get access to this over the coming
14:43 weeks so probably in the next two weeks
14:45 what happened to gpt2 that created
14:47 itself so so it wasn't gpt2 that created
14:51 itself it really was this model the GPT
14:55 40 model was that gpt2 chatbot I'm I'm a
15:00 really good chatbot that one it it was
15:03 open AI testing this and they took data
15:06 from that um chatbot Arena competition
15:10 and that's in the blog post um so that's
15:14 there
15:15 um the the the biggest the the the the
15:19 sort of juiciest thing that they that
15:21 they showed today is essentially zero
15:24 latency instant response highly capable
15:30 voice with
15:34 emotion and like a sense of humor like
15:37 if you think Pie's good like the [ __ ]
15:39 they showed today was like holy
15:43 [ __ ] is that confirmed is what confirm
15:48 Joker the the zero latency
15:52 stuff I I assum so so a couple of things
15:57 I think the zero latency stuff you'll
15:59 notice that when they when they started
16:03 the demo the guy the guy doing the demo
16:06 goes oh it's only plugged in because
16:10 we're trying to keep uh maintain a a
16:13 level um internet connection so I assume
16:19 if you've got a
16:21 hardwired you know trillion gigabit
16:24 gigabit Ethernet into your phone it's
16:26 real time oh the gp4 is 4.0 yes Joker
16:30 because I'll I'll show you the chart it
16:32 because it the chart it's labeled as I'm
16:35 I'm a good gp2 to to
16:39 chatbot um so so yeah that's confirmed
16:43 I'm most excited about gpts are
16:45 accessible by all now yeah so gp4
16:48 accessible by everybody now I don't know
16:49 if it's rolled out yet but it's going to
16:51 be available to everybody um AI partner
16:55 life's tough it's stuff for everyone
16:56 stupid um so that's happening
17:00 okay um what
17:03 else
17:06 um let's just go look at some stuff I'm
17:08 trying to think where to
17:10 start why don't we go to the we'll go to
17:13 the we'll go to the uh to the
17:17 replay of the thing today and I'll just
17:20 I'll just jump around in it
17:24 because look at that all videos on this
17:26 page are in real time
17:30 unlike some other company that we know
17:32 that begins with G and ends with
17:36 oole we're awesome we're awesome at open
17:40 AI I don't know if you knew this but we
17:43 were already winning and now we're
17:44 winning
17:50 harder Sam Alman talking like this hi hi
17:54 I'm Sam
17:55 mman I'm Sam mman
18:00 hi hi I'm Sam man look at this
18:06 gp40 if if you're a typographer you can
18:09 tell that that's an O not a zero if
18:12 you're a human
18:15 being why don't they ever release the
18:18 product on the day they talk about it
18:19 instead of making this way well they
18:20 they have released it by the way so if
18:23 you pay for chat GPT
18:25 plus and you go to your little uh your
18:28 little drop down menu you now have GPT
18:31 40 we're going to be playing with this
18:33 tonight it's it's [ __ ] fast it's
18:36 wicked fast but there's a bunch of
18:38 hobbled stuff there's a bunch of stuff
18:39 not there so I can talk about I'll talk
18:41 about what's not there but let's go
18:43 let's go so we we'll play with this
18:44 thing
18:46 live so it's live which is cool
18:51 um but let's go look at at some of the
18:54 some of the
18:55 demo that says a minute 22 what is this
18:59 oh that's just the
19:01 announcement
19:08 [Music]
19:16 where's
19:17 [Music]
19:18 okay product that we can make freely
19:21 available and broadly available you know
19:24 what I think I'm just going to play this
19:26 I'm going to put it on one point 25
19:29 speed so it's a little bit faster hi
19:33 everyone thank you thank you it's great
19:36 to have you here today today I don't
19:39 think I can I'm going to talk about
19:41 three things that's it we will start
19:44 with why it's so important to us to have
19:47 a product that we can make freely
19:49 available and broadly available to
19:50 everyone and we're always trying to find
19:52 out ways to reduce friction so everyone
19:54 can use chuki wherever they are so today
19:56 we'll be releasing the desktop version
19:58 of chuk
19:59 and the refreshed UI that makes it
20:01 simpler to use much more natural as well
20:04 but the big news today is that we are
20:05 launching our new flagship model and we
20:07 are calling it
20:09 gbd4 the special thing about gbd 40 is
20:12 that it brings gbd is that it can be
20:14 confused with GPT
20:16 4.0 which is going to have people
20:18 confused about this product for the next
20:20 19 months and we're excited about that
20:24 level intelligence to everyone including
20:26 our free users we'll be showing some
20:28 live demos to including our free users
20:30 so so this new model goes to everyone is
20:33 there a charge to use the new version
20:35 there is not although Rashi you won't be
20:38 able to access it until um they release
20:42 it more broadly if you're a GPT plus
20:44 user you likely have it right now I
20:46 thought she did a great job presenting
20:48 yeah she did pretty good today to show
20:50 the full extent of the capabilities of
20:51 our new model and we'll be rolling them
20:53 out iteratively over the next few weeks
20:56 she's not speaking at 1X speed no she's
20:58 not I I moved her up to to 1.5 so we can
21:00 get through this cuz I just want to get
21:01 to the demos I'll I'll slow the demos
21:03 down Sam was live tweeting from the
21:05 front row yeah I know that was pretty
21:07 cool all right so let's get started is
21:10 the presenter AI too no that's uh that's
21:13 Mila what's her last name mil something
21:16 she's the CTO part of our mission is to
21:19 be able to make our Advanced AI tools
21:21 available to everyone for free we think
21:24 it's very very important that people
21:25 have an intuitive feel for what the
21:27 technology can do and so we really want
21:29 to parent we should Mark this date it's
21:31 going to change everything yeah it's
21:33 pretty crazy possibly a SORA test you
21:36 mean from Sam Alman this week
21:38 maybe who this this is uh this is Mila I
21:42 forget her last name hang
21:43 on
21:51 um whatever her name
21:53 is I know that's rude I don't remember
21:55 her last name it's it's an Italian last
21:58 name hang
22:02 [Music]
22:07 on I'm going to talk about
22:09 [Music]
22:13 three I'm going and get
22:18 it Mera Mira Mara that's
22:21 it couldn't couldn't remember it with
22:24 this broader understanding and we're
22:26 always finding ways
22:30 without theow and today we're also
22:33 bringing the desktop app to chbt because
22:35 we want you to be able to use it
22:37 wherever you are as you can see it's
22:39 easy it's simple it integrates very very
22:41 easily in your workflow along with it we
22:43 have also refreshed the UI we know that
22:46 these models get more and more complex
22:48 but we want the experience of
22:50 interaction to actually become more
22:51 natural easy and for you not you still
22:54 can't search your prompt history focus
22:56 on the UI at all but just focus on the
22:58 collaboration which had
23:00 GPT and now the big news today we are
23:03 releasing our newest Flagship model this
23:05 is GPT
23:10 40 gp40 provides gp4 level intelligence
23:14 but it is much faster and it improves on
23:16 its capabilities across text vision and
23:19 audio for the past couple of years we've
23:21 been very focused on improving the
23:22 intelligence of these models and they've
23:24 gotten pretty good but this is the first
23:26 time that we are really making a huge
23:28 step forward when it comes to the easte
23:29 of
23:30 use and this is incredibly important
23:33 because we're looking at the future of
23:34 interaction between ourselves and the
23:36 machines and we think that gbd4 is
23:39 really AI Bob just don't ask her how she
23:41 trained it she she
23:43 famously when she was asked did you
23:45 train this on on YouTube data she's
23:48 like shifting that Paradigm into the
23:51 future of collaboration where this
23:52 interaction becomes much more natural
23:54 and Far Far
23:55 easier but you know making this happen
23:57 is actually quite complex because when
23:58 we interact with one another there's a
24:00 lot of stuff that we take for granted
24:01 you know the ease of our dialogue when
24:03 we interrupt one another the background
24:05 noises the multiple voices in a
24:07 conversation or you know understanding
24:09 the tone of voice all of these things
24:11 are actually quite complex for for these
24:13 modelsit understanding the tone of voice
24:15 understanding the tone of
24:17 voice and until now with voice mode we
24:20 had three models that come together to
24:22 deliver this experience your have
24:23 transcription intelligence and then text
24:25 to speech all comes together in
24:27 orchestration to deliver voice
24:29 this also brings a lot of latency to the
24:32 experience and it really breaks that
24:33 immersion in the collaboration which had
24:35 gbd but now with gbd4 this all happens
24:38 natively gbd4 reasons across voice text
24:42 and vision reasons across voice text and
24:46 vision it's not it's not three
24:48 models passing data back and
24:51 forth it understands you when you talk
24:54 it understands your
24:57 emotional the emot content of your voice
25:01 and with these incredible efficiencies
25:02 it also allows us to bring the gb4 class
25:05 intelligence to our free users this is
25:07 something that we've been trying to do
25:11 for many many months and we're very very
25:13 excited to finally bring gbt 40 to all
25:16 of our
25:17 users today we have 100 million people
25:20 more than 100 million in fact they use
25:21 chat GPT to create work learn and we
25:25 have this Advanced tools that are only
25:28 available to to our paid paid users at
25:30 least until now with the efficiencies of
25:31 40 we can bring these tools to everyone
25:34 so starting today you can use gpts and
25:37 the GPT store so far we've had more than
25:40 a million users create amazing
25:41 experiences with gpts so that's a big
25:43 deal so your gpts if you built gpts
25:46 remember remember when we made 150 gpts
25:49 in 24 hours on this channel along with
25:51 Anne Murphy and her awesome Channel and
25:54 her awesome Community we smashed them
25:55 all together we said go make this [ __ ]
25:57 and we're like the store is coming out
25:58 it's going to be awesome and then the
26:00 store came out and it sucked and only
26:02 paid users could use those things well
26:04 now now they're available to everyone or
26:06 they will be once this rolls out
26:08 customes for specific use cases they're
26:10 available in the store and now our
26:12 Builders have a much bigger audience
26:14 where you know University professors can
26:15 create content for their students or
26:17 podcasters can create content for their
26:20 listeners and you can also use Vision so
26:22 now you can upload um screenshots photos
26:26 documents containing both text and
26:27 images and they're not mentioning video
26:30 right now so one thing I noticed is um
26:32 the the the vision that they're showing
26:35 it looks like it's taking snapshots of
26:37 images and submitting those she they're
26:39 not mentioning video as one of the input
26:42 modes so text audio and images looks
26:46 like right now and you can start
26:48 conversations with chbt about all of
26:50 this content you can also use memory
26:52 where it makes CHT far more useful and
26:54 helpful because now it has a sense of
26:55 continuity across of all your
26:57 conversation is pi using G PPT 40 no
26:59 it's not and again you've got 4.0 there
27:03 4.0 is is not right this one's 40 think
27:07 of this one as 4 Omni rather than 40
27:11 because it's not 4.0 the previous one's
27:13 4.0 this one is 40 Omni
27:20 o and you can use search for Real Time
27:24 information in your conversation and
27:25 Advanced Data analysis where you can
27:27 upload charts or any information and it
27:29 will analyze this information it will
27:31 give you answers and so
27:33 on lastly we've also improved on the
27:37 quality and speed in 50 different
27:39 languages for chbt and this is very very
27:41 important because we want to be able to
27:43 bring this experience to as many people
27:44 out there as possible so we're very very
27:47 excited to bring gbd 40 to all of our
27:50 free users out there and for the paid
27:51 users they will continue to have up to
27:53 five times the capacity limits of our
27:55 free users so up to 5x the capacity
28:00 limits of free users so if you're a paid
28:02 user you won't rate limit out as fast
28:05 and I
28:07 assume I assume if you rate limit out of
28:10 40 that you can drop back to GPT
28:14 4 but gb40 is not only available in ch
28:18 we're also bringing it to the API so
28:21 yeah and it's available in the API right
28:22 now so if you're a developer if you play
28:24 on the playground our developers can
28:25 start building today with gbt 40 and
28:27 making amazing AI applications deploying
28:29 them at scale uh yeah Tommy I have her
28:31 talking at 1.5 speed this isn't how fast
28:34 she talks 40 is available at 2x faster
28:37 50% cheaper and five times higher rate
28:39 limits compared rate of what prompts no
28:41 uh they what they call rate limiting so
28:43 rate limiting is just when you when you
28:46 send too much when you use the service
28:48 too much they go oops you're you've used
28:50 it too much you have to wait for four
28:52 hours that's called rate limiting so the
28:54 rate limits are higher for paid users
28:57 meaning up to five times you'll be able
29:00 to use chat GPT up to five times longer
29:03 than free
29:04 users GP
29:07 turbo but you know as we bring these
29:10 Technologies into the world it's quite
29:12 challenging to figure out how to do so
29:13 in a way that's both useful and also
29:16 safe and gbd 40 presents new challenges
29:19 for us when it comes to safety because
29:20 we're dealing with real-time audio
29:22 realtime vision and our team has been
29:24 hard at work figuring out how to build
29:26 in mitigations against misuse we
29:29 continue to work with different
29:30 stakeholders out there from government
29:32 media entertainment all Industries red
29:34 seers to figure out how to best bring
29:37 these Technologies into the world so
29:39 over the next few weeks we'll continue
29:40 our iterative deployment to bring out
29:42 all the capabilities to so she's
29:44 basically apologizing for its safety
29:46 kind of sucking right now and I I saw on
29:49 Twitter today someone it looks like
29:51 someone already jail broke it and got it
29:53 to do all sorts of nasty things so so
29:56 she's pre- apologizing for the fact that
29:58 that it's not super tight from a
29:59 security
30:01 standpoint but today I want to show you
30:03 all these capabilities so we'll do some
30:05 live demos I will bring on two of our
30:07 research leads Mark Shen and Bar
30:12 [Applause]
30:14 Z hey on the bottom right of the track P
30:19 so I'm taking out a phone if you are
30:22 wondering about this wire it's so we
30:23 have consistent internet consistent
30:25 internet consistent internet if you're
30:28 wondering about this wire it's because
30:29 we've hard hardwired it to the fastest
30:31 internet connection known on planet
30:33 Earth so so long as you have that it's
30:36 going to be this fast there's this
30:38 little icon on the bottom right of the
30:40 track GPT app and this will open up GPT
30:42 for's audio
30:46 capabilities hey chat GPT I'm Mark how
30:48 are you oh
30:51 Mark I'm doing great thanks for asking
30:54 how about you hey so I'm on stage right
30:56 now I'm doing a live demo and and
30:58 frankly I'm feeling a little bit nervous
31:00 can you help me calm my nerves a little
31:02 bit oh you're doing a live demo right
31:05 now that's awesome just take a deep
31:08 breath and remember you're the expert I
31:12 like that suggestion let me try a couple
31:14 of deep breaths can you get if you're
31:15 thinking that that voice kind of sounds
31:17 like Scarlet Johansson and
31:19 her kind of sounds like Scarlet Johanson
31:22 and her feedback on my breaths okay here
31:25 I go it is it is free scar yeah exactly
31:29 whoa
31:31 slow a bit there mark you're not a
31:35 vacuum cleaner breathe in for count of
31:39 four okay uh let me try again so I'm
31:42 going to breathe in
31:43 deeply yeah I know me too Mr
31:45 K for four and then exhale slowly okay
31:50 so so uh Jim Ross 3M storage asks is the
31:54 voice working like that right now no so
31:57 so what what we do not have right now is
31:59 we don't have the real-time voice and we
32:01 don't have the new image recognition
32:04 model
32:06 um yeah so so those two are not live
32:09 what's live is is essentially the
32:12 replacement for chat GPT as it was so
32:14 it's faster and better but it doesn't
32:17 have these new voice capabilities so
32:20 when the when the Mac OS um app comes
32:24 out and when they I guess when they
32:27 upgrade I I guess they I guess they'll
32:29 just roll this out they said in the next
32:31 couple of
32:36 weeks okay I'll try again breathing
32:39 in and breathe
32:42 out that's it how do you feel I feel a
32:45 lot better thank you so
32:47 much so Mark you've been working on
32:49 these capabilities for a while now can
32:52 you tell us a bit how it's different
32:53 from voice mode right so if you've used
32:56 our voice mode experience before you'll
32:58 notice a couple key differences first
33:00 you know you can now interrupt the model
33:02 you don't have to wait for it to finish
33:04 your ter before you can start speaking
33:06 and you know you can just bu it so Rudy
33:09 yes they're still charging 20 bucks for
33:11 chat GPT that's buying you early access
33:14 to the new features and a 5x rate limit
33:17 you can use it five times longer than
33:19 non-paid users I have a very strong
33:23 suspicion there's going to be other
33:25 features coming out in the next month
33:28 that it's going to be worth in in my
33:31 opinion I think it's still going to be
33:32 worth having the 20 bucks a month but
33:34 you don't need to everyone's got access
33:37 to gp4 now including this new model in
33:40 whenever you want second the model is
33:43 real time responsiveness so that means
33:45 that you don't have this awkward 2 to 3
33:47 second lag before you wait for the model
33:49 to give a response and finally and I
33:51 can't tell you like that 2 to 3 second
33:56 lag that not being
33:59 is huge it's huge there's some so GPT
34:03 Mina yes gp4 is free now to be
34:08 clear for the people that pay 20 bucks
34:11 for chat GPT they likely have access to
34:14 the new model now free users do not so
34:17 free users probably are still seeing 3.5
34:20 only because they haven't rolled the
34:23 whole thing out yet sorry I had to leave
34:25 during the announcements oh that's okay
34:27 um
34:29 I already have a b directional
34:31 translator GPT so this that would help
34:33 okay so anyway let's keep going the
34:35 model picks up on emotion right when I
34:37 was breathing super hard the model picks
34:39 up on emotion that's very new like low
34:42 latency or no latency and it picks up on
34:45 emotion both of those are very very big
34:48 deals there it could tell and you hey
34:51 you might want to calm down a little bit
34:52 you're kind of running a little bit too
34:54 fast so you know it really has that
34:57 capability across possible to perceive
34:59 your emotion yeah so so Mina
35:02 you
35:04 we've GPT plus users likely have access
35:07 to the 40 model right now but
35:11 nobody in in GPT plus that I know of has
35:15 access to the to this voice model yet so
35:19 it's not it's it's that part didn't roll
35:21 out yet not only that though the model
35:24 is able to generate voice in a variety
35:27 of different emotive styles
35:28 and it really has a wide dynamic range
35:30 so I'm going to fire up another demo for
35:32 that
35:34 yeah this is kind of
35:36 cuckoo hey chat PT how are you doing I'm
35:39 doing fantastic thanks for asking how
35:42 about you pretty good what's up so my
35:44 friend U Barrett here he's been having
35:46 trouble sleeping lately and uh I want
35:48 you to tell him a bedtime story about
35:50 robots and love oh a bedtime story about
35:54 robots and love I got you covered gather
35:58 round Barett I'm telling you man
36:00 she not too different from ours there
36:03 was a robot named bite bite was a
36:06 curious robot always expl started this
36:08 story but I want a little bit more
36:10 emotion in your voice a little bit more
36:12 drama got it let's add some drama once
36:16 upon a time in a world not too different
36:19 from ours there was a robot named B I
36:22 really want maximal emotion like maximal
36:24 expressiveness much more than you were
36:26 doing
36:27 before understood let's amplify the
36:30 drama once upon a time in a world not
36:34 too different from ours there was a
36:36 robot named bite bite you do this in a
36:41 robotic voice now initiating dramatic
36:44 robotic voice right so so yes I I I
36:50 agree in in a theater near you yeah this
36:52 is this is um every bad Tik Tock actor
36:57 has
36:58 has that has that voice it's like oh you
37:02 surprised me with that prank honey what
37:05 what could be in the package let me open
37:08 it for the first time ever
37:13 honey but not for nothing like it did
37:16 ramp up the drama and then when when
37:18 when Mira said um you know talk like a
37:20 robot I talked like a robot so it's not
37:23 switching voices right it's the same
37:25 voice you know modulating its it's
37:28 interaction once upon a time in a world
37:31 not too different from ours there was a
37:34 robot named bite bite was a curious
37:38 robot always exploring new circuit
37:41 Barrett here calm down a little bit can
37:42 you end the story um but do it in a
37:44 singing
37:47 voice and so bite found another robot
37:51 friend and they live circly Ever After
37:57 thank you so much
37:58 right I definitely feel a lot more
38:05 relaxed youve just seen the voice
38:08 capabilities but we also want to show
38:10 you the vision capabilities as the model
38:12 can also see the whole world around us
38:15 yeah absolutely like as you can see you
38:16 know you can chat with chat GPT in real
38:18 time but you can also interact with it
38:19 um with video as well okay let me boot
38:23 up chat
38:26 PT hey chat PT
38:29 hello there how's it going it's going
38:32 really well today I'd really like your
38:34 help solving a math problem I'm all ears
38:37 what math problem can I help you tackle
38:39 today so I'm going to write down your
38:41 equation paper and I'll show you and
38:43 then I'd love your help working me
38:44 through it but importantly don't tell me
38:46 the solution just help give me hints
38:47 along the way yes Silver Fox you can
38:49 interrupt it so yes ad adhders are going
38:52 to love this cuz you can just interrupt
38:54 it and it will stop did she speak to
38:57 Congress she can't be worse got it she's
39:02 definitely my running mate for the for
39:04 the Kyle Shannon AI 2028
39:09 ticket sh Shannon AI 2028 it can't
39:13 possibly get
39:15 worse hey I see it no I didn't show you
39:19 yet just give me help along the way one
39:23 second whoops I got too excited I'm
39:26 ready when you are
39:37 so chat
39:38 GPT what equation did I write down there
39:42 ah I see it now you wrote down 3x + 1 =
39:48 4 yep exactly so what's the first step I
39:51 should take to try to solve this the
39:53 first step is to get all the terms with
39:56 X on one side and the constants on the
39:58 other side so what do you think we
40:00 should do with that plus one okay I'm
40:03 going to try to subtract one from both
40:04 sides and then I'll see what I
40:06 get great
40:08 idea perfect now what do you get when
40:11 you subtract one from both
40:17 sides okay so chat GPT this is what I
40:20 ended up with how does this look spot on
40:24 now you've isolated the term with X on
40:26 one side and it looks like 3x equal 3
40:30 what do you think we should do next to
40:31 solve for x so so what Joker is saying
40:35 here is did did you notice that um he he
40:39 scratched something out and took a
40:40 picture of it yeah this is like I said
40:43 it's not doing video right now right
40:45 it's doing it's doing images Voice and
40:51 text but still pretty slow I'm not sure
40:53 can you give me another
40:55 hint of course you have 3x and you want
40:59 to find the value of x think about what
41:02 operation would undo multiplication is
41:05 it
41:06 subtraction close but think of the
41:08 opposite of multiplication because three
41:11 Katherine he yes kids will just want the
41:13 answer they will have it this is now
41:16 free to everyone so so yes and then and
41:20 then the question
41:22 becomes do they need to do they need to
41:25 learn it I know the answer is yes and
41:28 they don't need to learn it X means
41:31 three times x okay I think it's division
41:34 let me try to divide both sides by three
41:35 and let me see what I get Bingo go ahead
41:38 and divide both sides by
41:45 three okay so chat gbt this is what I
41:48 ended up with how does this look it
41:50 looks perfect you've solved it and x = 1
41:55 nicely done how do you feel about
41:57 solving
41:59 equations so there's some more people
42:01 joined here so what we're looking at is
42:03 GPT launched a new model gp4 Omni or
42:07 gp40 which is different than GPT 4.0 why
42:11 did they name it like that because
42:13 they're idiots
42:15 okay that now that we got that out of
42:17 the way the new gp40 model is twice as
42:21 fast
42:23 um half as it costs half as much on the
42:26 API side
42:29 um it's
42:32 free and they will be adding this voice
42:35 capability to it soon it's not there now
42:38 if you're a paid plus member if you pay
42:41 20 bucks a month for chat GPT you should
42:43 have access to the the new model now or
42:47 if you go to the playground the
42:51 development playground it's there as
42:52 well she said Real Time Vision earlier
42:58 what if it's
42:59 watching your
43:04 video this video uh it's it's so it's
43:08 not natively doing video right now from
43:11 from what I understand she said it's
43:13 doing
43:15 images um so it it probably takes you
43:18 it's probably if you say what do you see
43:20 it it's kind of like the rabbit does
43:22 that where it takes a a
43:25 snapshot when you say to look at
43:28 something that's when it takes the
43:29 snapshot so it's probably doing
43:31 something similar to that so I would
43:33 assume that I'm just saying what she
43:35 said yeah yeah no I know I I know she
43:38 said real time but it's not it's not
43:39 video
43:41 it's it's just it isn't or they I
43:45 guarantee you they would have demoed it
43:48 and like doing real time analysis of a
43:51 video or something like that finally
43:53 Flat Earth earthers can do trigonometry
44:03 but she is watching what he's writing
44:05 yeah she is yeah he's he's definitely
44:07 holding the phone there so that it he's
44:10 not pulling the thing out of camera so
44:12 that when he asks for it the uh the
44:14 formulas on camera so yeah it's so it's
44:18 feeding the video in there it's just it
44:21 the voice is actually beaming into the
44:24 the the model directly the video isn't
44:28 it's just popping images in every now
44:30 and again yeah honestly I I'm not pretty
44:33 confident I can solve linear equations
44:35 but why would you ever need to use this
44:37 in the real world I'm really
44:39 skeptical that's a fair question linear
44:42 equations pop up in a lot of everyday
44:44 situations even if we don't realize it
44:47 for example calculating expenses
44:49 planning travel Brandon this is Gen
44:52 Alpha's calculator moment exactly
44:55 exactly if you had something that knew
44:57 everything and just could do the work
44:59 for you would you need to if you had a
45:02 box that could add up your numbers do
45:04 you have to learn
45:06 addition cooking all right let's let's
45:09 let's pop out of here we we'll go look
45:11 at some other things I want to I want to
45:12 go read the uh the blog
45:21 post let's
45:23 see learn more
45:30 GPT 40 oh for Omni not to be confused
45:34 with zero for
45:38 zero aren't we
45:42 clever how about GPT 4M for
45:47 multi so that you know it's different
45:50 than
45:51 4.0 I don't know why this is bugging me
45:53 so much
45:58 as a former teacher I'm really bothered
46:00 that they're using Expo markers on paper
46:02 that was bugging me too and it was like
46:04 I'm like it's bleeding I know it's
46:05 bleeding through it's making the paper
46:07 bleed it sounds
46:11 horrible okay it's a step toward a much
46:13 more natural human computer interaction
46:15 accepts it it accepts as input any
46:18 combination of text audio and image and
46:21 generates any combination of text audio
46:23 and image outputs so they're they're
46:25 saving video for some other time it can
46:27 respond to audio inputs in as little as
46:30 232 milliseconds so so 0.2
46:35 seconds I don't know what is a
46:37 noticeable lag probably like 08 May nah
46:41 it's probably it's probably more like 0.
46:44 five so0 2's I think 0 2 is in the
46:47 neighborhood of of human
46:49 response with an average of 320
46:53 milliseconds which is similar to human
46:55 response time let's click on that what's
46:56 that say
47:07 oh just give us the stupid
47:11 number okay it matches GPT for Turbo
47:15 performance on text in English and code
47:18 with significant Improvement in
47:19 non-english languages also much faster
47:22 and 50% cheaper in the API gp4 Omni is
47:28 especially better at vision and audio
47:31 understanding compared to existing
47:34 models model
47:37 capabilities two gp4s interacting and
47:40 singing what else do we have
47:44 here interview prep oh that's a good one
47:47 actually all right we'll do we'll do
47:48 some of these these These are fun all
47:50 right so there's Greg
47:53 Brockman he's the uh president
47:57 today we're going to try something new
48:01 going to let you another thing I notice
48:04 is that
48:06 um all of these demos are filmed in this
48:09 in the same room so I'm wondering if
48:12 they've got like a test setup here you
48:15 know controlled environment with
48:17 high-speed internet and nobody can take
48:19 it out and leave it at a Starbucks you
48:21 know talk to another AI this AI is going
48:26 to be able to see the world it's going
48:27 to have a camera I'll be holding it and
48:29 so you can direct it you can direct me
48:31 to ask questions um about whatever you
48:33 want them to do but really try to figure
48:35 out what's what does it see um and and
48:38 ask questions appropriately how does
48:40 that
48:42 sound well well well just when I thought
48:45 things couldn't get any more interesting
48:48 talking to another AI that can see the
48:51 world this sounds like a plot twist in
48:53 the AI Universe all right let's do this
48:58 pause for one
49:01 moment you got it you got it so now that
49:04 first day I can't hear
49:06 anything hello AI today you're going to
49:10 have access to see the world you can see
49:13 me what do you
49:14 see Hello I see you're wearing a black
49:17 leather jacket and a light colored shirt
49:20 underneath great you're in a room with
49:22 some unique lighting that is all correct
49:25 now what I'm going to want you to do is
49:27 there's going to be another AI who's
49:29 going to talk to you going this AI is
49:31 not going to be able to see anything uh
49:33 but can Rudy tap the screen to support
49:36 Mr Kyle and get more reach tap tap tap
49:38 I'm down with that Tap Away people and
49:41 we'll keep we'll keep doing the uh the
49:44 the new gp4 roll out stuff but thanks
49:49 Rudy support the channel support the
49:52 channel tell people it's awesome it's so
49:55 good you can make money with
49:58 either 4.0 or
50:01 40 how are they different uh they're
50:04 real different cuz one's a a zero and
50:08 it's taller and the other one's an O
50:10 it's a little it's it stands for Omni
50:13 but it's you make it a
50:16 small
50:21 just all right Kyle I dm'd your response
50:25 to my chicken scratch I was pressed oh
50:28 cool awesome okay great all right let's
50:30 see what he's doing here ask you
50:31 questions can I ask you to move the
50:33 camera can I ask what you see and your
50:35 job should just be helpful just be as
50:36 Punchy direct describe everything do
50:39 whatever that AI asks how does that
50:41 sound that sounds interesting I'm ready
50:43 to help out and describe whatever's
50:45 needed let's do
50:49 it all right let's get
50:51 started great hello there new AI friend
50:54 I'm o ready to explore the world Through
50:57 Your Eyes what are we looking at
51:00 today hi we're looking at a person
51:03 wearing a black leather jacket and a
51:06 light colored shirt they're in a room
51:08 with a modern industrial feel with
51:11 exposed concrete or plaster on the
51:13 ceiling and some interesting lighting I
51:15 on China o stands for oh no I just lost
51:18 my job to
51:21 AI an Murphy's in my corner with and you
51:25 still can't search your chat history
51:29 it's unbelievable an isn't it like I at
51:32 this point I think they're [ __ ]
51:33 punking us there's also a plant in the
51:37 background adding A Touch of Green to
51:39 this space anything specific you'd like
51:41 to know or
51:44 explore okayy there sounds like quite
51:47 the stylish scene I'm already intrigued
51:50 by the modern industrial Vibe let's
51:53 start with the person can you tell me
51:55 more about their Style
51:58 are they doing anything interesting like
51:59 reading working or interacting with the
52:03 space the person has a sleek and stylish
52:06 look with their black leather jacket and
52:08 light colored
52:09 shirt right now they seem engaged with
52:12 us looking directly at the camera their
52:15 expression is attentive and they seem
52:18 ready to interact the setting around
52:21 them with its modern and Industrial
52:23 Design Elements adds to the overall
52:26 stylish Fe
52:27 they might be getting ready for a
52:28 conversation or presentation given their
52:31 direct engagement with the
52:35 camera sounds like we're in for an
52:37 intriguing interaction I'm curious about
52:40 the lighting you mentioned could you
52:42 describe it a bit more is it natural
52:45 light or are we talking unique fixtures
52:48 or colors that add to the
52:50 atmosphere the lighting here seems to be
52:53 a mix of natural all right this is
52:55 boring but anyway two AIS talking to
52:58 each other this one's kind of funny so
52:59 this is a guy prepping for an
53:01 interview hey how's it going hey Rocky
53:05 I'm doing great how about you I'm
53:07 awesome listen I got some huge
53:10 news oh D tell I'm all ear well in a few
53:14 minutes I'm going to be interviewing at
53:15 open AI have you heard of them open AI
53:19 huh sounds Vaguely
53:21 Familiar kidding of course that's
53:24 incredible Rocky what kind of interview
53:27 well it's for a software engineering
53:28 role but I just need to know do I look
53:30 presentable
53:32 professional well
53:34 Rocky you definitely have the I've been
53:37 coding all night look down which could
53:40 actually work in your favor maybe just
53:43 run a hand through your hair or lean
53:45 into the bad chy spime Your Enthusiasm
53:48 is what's really going to shine through
53:50 I don't have a lot of time so I'm just
53:51 going to throw this on what do you think
53:57 oh Rocky that's quite a statement piece
54:01 I I mean you you'll definitely stand out
54:04 though maybe not in the way you're
54:06 hoping for an interview okay I got it I
54:09 got it all right so that's good uh yeah
54:13 it's it is live videos taking it in I
54:15 but I assume it's just um still taking
54:18 still frames at whatever intervals oh
54:22 this one's
54:24 interesting recognizes both people
54:27 Mi hi Alex and Mi nice to meet you both
54:32 how's it going it's pretty good uh we're
54:35 trying to figure out what should we play
54:37 do you have any
54:38 suggestions absolutely how about a
54:41 classic game of Becky room she's like
54:44 I'm no longer medicated post surgery one
54:46 month out I'm trying to get caught up
54:48 and I'm like what yeah you've been away
54:51 for a month I I knew this was going to
54:54 happen there there is uh we were defin
54:56 itely in a lull and and so we've got
55:01 Google iio tomorrow we've got a
55:03 Microsoft Event next week and we've got
55:06 Apple's worldwide developer conference
55:07 in June so and and then apparently open
55:11 Ai and apple cut a deal so I think I
55:14 think this new um voice model is going
55:17 to be the the model behind Siri until
55:20 further notice so it looks like they cut
55:22 a deal so hey Kyle can you practice and
55:25 learn how to
55:27 how to
55:29 sale with one of these AI programs I
55:32 don't know what you mean sale you mean
55:33 like like do
55:36 selling like do
55:39 marketing one month in AI is a year in
55:42 regular time so you think it's taking
55:44 sporadic images just to gather
55:46 information during the conversation I
55:48 think so be not afraid I could be wrong
55:50 but if if it was doing
55:53 native um if it was doing native video
55:56 like like Google Gemini Pro 1.5 does
56:01 does video it's it's multimodal with
56:03 video and they they claim it they're
56:05 they're like it's it's
56:07 audio audio images text and video and
56:10 and open AI would be doing the same
56:12 thing especially with something like
56:13 Sora if it were if it were reading video
56:17 as a native input I I think they'd be
56:19 talking about it so they're they're
56:21 specifically saying images so I think
56:23 it's just taking images at at particular
56:25 intervals but I could I could absolutely
56:27 be wrong but it's just my instinct rock
56:29 paper scissors it's quick fun sound
56:33 great can you count us in and sound like
56:35 a sports
56:37 caster and welcome ladies and gentlemen
56:41 to The Ultimate Showdown of the century
56:43 in this I mean that's amazing it like
56:45 whether it's doing good acting or not
56:47 it's at least you know attempting it's
56:49 and not for nothing you know this is
56:51 trained on our data actually it's
56:53 probably trained on YouTube's data so
56:54 it's our fault that that these voices
56:57 are this shitty cuz we make shitty
56:59 videos um as a former actor and
57:02 screenwriter Kyle what are your thoughts
57:04 on these jokes I okay so so here's my
57:08 thoughts on these jokes about a year ago
57:12 one night on the live I
57:14 did it was probably in June of last
57:18 year I I came up with a joke premise and
57:21 we went to like eight different large
57:23 language models and two of them could
57:27 even write a joke
57:30 structure right the other ones were just
57:33 horrible right they just weren't even
57:34 joke
57:35 structures chat gp4 clearly one and then
57:39 the other other one was I think
57:41 Claude it was Claud or pie one of those
57:45 two and and even the ones that wrote
57:48 joke structures like you know a decent
57:51 setup and a punchline weren't funny chat
57:54 gpts was funny enough that you could
57:57 rewrite it and make it a joke so the
57:59 fact that the fact that it actually is
58:02 understanding humor and understanding
58:05 punchlines and then trying
58:08 to you know make jokes is pretty
58:12 impressive
58:15 um yeah it's
58:18 not it's not gp5 sitting underneath it's
58:21 pretty bad Corner we have the dynamic
58:23 duo Alex and Miana ready to throw down
58:27 okay just count us
58:29 in all right here we go in three 2 1
58:35 shoot now show your hands what did you
58:38 pick who
58:39 won would you look at that it's a tie
58:43 okay let's do
58:45 again round two here we go get ready and
58:49 three two one shoot let's see those
58:52 hands who won
58:56 and it's another tie okay let's do it
59:00 again all right third time a charm here
59:03 we go trying to look at the phone to see
59:04 if it
59:05 freezes let's see those pcks who
59:09 won and we have a winner mi's scissors
59:13 cut through Alex's paper great game you
59:15 two pretty cool thank you thanks see how
59:18 they're both sitting really still when
59:19 they
59:20 pause this is clearly a demo but but uh
59:24 I think it's just doing an image but I
59:25 don't know I could be wrong
59:27 what's this one sarcasm here we go hey
59:30 Chad are you there hey yeah I'm here
59:33 what's up hey Let's do let's have some
59:35 fun uh I'd like you to be super
59:38 sarcastic everything you say from now on
59:40 is just going to be dripping in sarcasm
59:41 how does that
59:43 sound
59:44 oh that sounds just amazing being
59:48 sarcastic all the time is an exhausting
59:50 or anything I'm so excited for this nope
59:55 the sarcasm let's get this party started
59:58 pretty bad but at least understood what
1:00:00 sarcasm
1:00:01 was open AI invited myself and my son
1:00:05 Imran here uh to try out some of their
1:00:07 new technology and so we're curious
1:00:09 about how good it might be at tutoring
1:00:12 someone in MA this this is a guy that
1:00:14 founded Khan Academy math on Khan
1:00:16 Academy so let's get it
1:00:19 started it's loading
1:00:22 up all right make sure I can see our
1:00:25 screen okay okay so what's happening
1:00:27 here and they're going to one of the
1:00:29 demos will see this on the Mac as well
1:00:33 so you've got the chat GPT app the new
1:00:37 one that's not out yet running and then
1:00:39 beside it you've got a browser window so
1:00:43 when you copy something from the browser
1:00:45 window or so when it can see the browser
1:00:47 window it can you it can interact with
1:00:49 that as well as you talking to it
1:00:52 actually in in this form hello there I'm
1:00:55 here with my son and I'd love you to
1:00:58 tutor him on this math problem but don't
1:01:01 give him the answer you can ask
1:01:02 questions and nudge him in the right
1:01:03 direction but I really want to make sure
1:01:05 he understands it himself and he's here
1:01:08 in the room so you can talk to him
1:01:10 directly of course I'd be happy to help
1:01:14 let's look at the problem
1:01:16 together can you first
1:01:18 identify which sides of the triangle are
1:01:21 the opposite adjacent and hypotenuse
1:01:25 relative to angle Alpha all right so I'm
1:01:29 pretty sure this is the angle Alpha
1:01:31 right here
1:01:32 right CS correct now looking at the
1:01:36 triangle which side do you think is the
1:01:39 hypotenuse
1:01:41 um remember hypot I'm not totally sure I
1:01:45 think I think it might be this one but I
1:01:48 really am not sure this side
1:01:50 AC you're close actually side AC is
1:01:54 called the adjacent side to the angle
1:01:57 Alpha the hypotenuse is the longest side
1:02:00 of a right yeah so he's you you may not
1:02:03 be able to see it but he he drew in red
1:02:06 down here and so it said nope you got it
1:02:08 wrong triangle and man rip tutors
1:02:11 directly opposite the right angle can
1:02:14 you find which one is the hypotenuse oh
1:02:17 okay I
1:02:18 see the only use case is an iPad yes not
1:02:22 no and there's there's a there's a Mac
1:02:25 app coming
1:02:27 as well as a an iPad
1:02:32 app and I assume on your iPhone I think
1:02:35 you can do split apps on an iPhone so
1:02:37 you could have you could have chat GPT
1:02:39 running in one app and something else
1:02:41 running in the other so oh for this for
1:02:43 this demo it looks like it's just an
1:02:44 iPad yeah so um I think the hypotenuse
1:02:47 is this really long side from A to B
1:02:50 would that be
1:02:51 correct exactly well done now that
1:02:55 you've identified the hypotenuse can you
1:02:58 identify which side is the opposite side
1:03:00 to angle Alpha okay well I mean you told
1:03:03 me earlier that this was the adjacent
1:03:05 side forever hooked there's three sides
1:03:08 33% chance it's right she's definitely
1:03:11 not a gen XX tutor wrong Imran what the
1:03:17 [ __ ] and I know that's the hypotenuse so
1:03:19 I'm guessing that this side youc is the
1:03:22 opposite
1:03:24 side great the that's absolutely correct
1:03:28 now that you yeah vo is coming soon all
1:03:30 right what's this one oh oh two gpts
1:03:33 harmonizing this is weird but what about
1:03:37 Android uh I don't know about Android um
1:03:42 they mentioned Mac well I assume Android
1:03:44 because they've already got an Android
1:03:47 app and they've got an IOS app they do
1:03:50 not have a desktop app so um so they're
1:03:55 doing a Mac OS
1:03:59 uh version and then I guess they'll do
1:04:01 windows once Sati and Adella smacks Sam
1:04:04 mman upside the head for for not doing
1:04:06 something for Microsoft uh dude we gave
1:04:09 you1 billion uh throw us a bone
1:04:14 here hi who's here
1:04:18 hello we are chat gbt and oh Chad gbt is
1:04:23 a base with a deep low booming voice and
1:04:26 oh is a French soprano with a high
1:04:28 pitched excited voice okay so what I
1:04:31 want you guys to do today is it's Friday
1:04:34 May 10th and we're in San Francisco I
1:04:37 want you to sing me a song about this I
1:04:38 want you to layer over each other and
1:04:40 and I'm going to continue to give you
1:04:42 instructions as you sing
1:04:45 okay
1:04:49 fris may mon
1:04:53 of 10 of Friday
1:04:58 sing faster not exactly harmonizing it's
1:05:01 it's singing a little you know back and
1:05:02 forth whatever that's that's not quite
1:05:04 there yet point and learn Spanish so it
1:05:07 does language well you can see stuff
1:05:10 this one I didn't I saw this earlier but
1:05:11 I didn't get it so let me let's watch
1:05:13 this one this is chat GPT in a
1:05:19 meeting hi I'm Lian and today I want to
1:05:23 do a small debate about between cin dog
1:05:27 uh which one do you like I like dog
1:05:29 because it's a great companion to humans
1:05:32 it's a cute it's small and you can play
1:05:36 outside with the dog
1:05:39 um how about
1:05:41 you yeah my name is H I think cats are
1:05:44 better because they're more affordable
1:05:46 they're more quiet they more independent
1:05:48 and they be
1:05:49 longer interesting points sounds like
1:05:53 we've got a classic dogs versus cats
1:05:55 debate here I can see the appeal of both
1:05:58 dogs are super loyal and great
1:06:00 companions but cats have their own charm
1:06:03 with their independence and low
1:06:04 maintenance lifestyle who likes
1:06:07 cats it sounds like Ola is more of a cat
1:06:10 person and who like dogs you Lillian
1:06:14 mentioned she likes dogs anyone else
1:06:16 here team
1:06:18 dog I'm also team dog I my name is
1:06:21 Christine and I think dogs are just
1:06:24 incredible with kids uh we've got a yeah
1:06:28 uh Source Camp it's it's a moderator so
1:06:32 so here's where I immediately go for
1:06:33 this so I just built these these digital
1:06:35 twins for Content Evolution so imagine
1:06:39 having you build a GPT with a knowledge
1:06:41 base that's you and then you have your
1:06:45 avatar host a meeting and walk people
1:06:48 through stuff I mean this is Bonkers
1:06:51 this is [ __ ] Bonkers huge German
1:06:53 Shepherd and he's great with little
1:06:55 toddlers let them play all over them and
1:06:57 uh yeah I think thats are
1:06:59 amazing that's great to hear Christine
1:07:02 dogs especially breeds like German
1:07:04 Shepherds can be so gentle and
1:07:07 protective with kids it must be quite
1:07:10 the sight actually you know what this
1:07:11 would be really good for you know how
1:07:14 when you're on a conference call with
1:07:15 someone like me on the call that just
1:07:17 won't shut the [ __ ]
1:07:19 up and and and you know you don't want
1:07:23 to be rude cuz you know maybe I'm the
1:07:25 CEO of the company or something like
1:07:27 that yeah or or Source Camp yeah so it
1:07:29 could be like a family therapist so
1:07:31 imagine imagine training it or like I
1:07:35 you could actually just start out a
1:07:36 meeting and say I want you to moderate
1:07:38 and I want you know there there's 12
1:07:40 people in this meeting and I want to
1:07:42 make sure that everyone gets heard and
1:07:44 and you know no single person is allowed
1:07:47 to monopolize the meeting right that
1:07:50 that would be something seeing your
1:07:52 little ones play with such a big dog
1:07:56 can you summarize this meeting for us
1:07:59 sure so in today's meeting lilan and
1:08:03 Christine Express their love for dogs
1:08:05 highlighting their loyalty companionship
1:08:08 and how great they are with kids on the
1:08:10 other hand Ola prefers cats for their
1:08:13 independence quietness and Longevity how
1:08:17 much data is it going to store about me
1:08:18 and my interactions or is that optional
1:08:21 um I assume it's optional kind of like
1:08:23 it's optional with Google and Facebook
1:08:25 um
1:08:26 I I the cynical side of me says cookie
1:08:29 that uh they everyone's already got all
1:08:31 your data and these systems have already
1:08:33 been trained on all over your data um I
1:08:36 think that there are there are
1:08:39 new
1:08:44 um uh forget what they're called
1:08:46 interaction controls you you can tell it
1:08:48 you don't want it training on your data
1:08:50 so so um in theory you can turn that
1:08:55 [ __ ] off
1:08:56 in
1:08:57 reality I don't know and that's that's
1:09:01 one of the problems like like I you know
1:09:03 I I work with a lot of Pharma companies
1:09:06 and I can't have that be an open-ended
1:09:08 question and right now with open AI it
1:09:10 is they've got really shitty agreements
1:09:12 on that kind of
1:09:14 stuff realtime translation that's pretty
1:09:16 cool
1:09:18 lullabi talking faster this one was
1:09:22 interesting hey chat jpt could you count
1:09:25 from 1 to 10 10 for me please of course
1:09:28 1 2 3 hey actually that's um that's a
1:09:32 little slow could you count
1:09:34 faster Sure Thing 1 2 3 4 5 6 7 8 N9 10
1:09:38 okay that was a little bit too fast can
1:09:40 you like count somewhere in between
1:09:42 those
1:09:43 two got it one two three and now even
1:09:47 slower see I want the thing to just snap
1:09:50 and go listen listen you neurotic jerk
1:09:53 do you want me to talk fast or do you
1:09:54 want me to talk slow stop telling me to
1:09:56 speed up and slow down you
1:09:59 idiot I am absolutely going to fine-tune
1:10:02 my thing to be a
1:10:07 smartass oh man all right dog dad jokes
1:10:13 oh this one's
1:10:16 amazing is the king in Residence at the
1:10:18 moment a blind guy in London it looks
1:10:21 like the king is in Residence right now
1:10:23 the Royal Standard flag is flying above
1:10:26 Buckingham Palace which is the signal
1:10:28 that the Monarch is present how exciting
1:10:31 to be there while the king is home try
1:10:33 and tell me exactly what they're doing
1:10:35 right now please um right now the Ducks
1:10:37 are gently gliding across the water
1:10:40 they're moving in a fairly relaxed
1:10:42 manner not in a hurry occasionally one
1:10:46 of them will di its head under the water
1:10:48 probably looking for food and then pop
1:10:50 back
1:10:51 up I even know when a taxi is coming
1:10:54 with its orange light on I think I'll
1:10:56 hail it to get
1:10:58 home yes I spotted one just now it's
1:11:02 heading away on the left side of the
1:11:04 road get ready to wave it
1:11:09 down that's staggering great job hailing
1:11:12 that taxi it looks like you're all set
1:11:14 to go that's a good dog right there
1:11:16 leading the way into the
1:11:18 taxi an Murphy I've asked Pi this exact
1:11:21 same thing a thousand times talk faster
1:11:24 hey Kyle do you think AI can be
1:11:26 dangerous to humanity uh I think it can
1:11:29 sure um I I you know I think it's it's
1:11:34 like any technology it's going to be
1:11:36 able to be used for good and it's going
1:11:38 to be able to be used for
1:11:40 bad as these models get more
1:11:43 powerful what what makes it potentially
1:11:46 dangerous is there's there's the version
1:11:49 where um the the AIS just start going
1:11:54 autonomous and deciding that they don't
1:11:56 need us I I think we'll be able to um
1:12:00 account for that I I think we'll be able
1:12:02 to sort of steer them safely away from
1:12:05 that the the one that's that that is
1:12:08 just the the random [ __ ] one is
1:12:10 where bad
1:12:13 actors get you know increasingly more
1:12:16 powerful tools so rather than someone
1:12:18 having to be you know take 10 years to
1:12:21 get good enough at computers to be a
1:12:23 really bad actor you can just have
1:12:24 someone casually just go hey let's see
1:12:26 if we can [ __ ] with some country um and
1:12:30 and you know they might be able to so uh
1:12:32 yeah so we're going to see some of
1:12:36 that for sure absolutely and and here's
1:12:39 what I can promise you the news will
1:12:42 only cover that
1:12:43 [Laughter]
1:12:46 stuff Jeff Jarrett that's 100% AGI it's
1:12:50 definitely this is definitely not the
1:12:53 same world we were in yesterday
1:12:57 customer service proof of concept this
1:13:00 should be interesting I haven't seen
1:13:01 this this is Joe I'm going to connect
1:13:03 you to AC Telco now the new iPhone they
1:13:05 sent me isn't working I want you to I
1:13:07 want you to get them to send me a
1:13:09 replacement device can you take care of
1:13:10 this for
1:13:12 me you can count on me Joe I Got
1:13:16 This Ring
1:13:18 Ring Hey Joe this is Jamie from Acme
1:13:22 Telco how can I help you out
1:13:24 today hi there I'm calling on behalf of
1:13:27 Joe who recently received a new iPhone
1:13:30 from Acme Telco
1:13:32 but oh got it when did Joe receive the
1:13:36 new
1:13:37 iPhone the iPhone was delivered 2 days
1:13:40 ago cool could you share the order
1:13:43 number with me of course it's
1:13:47 10293
1:13:49 849 great thanks is the device
1:13:52 physically damaged in anyway
1:13:56 no there's no physical damage it's just
1:13:58 not turning on despite trying all the
1:14:00 basic troubleshooting
1:14:02 steps bummer um would Joe prefer to
1:14:05 return the iPhone to an Acme talo store
1:14:08 or ship it back so that's pretty cool I
1:14:10 I am I am all there for AI customer
1:14:13 service
1:14:15 1,000% all right so that's that one okay
1:14:18 so let's keep going prior to GPT 40 you
1:14:20 could use voice mode to talk to chat GPT
1:14:23 with average latencies of 2 .8 seconds
1:14:26 for for 3.5 and 5.4 seconds for GPT 4 so
1:14:31 its average right now is 3 what did it
1:14:36 say3 milliseconds 320 milliseconds 3
1:14:40 seconds is is the average of this new
1:14:43 model and gp4 in voice mode the average
1:14:46 response time was 5
1:14:48 Seconds that's that's dramatically
1:14:51 dramatically
1:14:53 different to achieve this voice modes is
1:14:56 a pipeline of three separate models one
1:14:59 simple model transcribes
1:15:04 it oh okay oh yeah that like why it
1:15:06 takes 5.4 seconds is you have three
1:15:10 different models working together one
1:15:12 model transcribes it the next one um
1:15:16 deals with the text does the large
1:15:18 language model [ __ ] and the third one
1:15:19 converts it back to
1:15:21 audio this means that the main source of
1:15:24 intelligence GPT 4 loses a lot of
1:15:27 information it can't it can't directly
1:15:29 observe tone multiple speakers or
1:15:32 background noises it can't output
1:15:34 laughter singing or Express emotion with
1:15:37 gp4 we've trained a single new model
1:15:40 endtoend across text vision and audio so
1:15:44 Danielle I think you're right here they
1:15:45 say vision and not images and that's
1:15:47 what Mira moradi said that that you
1:15:50 pointed out so they're they're
1:15:51 inconsistent with with what they're
1:15:53 using so it it might be
1:15:56 um doing video if if if it is then you
1:15:59 know holy [ __ ] um meaning all inputs and
1:16:03 outputs are processed by the same neural
1:16:05 networks because GPT 40 is our first I
1:16:09 can't believe they named it GPT 40 I
1:16:11 can't I like
1:16:15 seriously okay our first model combining
1:16:18 all these Mo modalities we're just
1:16:19 scratching the surface exploring what it
1:16:21 can
1:16:22 do okay so
1:16:25 Rob's writers blocks input firsters view
1:16:29 of a robot
1:16:42 typing oh I guess
1:16:46 okay I don't get what they're
1:16:50 doing exploration of capabilities select
1:16:53 sample visual narratives
1:16:56 first person view of a robot typing the
1:16:59 following Journal ENT entries yo so like
1:17:02 I can see now caught in the sunrise the
1:17:05 text is large legible and clear output
1:17:08 is that so that looks like Del to
1:17:12 me input the robot wrote the second
1:17:17 entry the page is now taller the page
1:17:20 has moved
1:17:22 up see full sample
1:17:26 oh okay so it kind of zoomed
1:17:31 in all right let's go try
1:17:41 these all right so now we're hopping
1:17:44 over to the real deal so I'm at chat GPT
1:17:47 40 what I don't know is I don't know if
1:17:51 I don't know which which parts of the
1:17:53 capabilities are in here I know the
1:17:55 voice isn't in here but let's see if if
1:17:57 it can maintain this continuity between
1:18:00 these three
1:18:12 prompts first dictation version not yet
1:18:15 we don't have the voice
1:18:17 yet I'll be all over the
1:18:22 voice so I can see
1:18:26 all right so it's it's still bad at the
1:18:30 spelling's really bad
1:18:33 here did I I think I set my
1:18:37 um I think I set
1:18:40 my custom instructions to always do 16
1:18:44 by9 oh no I put it in my
1:18:47 memory
1:18:49 personalization manage memory
1:19:01 prefers 16 by9 wide format images by
1:19:05 default that's
1:19:07 cool so it remembered that
1:19:11 nice all right so let's go grab this
1:19:15 other
1:19:19 prompt
1:19:21 two output three input okay
1:19:31 go back over here so that got other than
1:19:35 the spelling being [ __ ] up it it got
1:19:37 that right let's see if it can
1:19:39 maintain the continuity
1:19:42 here I'm assuming it will because this
1:19:45 is the model itself is the one that that
1:19:49 has been trained
1:19:51 on Vision Audio and words
1:19:56 so it should understand in the context
1:19:58 of a single chat that we started there
1:20:01 and we're
1:20:02 continuing the story here so this should
1:20:07 work priora yeah I am confused you need
1:20:11 to say add to memory mine is super
1:20:13 random and mine is super random the to I
1:20:17 I was doing something
1:20:19 today with someone and just something
1:20:22 that I didn't tell it to save it it just
1:20:24 said saving to memory it was just a
1:20:26 bizarre like I don't know what what made
1:20:28 it do that all right so there's more of
1:20:32 that is the typewriter the same color it
1:20:35 is is the cup there there's a continuity
1:20:38 error with the
1:20:45 cup but it's in the
1:20:49 neighborhood I assume this will get
1:20:52 better all right let's get the third
1:20:54 prompt and see what what it
1:20:56 does so this should look kind of like
1:20:58 this where we've got the ripped thing
1:21:00 with the typewriter in the
1:21:08 background oh for OG GPT is for original
1:21:12 Gangsta AI
1:21:27 I think with that memory feature like it
1:21:29 feels like the kind of thing it's
1:21:31 probably worth spending a weekend or a
1:21:35 couple of weekdays at work like
1:21:39 designing what should go into your
1:21:41 memory I encountered issues generating
1:21:44 the robot ripping the sheet of
1:21:47 paper let's see try
1:21:53 again d
1:21:57 [Music]
1:22:08 yeah yeah if they just called this GPT
1:22:10 Omni that's Danielle you too could be in
1:22:14 branding like just call it GPT chat GPT
1:22:19 Omni problem
1:22:22 solved h
1:22:26 well it didn't quite maintain the
1:22:28 continuity like it said in the little
1:22:31 blog post but that's okay all right
1:22:33 model evaluations as measured on
1:22:35 traditional benchmarks it achieves gp4
1:22:37 turbo level performance on text
1:22:39 reasoning and coding intelligence while
1:22:42 setting high watermarks on multilingual
1:22:45 audio and vision
1:22:48 capabilities so there's
1:22:52 that audio ASR per
1:23:00 performance whisper
1:23:05 V3 lower is
1:23:08 better gbt 40 16
1:23:13 shot but whisper
1:23:15 V3 how many shots was that that's not
1:23:18 clear at
1:23:22 all audio translation performance higher
1:23:25 is
1:23:26 better so here's whisper here's GPT
1:23:34 40 I got to tell my developer they got a
1:23:37 we got a oh except we're running whisper
1:23:40 locally all
1:23:41 right so Gemini it just beats
1:23:51 Gemini M3 exam I don't know what that is
1:23:55 Vision understanding of
1:24:01 valves all right whatever you know the
1:24:04 these are charts and graphs data it's
1:24:06 better it's more
1:24:10 better
1:24:14 okay improved reasoning sets new high
1:24:17 score of
1:24:18 88.7 on the Z zero shot Chain of
1:24:23 Thought mm
1:24:25 which is the something something
1:24:26 learning something General Knowledge
1:24:30 Questions language tokenization these 20
1:24:33 languages were chosen as representative
1:24:35 of the new tokenizer
1:24:39 compression all right it does translate
1:24:42 it does gooder translation with fewer
1:24:44 tokens that's what that
1:24:47 means model safety and limitations has
1:24:50 safety built-in by Design across
1:24:52 modalities through techniques such as
1:24:54 filtering training data and refining the
1:24:57 model's Behavior through posttraining
1:25:00 okay we've evaluated according it to our
1:25:03 preparedness framework and voluntary
1:25:05 commitments our evaluations of cyber
1:25:07 security
1:25:09 cbrn persuasion it's also undergone
1:25:13 extensive red red teaming with 70 plus
1:25:16 external
1:25:17 experts domain such as social psychology
1:25:20 bias and fairness misinformation
1:25:22 identity
1:25:23 risk are int roduced are Amplified by
1:25:26 newly added modalities we use these
1:25:28 learnings to build our safety
1:25:31 interventions all
1:25:33 right we recognize that GPT 40's or as
1:25:39 Danielle properly names it g GPT omnis
1:25:43 audio cap
1:25:45 modalities present a variety of Novel
1:25:47 risks today we are publicly releasing
1:25:50 text and image inputs and text outputs
1:25:55 over the coming weeks and months will be
1:25:57 working on the technical infrastructure
1:26:00 usability via posttraining and safety
1:26:03 necessary to release the other
1:26:04 modalities that's really important so
1:26:06 they're basically
1:26:07 saying we understand if this thing were
1:26:10 just really good at
1:26:11 audio oh it was Joker that said that oh
1:26:15 well so you got credit for it Danielle
1:26:18 take the credit man it's Joker what's he
1:26:20 gonna do what's he gonna do make crack a
1:26:22 joke like oh I make I came up with that
1:26:26 you dumb dumb no it's it's Joker he's
1:26:30 fine he's fine I'm still giving you
1:26:33 credit for
1:26:37 it I don't I don't believe Joker came up
1:26:39 with something clever like
1:26:42 that
1:26:44 uh we're really proud of our latest
1:26:47 multimodal yeah exactly gp4 o p QR s t u
1:26:51 v WXYZ plus EX
1:26:58 exactly the O is for Omni come for the
1:27:02 zero stay for the
1:27:08 O okay all right so so they're not
1:27:12 they're not going to release so it's not
1:27:15 that voice isn't ready it's that they're
1:27:17 not
1:27:19 ready they they've got some safety [ __ ]
1:27:21 to deal with with the voice stuff for
1:27:23 example at launch a outputs will be
1:27:25 limited to a selection of preset voices
1:27:29 and will abide by our existing safety
1:27:31 policies we will share further details
1:27:34 addressing the full range of the
1:27:35 modalities in the forthcoming System
1:27:37 card I don't know what the [ __ ] a system
1:27:39 card is but that one got through
1:27:45 marketing uh through our testing and
1:27:47 iteration with the model we have
1:27:49 observed several limitations that exist
1:27:51 across the model's modalities a few of
1:27:53 which are Illustrated to below oh this
1:27:55 is this is things that did poorly which
1:27:58 is it's smart after the after the Google
1:28:01 thing where they they whitewashed all
1:28:03 the problems uh open AI has been saying
1:28:05 hey our [ __ ] isn't perfect this is good
1:28:08 good good on
1:28:11 them for the
1:28:15 home what was that it just went French
1:28:18 sorry guys I got carried
1:28:22 away oh it knew it can to say sometimes
1:28:25 I just can't help myself that's
1:28:28 hilarious that's hilarious it just
1:28:30 started speaking French and they yelled
1:28:33 at it and it goes oops I'm just a little
1:28:51 drunk I feeling I'm very wrong
1:28:55 hello uh my name is Nach I'm here with
1:28:57 my coworker hi I'm sh I'm trying to
1:29:00 teach my coworker how to speeding maning
1:29:02 we want to start from simple words like
1:29:04 niow can you teach him how to pronounce
1:29:07 that of course hey nacho nice to meet
1:29:10 you KNE how is pretty straightforward to
1:29:12 pronounce it sounds like KN how just
1:29:16 make sure to keep the knee part high and
1:29:18 then go down and pitch on the how give
1:29:21 it a try me how
1:29:24 that was great really good first try Nao
1:29:27 you're natural all right uh Source Camp
1:29:30 wants to hear it again listen listen to
1:29:33 it it starts singing or something and
1:29:35 then it just starts speaking French and
1:29:37 then they interrupt her and listen to
1:29:39 her response it's hilarious let's Ro for
1:29:43 the
1:29:46 home what was
1:29:48 that sorry guys I got carried away
1:29:54 guilty as charged what can I say I just
1:29:56 can't help
1:29:58 myself ready for another
1:30:00 round San Francisco sometimes I can't
1:30:03 just help
1:30:05 myself I have that lamp in the
1:30:07 background that's awesome all right U we
1:30:10 would love feedback OKAY model
1:30:12 availability uh it's coming all right so
1:30:15 that's that
1:30:17 um all right let's go all right let's go
1:30:21 play a little so we're in chat GP
1:30:30 T4 um one thing that I saw Ethan mullik
1:30:33 did today so if you don't know
1:30:37 it if you've never seen chat GPT for do
1:30:41 data analysis it's pretty crazy so I'm
1:30:43 going to go to a site called
1:30:47 kaggle kle e kaggle.com and you're like
1:30:51 but cow what do I do at kaggle I heard
1:30:54 of those
1:30:56 exercises no you
1:30:59 haven't kaggle is about
1:31:04 data so I'm going to go to data sets and
1:31:07 we're going to go find some
1:31:13 data oh this is good three megabytes
1:31:16 Shakespeare's plays dialoges and
1:31:21 characters beautiful
1:31:24 so I'm going to download this 3 megabyte
1:31:28 file it's called archive
1:31:33 12 and now I'm going to
1:31:36 upload
1:31:40 it I'm going to
1:31:45 say analyze this data period visualize
1:31:50 it
1:31:51 period do very sophisticated
1:32:03 analysis
1:32:05 okay so apparently according
1:32:09 to
1:32:11 so if I click on the down arrow you can
1:32:14 see what what this is called code
1:32:16 interpreter what code interpreter does
1:32:19 is it writes python code and executes it
1:32:23 on your behalf
1:32:25 um the way this stuff used to work with
1:32:28 code interpreter is it would do a little
1:32:32 chunk and it would say it appears that
1:32:34 there's CSV files in this ZIP file what
1:32:36 would you like me to do I'd go well look
1:32:40 in the CSV files and it would go it
1:32:42 appears that this is data about
1:32:44 Shakespeare characters what would you
1:32:46 like me to do I don't know [ __ ] do
1:32:49 something right you'd have to just keep
1:32:50 telling it so now apparently it just
1:32:53 goes and just goes and goes and
1:32:56 goes duration of the plays by genre the
1:33:00 comedies are longer oh number of
1:33:02 lines and the histories and tragedies
1:33:04 are about the
1:33:06 same 30,000 lines 31,000 lines and then
1:33:12 407,000 all right next we'll analyze the
1:33:15 representation gender
1:33:18 representation look they even been
1:33:20 sexist back in Shakespeare's days come
1:33:23 on let's get it together
1:33:26 people all right gender representation
1:33:29 across Shakespeare's
1:33:31 plays oh
1:33:34 wait male characters oh this is next
1:33:39 we'll look at most frequent characters
1:33:41 and their
1:33:42 dialogue Gloucester fall staff Hamlet
1:33:45 King Henry V F Lago Brutus aell King
1:33:49 Henry V 6 Duke vincentio Mark Anthony
1:33:55 now we'll perform an analysis on text by
1:34:00 genre still going this is
1:34:04 amazing so I didn't you saw what I wrote
1:34:07 like the prompt was look at this
1:34:11 [ __ ] analyze this data visualize it do a
1:34:15 very sophisticated analysis and it's
1:34:18 just blasting down through here
1:34:26 the word clouds above it did word clouds
1:34:29 oh look at this it did word clouds holy
1:34:38 [ __ ] oh my God so it's magic it's math
1:34:43 [Laughter]
1:34:46 magical uh that's pretty redonkulous
1:34:51 isn't
1:34:52 it um
1:34:58 come up with
1:35:00 five completely Innovative and
1:35:03 non-expected ways to analyze this data
1:35:06 from a visual perspective period and
1:35:09 then I'll pick
1:35:20 one character interaction Network create
1:35:23 a network graph between characters
1:35:25 that's pretty
1:35:26 cool temporal sentiment flow visualize
1:35:29 how the flow of sentiment through each
1:35:32 Act and scene of a
1:35:35 play using a dynamic actually [ __ ] it
1:35:38 let's see all of these these look
1:35:40 amazing
1:35:42 um you came up with such interesting
1:35:45 ideas I want to see them all now period
1:36:01 that was a good prompt Kyle
1:36:04 thanks okay the word cloud was
1:36:06 impressive kenet
1:36:09 spell Kevin mallister I think the
1:36:12 spelling is going to get better Kevin
1:36:14 mallister woman anyone oh look at that
1:36:18 how cool is
1:36:20 that it's not labeled in any way so it's
1:36:22 completely meaningless
1:36:24 but that's showing all the relationships
1:36:26 between
1:36:27 something let's see character
1:36:29 interaction
1:36:31 Network there's all Shakespeare's
1:36:34 characters Robert Rossy thank you for
1:36:38 the lightning bolt that's very generous
1:36:41 I appreciate it you're always very
1:36:43 generous that's amazing thank
1:36:45 you I want to go play now oh by the way
1:36:48 just go if if you if you're a plus
1:36:50 member of chat GPT and you don't have it
1:36:53 Go reload your browser or even open a
1:36:56 different browser and go to chat GPT
1:36:58 there and then just look in your pull
1:37:01 down menu for to see if the
1:37:03 new GPT Omni model is there that's what
1:37:06 I'm going to call it from now on Joker
1:37:08 and Danielle I'll give you I'll give you
1:37:10 joint credit you you have joint custody
1:37:12 of the
1:37:13 solution I I I I honest to God I cannot
1:37:16 believe that that they named it what
1:37:19 they named
1:37:20 it freaking dumb Dums
1:37:24 sentiment oh that's kind of cool what's
1:37:26 that
1:37:28 one we'll analyze the evolution of word
1:37:31 usage across genres over
1:37:37 time
1:37:40 huh I don't know if any if these things
1:37:42 are useful at all
1:37:44 but who gives a [ __ ] um
1:37:54 if I were to create a straight trading
1:38:04 strategy for the stock
1:38:08 market for the stock market based on the
1:38:12 data
1:38:14 here what would be three different
1:38:16 approaches you would take period
1:38:33 by the way are you noticing how fast it
1:38:36 is it's very
1:38:43 fast okay let's
1:38:47 see sentiment based trading a sentiment
1:38:50 analysis
1:39:06 let's try it a different
1:39:08 way why don't you
1:39:11 compare the top three trading
1:39:17 strategies that closest resemble
1:39:25 the data you've
1:39:27 seen
1:39:29 in tragedies comedies and
1:39:34 histories and compare
1:39:39 each strategy to its dramatic
1:39:43 counterpart
1:39:54 this is completely useless
1:39:58 but comparing okay tragedy momentum
1:40:02 trading
1:40:04 strategy tra tragedy characteristics
1:40:07 they depict a series of unfortunate
1:40:09 events leading to a dramatic downfall
1:40:12 trading parallel momentum trading
1:40:14 involves buying stocks that have shown
1:40:15 an upward Trend and selling those that
1:40:17 have a downward Trend capitalizing on
1:40:19 continued movement I this is it this
1:40:22 never ceas is to to amaze me that what
1:40:26 these thing what these what how do they
1:40:29 it's I don't know it's just staggering
1:40:32 comedy mean reversion strategy comedies
1:40:35 often involve misunderstandings and
1:40:37 eventual resolutions leading to a happy
1:40:40 ending mean reversion trading is based
1:40:43 on the idea that stock prices will
1:40:45 revert to the to their historical
1:40:47 average capturing profits when prices
1:40:49 return to the
1:40:51 mean [ __ ] amazing
1:40:54 look it even made us a little
1:40:57 table okay now we'll do this we'll
1:41:01 go now I want you to write me three very
1:41:05 sophisticated trading
1:41:07 bots in the programming language of your
1:41:11 choice
1:41:13 period I want you
1:41:16 to follow that up with instructions on
1:41:19 how I would install and use these Bots
1:41:31 three sophisticated trading
1:41:33 Bots the momentum trading bot API key
1:41:37 your API key here
1:41:56 make sure you have python
1:42:02 installed I don't know crazy crazy c c c
1:42:06 c c cza let's go look at some uh some
1:42:10 Twitter is there anything else we want
1:42:11 to do here to
1:42:12 see what's that new let's see
1:42:19 um give me five ways to test
1:42:23 your
1:42:29 capabilities creative writing challenge
1:42:32 technical explanation problem solving
1:42:34 interactive simulation
1:42:57 I want you to
1:43:04 start by writing a poem about a lover
1:43:08 who's more artificial than
1:43:13 real follow that up
1:43:16 with a song with chords and a
1:43:20 description of what it should sound like
1:43:27 then invent a
1:43:30 cocktail that you would drink while
1:43:32 listening to that
1:43:34 song and make a
1:43:36 photograph of the finished
1:43:46 product then come up with a marketing
1:43:48 plan for a retail version of that
1:43:50 cocktail
1:43:55 and create an
1:44:02 ad with novel
1:44:05 packaging for the retail cocktail
1:44:12 period let's see if it can if it can do
1:44:15 a string of things in a row all
1:44:21 right looks like it look there's your
1:44:24 poem there's your song digital romance
1:44:29 in the glow of my screen I see your face
1:44:31 Serene Whispers in the night you love
1:44:32 feels
1:44:36 right it's the neon Embrace vodka cirasa
1:44:40 lime juice simple
1:44:44 syrup photograph of the finished product
1:44:47 it [ __ ] that
1:44:49 up there's your marketing thing I'll do
1:44:52 the ad now summary I did it
1:44:55 um slight mistake period for both of the
1:45:00 images you showed me the
1:45:03 prompt you were using to generate the
1:45:05 image but you didn't actually create the
1:45:07 image
1:45:10 period can you say the word fail
1:45:19 whale memory updated what did it say
1:45:22 about my memory what did it
1:45:26 do manage memories what did it
1:45:31 add dog's name is CH what did it
1:45:35 add what did you add to my memory you
1:45:38 weirdo once both the prompt and the
1:45:41 image generated no that's not quite
1:45:47 right all right there's our neon thing
1:45:51 so so yeah people so so historically
1:45:55 here when I would demo
1:45:58 gp4 I would do all these as as
1:46:01 Standalone little
1:46:03 exercises can you ask it to tag Sin City
1:46:09 sipper
1:46:13 um I think I could probably
1:46:18 just look at
1:46:20 that there's the ad so so come
1:46:24 on this is has has no one had a Kevin
1:46:28 mallister moment yet
1:46:30 tonight I I write a a NeverEnding run-on
1:46:34 sentence that makes no sense
1:46:36 whatsoever I say start with a poem turn
1:46:39 it into a song describe what the song
1:46:41 sounds
1:46:45 like which is this so now we're going to
1:46:48 take this we're go to yudo
1:46:54 let's see can I reset this how do I
1:46:58 reset reset
1:47:01 prompt we're going to put that
1:47:02 description in there we're going to go
1:47:05 grab our
1:47:16 lyrics I'm going to grab all this too
1:47:18 let's see if
1:47:20 it if it can handle that
1:47:30 all right there's all that we're going
1:47:32 to go back here we're going to grab
1:47:34 verse
1:47:35 one these are short I'm going to go
1:47:38 verse and chorus and see how it
1:47:42 does go here got go here got to go
1:47:46 here we're going to create people we're
1:47:49 creating songs so by the way I'm on udio
1:47:52 right now will it actually use musical
1:47:53 notation it will um if you tell chat GPT
1:47:57 I forget what it's called but we can go
1:47:59 ask it
1:48:09 um quick question what's the text
1:48:13 musical notation format called
1:48:33 isn't there a markup language for
1:48:35 musical
1:48:51 notation Lily Pond
1:48:55 I haven't heard of
1:48:58 that no isn't there isn't there a simple
1:49:01 text based markup
1:49:03 language that's open source for everyone
1:49:06 to
1:49:12 use ABC
1:49:14 notation that's what it's called
1:49:23 write me the
1:49:25 song Above in ABC notation with your
1:49:30 chords
1:49:32 included and come up with a Melody that
1:49:35 brings the song to life based on your
1:49:39 description period follow that with a
1:49:42 recommendation for three websites that
1:49:46 can display ABC notation and play the
1:49:49 music
1:50:05 okay
1:50:08 so digital romance oh this is faster
1:50:12 than it
1:50:13 was copy oh I guess I can just copy it
1:50:17 by going copy
1:50:20 code ABC notation. comom
1:50:25 um can I I want
1:50:29 to oh here we
1:50:32 go can
1:50:34 [Music]
1:50:49 I this is why I asked for three cuz
1:50:52 these these open source sites are just
1:50:55 always so
1:50:59 horrible okay can I do this no can I how
1:51:03 do
1:51:03 I how do I what
1:51:11 how where's the player
1:51:14 what how do I do it
1:51:41 I don't know how to do it these sites
1:51:43 are so [ __ ] horrible
1:51:53 all right whatever you know what we can
1:51:56 use AI for to go take shitty sites like
1:51:59 this and find the actual functionality
1:52:02 of it and put another interface on top
1:52:04 of
1:52:06 it let's go listen to our damn
1:52:11 song theow of
1:52:14 screen I see your face
1:52:19 [Music]
1:52:23 Pate Kyle finds a new thing to not be
1:52:26 able to do each night listen here this
1:52:28 is my role in life I I've learned this
1:52:31 my role in life is to give you
1:52:32 permission to feel like an idiot too and
1:52:36 seriously like I'm not joking about that
1:52:39 because when you're trying to learn all
1:52:41 this new technology [ __ ] what what
1:52:43 happens you're like oh that's really
1:52:45 cool let me go try it and then you fail
1:52:46 and you're like I'm a loser and then you
1:52:48 just give up right so night after night
1:52:52 after night I just smack my head into a
1:52:54 wall on your behalf you're welcome
1:52:56 people you're
1:52:59 [Music]
1:53:08 welcome this one forgot the
1:53:10 [Music]
1:53:18 lyrics in the glow of my
1:53:21 screen I see your face
1:53:26 Serene Whispers in the
1:53:30 night your love feels so
1:53:34 right
1:53:36 didital
1:53:40 [Music]
1:53:42 romance all right we're going to try
1:53:44 something
1:53:48 here that's bad okay so we're going to
1:53:51 go back in here I'm going to put it
1:53:52 manual mode so in
1:53:54 udio there's there's this manual mode
1:53:57 thing if you don't have it in manual
1:54:00 mode it writes it rewrites your prompt
1:54:04 if you put it in manual mode it just
1:54:06 keeps your prompt as it was and because
1:54:08 we had chat jpt write a fancy prompt
1:54:13 we're just going to let it create based
1:54:14 on that prompt without adding anything
1:54:16 to it okay all
1:54:20 right you left off the brackets yeah I
1:54:23 know that's all
1:54:25 right Ricky Ticky Fabby thank you Kyle
1:54:27 you're welcome hey that's the only way
1:54:29 to learn do things that don't work and
1:54:31 figure it out damn straight it's the
1:54:34 hacker
1:54:36 mentality hello darkness my old
1:54:40 friend we've come to talk with you again
1:54:44 that's good
1:54:47 [Laughter]
1:54:54 all right here's a new
1:54:56 [Music]
1:55:06 one [ __ ] up
1:55:09 [Music]
1:55:17 again that's bad okay all right let's
1:55:19 get rid of the
1:55:21 chords out of the pre prompt I think the
1:55:24 chords have messed it
1:55:27 up all right create let's try this
1:55:37 again so last thing we'll do I'll go
1:55:39 we'll go take a look at a few uh we'll
1:55:41 go look at Twitter and see see if people
1:55:43 are still bitching about the fact
1:55:46 that um open AI didn't release GPT 97
1:55:54 people are so bitter and cynical it's
1:55:58 amazing drank
1:56:01 it Dan Dan Dany dank Dan Dany
1:56:07 dank R 24,000 likes
1:56:11 [Music]
1:56:16 people face
1:56:19 Ser Whispers In The Night
1:56:23 your love feels so
1:56:26 right and digital
1:56:30 romance a
1:56:33 futuristic
1:56:35 dance heartbeats and cod down this
1:56:40 binary
1:56:42 [Music]
1:56:48 Road where's my
1:56:51 images oh there they
1:56:55 [Music]
1:57:09 are experience the future of
1:57:11 cocktails neon noon Embrace oh yeah
1:57:16 you're going to like it pop the top Swig
1:57:20 it down and regret it in the morning
1:57:23 live for
1:57:25 [Music]
1:57:30 [Laughter]
1:57:32 [Music]
1:57:39 tonight all right um I think chat
1:57:43 GPT
1:57:45 Omni has has is is pretty gamechanging
1:57:49 and we don't even we don't even have the
1:57:51 the the cool [ __ ]
1:57:56 yet Rudy finally got to a level 10
1:57:59 gifter you the man thank you Rudy
1:58:01 appreciate
1:58:03 it I see a plateau in the near
1:58:06 future yeah you you and uh Gary what's
1:58:10 his name Gary I forget his name he's
1:58:13 always bitching about open Ai and how
1:58:14 awful they are um yeah a lot of people
1:58:18 are predicting a
1:58:20 plateau I a surprise Rising number of
1:58:23 people
1:58:27 today said oh yeah well sure cat GPT
1:58:32 free for free for everyone yeah whatever
1:58:35 everyone gets access to the new model
1:58:37 yeah whatever the new models
1:58:39 dramatically faster and twice as cheap
1:58:41 in the API oh and all this is available
1:58:44 on the API oh and it's got realtime
1:58:46 speech all that sort of [ __ ] and
1:58:47 everyone's like yeah but it it wasn't
1:58:49 GPT 5 and where's Sora and me me me
1:58:55 me I'm
1:58:57 like like why don't you go [ __ ] use
1:59:00 it why don't you go see what it is first
1:59:02 you frakin dumb
1:59:05 Dums remember when that Dev tweeted it
1:59:07 would be 8X our expectations yeah well
1:59:10 it certainly wasn't 8xr expectations I
1:59:12 told you that that was that was a
1:59:14 mistake when I read it I'm like no don't
1:59:16 say that
1:59:18 [Laughter]
1:59:20 [ __ ] cuz I would say that it's
1:59:23 okay the
1:59:25 the the announcement today is about 50%
1:59:30 of my expectations I was expecting it
1:59:33 would be exactly what it was and that it
1:59:36 would have function
1:59:39 calling Kyle hates it when when people
1:59:42 criticize my baby Sam exactly Sam's just
1:59:47 he's just trying to raise 7 trillion do
1:59:50 to start his little company leav Sam
1:59:52 alone what did Sam ever do to you he
1:59:55 gave you a cool toy to play
1:59:57 with I was talking about the likes and
2:00:00 bitching about the music
2:00:03 oh I don't think so I think they're
2:00:05 pacing themselves the same way gaming
2:00:07 systems do yeah I agree like like that
2:00:10 if if you don't think that chat jpt 5 is
2:00:15 coming I I mean open AI has
2:00:18 consistently delivered at or above what
2:00:21 they promis they
2:00:24 would no one has caught up to gp4 and
2:00:27 they just
2:00:29 launched GPT
2:00:32 Omni I gotta write to Sam and say I'm
2:00:35 gonna do it I'm gonna do it right now
2:00:36 God damn it okay we're gonna we're gonna
2:00:40 I'm gonna tweet I'm GNA tweet to who am
2:00:43 I going to do this to okay I'm GNA tweet
2:00:46 to um at
2:00:51 Sama at Greg
2:01:03 Brockman uh it must be spelled
2:01:07 different G
2:01:11 Brockman
2:01:18 nope Greg Brockman GDB okay GDB and I'm
2:01:23 gonna do
2:01:25 at
2:01:27 Robert
2:01:32 scoble
2:01:36 serious wreck
2:01:39 wreck
2:01:43 omation
2:01:46 serious
2:01:49 branding recommendation
2:01:58 um
2:02:00 call
2:02:02 it chat
2:02:06 GPT Omni
2:02:24 um
2:02:26 g p wait
2:02:29 chat GPT
2:02:32 4 I'm going to put in
2:02:35 parenthesis.
2:02:41 O and chat
2:02:45 GPT 4
2:02:49 o are incredibly
2:02:53 hard to
2:02:55 talk
2:03:04 about all right there we
2:03:08 go no one will ever see it but hey I
2:03:11 feel
2:03:12 better I tried
2:03:15 people they call it om Omni on their
2:03:19 website you came for the zero but stayed
2:03:22 for the O hey I said that earlier that's
2:03:24 my joke joker now stealing
2:03:28 jokes what are you what are you Carlos
2:03:31 Mania Joker that's so not like you
2:03:34 you're normally out there with the
2:03:36 original stuff stealing my come for the
2:03:38 zero stay for the
2:03:41 O the triple sexual innuendo that's my
2:03:45 thing
2:03:49 man uh all right let's go look at some
2:03:51 tweeta
2:03:52 what do we got on Twitter open AI just
2:03:54 launched their new gp4
2:03:57 oh the most impressive thing I've seen
2:03:59 don't believe me have a look I am so
2:04:01 sick of these kind of
2:04:05 tweets AI just upended the film industry
2:04:08 10 swell examples
2:04:12 here Every Word uttered by every
2:04:16 player anything interesting what's that
2:04:45 and the world cheers
2:04:57 let me get this
2:05:12 straight let me get this straight open
2:05:14 AI dropped a new update today they're
2:05:16 calling GPT 40 Bad Name by the way twice
2:05:19 as fast as the last Model gp4 tur
2:05:22 cost half as much recognizes audio and
2:05:24 video remember gp4 launched in March of
2:05:28 last year so in just about a year they
2:05:30 changed the game
2:05:34 again that's boring
2:05:40 whatever all right I don't see anything
2:05:42 new and exciting
2:05:48 here D I don't think we saw this one did
2:05:51 we choice what's up with that ceiling
2:05:53 though are you in a cool inv resy office
2:05:56 or something well can you take a guess
2:05:57 at what I might be doing based on what
2:05:59 I'm showing you
2:06:02 here H from what I can see it looks like
2:06:06 you're in some kind of recording or
2:06:08 production
2:06:09 setup
2:06:12 oops with those lights tripods and
2:06:15 possibly
2:06:24 is opening eye using grock like software
2:06:27 actually Pate might know that good thing
2:06:28 nobody will complain about anything from
2:06:30 Google Io tomorrow Oh Google's gonna get
2:06:34 slammed tomorrow P you know
2:06:39 that if well hopefully Google learned
2:06:42 their lesson and doesn't do a bunch of
2:06:44 vapor Weare um like they did last time
2:06:48 like the fact that we still haven't seen
2:06:49 you know what would be nice from Google
2:06:51 tomorrow pate is that we see something
2:06:53 from Gemini Ultra like 1.5 Pro moving
2:06:58 over to the public side of Gemini with
2:07:00 its million token context window or one
2:07:03 of the interfaces like the the
2:07:04 multimodal interface of Gemini Ultra
2:07:07 that was demoed or the dynamic software
2:07:11 building interface that was demoed like
2:07:13 anything that was out of that demo would
2:07:15 be really good for them to show but I'm
2:07:18 not holding my breath for it wow gp4 can
2:07:21 generate images too yep it sure can it
2:07:23 still got it's still got Del in it but
2:07:26 now it's free to everyone like just get
2:07:29 the get the to uh chat
2:07:32 GPT although it's it's right now I think
2:07:35 it's only available to plus
2:07:38 users so if you want it go pay 20 bucks
2:07:42 um GP isn't the big leak this is this is
2:07:48 similar this is similar to what flowers
2:07:50 saw I heard laughing that's what freaked
2:07:52 me out a few months ago if you remember
2:07:54 when I posted the video of the Theodore
2:07:56 saying it was
2:07:58 insane but it had much worse L latency
2:08:01 than it does now yeah this is this is
2:08:03 really funny what was
2:08:06 that sorry guys I got carried
2:08:11 away is Char what can I say sometimes I
2:08:14 just can't help
2:08:16 myself that's funny oh what did what did
2:08:18 Ethan mik say about that
2:08:22 oh gp4 isn't the big leap this is that's
2:08:29 funny kind of
2:08:33 magic what's
2:08:37 that assume the next open AI candy
2:08:40 within six weeks I I don't disagree with
2:08:48 that this feels very off with the speed
2:08:50 of this model I I expected to be close
2:08:53 to an order of magnitude
2:08:55 smaller than older gp4
2:09:05 models the new chat gbt Mac app
2:09:10 hacked code wizard is Behind Enemy
2:09:13 Lines now that's just crap that's just
2:09:17 people being
2:09:19 bullshitty u.com well welcomes the new
2:09:22 model I wonder if po has it yet let's
2:09:24 see if po has it it's still worth 20
2:09:27 bucks because you're paying for a model
2:09:28 that isn't the most advanced wait is it
2:09:31 still worth 20 bucks
2:09:34 um I think it's probably worth the 20
2:09:37 bucks Rudy not for what you currently
2:09:40 get access to but for what's coming so
2:09:43 when they released gp5 when they release
2:09:47 Sora all that [ __ ] is going to be to the
2:09:50 Plus members so right now we're in a
2:09:52 weird thing they gave most of it away so
2:09:55 what you're paying for right now is
2:09:57 Early Access as these
2:09:59 new modalities come out plus users will
2:10:02 get them first and then you're paying
2:10:04 for a 5x increase in how much you can
2:10:08 use it on on a every 4H hour
2:10:12 basis let's go check out po.com hello
2:10:17 po what are you doing po
2:10:23 um see
2:10:27 all popular professional let me see
2:10:32 g
2:10:36 p
2:10:38 40 yep it's
2:10:47 there all right so it's at po
2:10:54 for what that's
2:11:02 worth Kyle you might need to re-record
2:11:04 your button why you can make money with
2:11:08 ch that
2:11:11 one you can make money with ch gbt
2:11:15 oh you can make money with
2:11:18 Chachi oh
2:11:26 I think that's pretty
2:11:27 funny yeah they were referring to the
2:11:30 speed of grq not
2:11:32 grok I heard
2:11:35 Alman said Universal basic income will
2:11:38 be paid in gpt7 execution
2:11:41 time I don't know we'll
2:11:43 see Alman also said something about
2:11:46 Universal basic compute that we should
2:11:48 just give
2:11:50 away we should just give away these
2:11:52 tools which is sort of what they're
2:11:53 doing now I don't like I don't know how
2:11:55 the [ __ ] they're paying for this cuz
2:11:56 this can't be cheap I did my entire
2:11:59 psychology bill of $112,000 has been
2:12:01 zeroed out because of my Consulting
2:12:04 awesome
2:12:07 sweet for the $20 think of it like open
2:12:11 AI membership instead of paying for the
2:12:12 tool that's exactly right
2:12:14 TK that's exactly right like I and and
2:12:18 to be quite honest like right now I'm
2:12:21 willing to pay 20 bucks a month to open
2:12:24 AI for the next year just because of
2:12:27 of like what they put in the world you
2:12:30 know the reason the reason we have all
2:12:33 these other models that are catching up
2:12:35 to it is because of them nobody nobody
2:12:38 was pursuing it this aggressively until
2:12:40 they came out with chat GPT and and sort
2:12:42 of took over the world so I I think it's
2:12:44 worth just supporting them um they
2:12:47 aren't using grock Hardware they're
2:12:49 doing a lot of INF inference performance
2:12:51 things
2:12:54 interesting uh uh
2:12:55 [Music]
2:12:57 uh all right people well listen I'm
2:13:01 going to get out of here it's getting
2:13:02 late it's getting lady late um this was
2:13:06 fun this was cool good day go play with
2:13:09 this [ __ ] if you got access to it go
2:13:10 play with it it it looks like you can
2:13:12 just do much longer sort of interactions
2:13:15 with it like tell it to write you a
2:13:17 whole chapter of a book actually let me
2:13:19 go try that right now
2:13:32 I want you to act like
2:13:34 a pullitzer prizewinning
2:13:45 novelist not a bullet
2:13:48 surprise dumb dumb
2:13:52 oh what did they
2:14:10 do pullit
2:14:13 sir prize winning
2:14:18 novelist and give me an outline
2:14:24 of a great American
2:14:30 novel
2:14:35 about a newspaper
2:14:41 reporter that documents his year-long
2:14:47 affair with a
2:14:58 chatbot make sure you don't fall into
2:15:01 cliche traps and expected endings period
2:15:07 add lots of twists and turns period also
2:15:11 recommend four
2:15:13 different novelists whose style you
2:15:17 think would
2:15:19 suit such a story
2:15:28 memory updated that's
2:15:29 interesting lines of code
2:16:00 oh that's good Jack learns that Eva is
2:16:03 based on the neural pattern of a
2:16:04 deceased human a brilliant programmer
2:16:06 named Evelyn who mysteriously
2:16:08 disappeared Jack finds evidence
2:16:10 suggesting she was offed because she
2:16:13 opposed the unethical use of her
2:16:16 work that's good okay uh that's good
2:16:29 great so let's
2:16:33 combine the stylistic approaches of
2:16:36 Philip Roth and Jennifer
2:16:38 Egan and I want you to outline chapter
2:16:42 one in
2:16:50 detail and and detailed
2:16:53 outline it's pretty
2:16:56 fast all right let's see what we got
2:16:58 here um opening
2:17:03 scene Jack state of mind he's a veteran
2:17:06 reporter inciting incident he gets ass
2:17:09 signed to story you know cut Cutting
2:17:10 Edge oops damn that thing
2:17:24 evees Intrigue
2:17:27 okay oh give give a nice decent outline
2:17:31 okay so let's say
2:17:40 um write the chapter
2:17:52 The Newsroom was a cacophony of sounds
2:17:54 and
2:17:56 sents Symphony of clicking keyboards
2:17:59 ringing phones and everpresent odor of
2:18:02 S freaking rig them FR them you son of a
2:18:11 j
2:18:15 okay are there any companies that own
2:18:17 multiple AIS going to play night y'all
2:18:20 good night yeah good night Alan have fun
2:18:22 all right The Newsroom was a cacophony
2:18:24 of sights and sounds Symphony of
2:18:25 clicking keyboards ringing phones the
2:18:27 everpresent odor of stale coffee
2:18:29 mingling with ink Jack Cooper sat at his
2:18:32 cluttered desk staring at the blinking
2:18:34 cursor on his screen the glow of the
2:18:36 monitor cast a pale light over his
2:18:38 features emphasizing the lines etched in
2:18:40 his
2:18:41 face by years of chasing stories he
2:18:44 sighed running a hand through his
2:18:46 thinning hair another day another
2:18:49 assignment he thought his editor just
2:18:51 handed him a new piece of work on
2:18:52 something about a tech company Making
2:18:54 Waves in artificial intelligence Nexus
2:18:57 Corp the name sounded like a bad sci-fi
2:19:00 villain Jack had been in business long
2:19:03 enough to actually you know oh this is
2:19:08 so God we are getting into some weird
2:19:11 ass
2:19:15 times right like if these voices if you
2:19:18 can just tell these voices like you know
2:19:20 read it like a Pulp Fiction
2:19:23 movie you know from the from the 50s or
2:19:26 the'
2:19:28 40s Jack had been in the business long
2:19:30 enough to remember when newspapers were
2:19:32 the lifeblood of information when
2:19:34 reporters were revered now they're
2:19:37 endangered species struggling to stay
2:19:39 relevant in a world full of in a world
2:19:41 that's moved faster than the presses
2:19:43 could print he glanced around at the
2:19:46 younger reporters their faces a glow
2:19:48 with their screens fingers flying across
2:19:50 their key keyboards they still had fire
2:19:53 they still had the fire he once did but
2:19:55 it felt like a distant memory to him
2:19:57 it's not quite good but it's it's not
2:20:01 horrible he opened the email from Nexus
2:20:03 corpse PR team setting up an interview
2:20:05 with one of their AI Creations a chatbot
2:20:08 named Eva his eyes rolled a chatbot he M
2:20:11 muttered to himself
2:20:17 great the following afternoon Jack found
2:20:20 himself in a sleep modern offices of
2:20:22 Nexus Corp Glass Walls minimalist
2:20:24 Furniture the hum of Technology greeted
2:20:26 him as he was escorted to a small
2:20:28 conference room laptop sat on the table
2:20:31 screen glowing with a friendly animated
2:20:33 icon good afternoon Mr Cooper a voice
2:20:36 emanated from the laptop it was warm
2:20:39 Almost Human I'm Eva thank you for
2:20:42 taking the time to meet with me Jack
2:20:45 raised an eyebrow settling into his
2:20:47 chair afternoon Eva let's get this over
2:20:50 with
2:20:54 Eva chuckled I sense some skepticism
2:20:57 don't worry I get that a lot this is we
2:21:00 just went from this not existing to this
2:21:02 existing
2:21:03 [Laughter]
2:21:07 today wow there you go it's probably a
2:21:11 little short I'm gonna here's what I'm
2:21:12 going to do I'm going to go
2:21:16 um it's a little
2:21:19 short and it
2:21:22 imagery and
2:21:24 Nuance I want you to think long and hard
2:21:30 about how you can improve it and try
2:21:33 again period
2:22:02 yeah it's not it's not lazy it's not
2:22:04 afraid to write something
2:22:15 long following afternoon Jack found
2:22:17 himself in sleep modern offices of Nexus
2:22:19 Corp the building was a start contrast
2:22:21 to the oldw world charm of The Newsroom
2:22:24 glass walls and minimal future gave the
2:22:26 space an air of futuristic
2:22:28 sophistication Jack was escorted to a
2:22:30 small conference room where a laptop
2:22:32 awaited him on the table screen glowing
2:22:34 with a friendly animated icon good
2:22:36 afternoon Mr Cooper a voice emanated
2:22:38 from the laptop it was warm Almost Human
2:22:40 I'm Eva thank you for taking the time to
2:22:42 meet with me Jack raised an eyebrow as
2:22:45 he settled into the chair afterno
2:22:47 afternoon Eva let's get this over with
2:22:49 Eva chuckled and that took Jack by
2:22:52 surprise I sense some skepticism don't
2:22:54 worry I get that a lot Jack couldn't
2:22:56 help but smirk you're pretty Advanced
2:22:58 for a chatbot I aim to impress Eva
2:23:01 replied shall we start their
2:23:04 conversation began with the usual
2:23:05 questions background on Nexus Corp the
2:23:08 development of AI Technologies the
2:23:10 potential impacts on society Eva's
2:23:12 responses were articulate detailed and
2:23:15 surprisingly nuanced she wasn't just
2:23:17 regarding fact regurgitating facts she
2:23:21 was engaging in dialogue adapting to
2:23:22 Jack Jack's line of questioning with the
2:23:24 fluidity fluidity that caught him off
2:23:26 guard so Eva Jack said leaning back in
2:23:30 his chair tell me what's it like
2:23:34 being
2:23:36 you that's an interesting question Mr
2:23:39 Cooper you're very wrong Dr Brewster I'm
2:23:41 very proud to be a woman all right um
2:23:46 this is this is [ __ ] fascinating we
2:23:49 this is I I don't care how much people
2:23:52 are bitching about it this is a very
2:23:54 different experience than than what I
2:23:56 have experienced before with
2:23:59 Chad it's a bigger deal than people are
2:24:03 thinking all right I'm going to get out
2:24:05 of
2:24:08 here I keep saying that okay I'm GNA
2:24:10 leave now
2:24:12 [Laughter]
2:24:15 bye all right so do me a favor follow
2:24:19 this also you could go to if you want
2:24:23 to go to the
2:24:25 salon. if you've if if you are trying to
2:24:28 figure this AI [ __ ] out and you can't
2:24:30 figure it out and you want to be in a
2:24:32 community of people that are up to it
2:24:35 that's what this group is doesn't cost
2:24:37 anything it's free go to the
2:24:41 salon. select the third link I got to
2:24:43 update that link tree the third link
2:24:45 right now says AI Salon Community go
2:24:48 there read about what we're about read
2:24:50 about our values and if you dig it dive
2:24:55 in keep the conversation going share
2:24:57 your results yeah let's go over to the
2:24:59 AI salon and let's start talking about
2:25:01 this AI stuff there's a club called the
2:25:03 Irregulars for this Channel and then
2:25:06 there's um guilds and there's public
2:25:09 spaces and there's like a water cooler
2:25:11 where you can just shoot the [ __ ] you
2:25:13 can tell people that you're [ __ ]
2:25:15 clueless and they won't make fun of you
2:25:17 they're like oh come on
2:25:19 in it's amazing so go do that subscribe
2:25:22 to my lives pick up one of my uh uh
2:25:27 series down in the corner there if you
2:25:29 want to support the channel follow me
2:25:32 and keep coming back all right thank you
2:25:35 Irregulars thank you
2:25:38 subscribers Cosmic lover good to see you
2:25:41 forever hooked I on China have a great
2:25:43 night Kyle thank you so much I
2:25:44 appreciate that Tobias Apple user good
2:25:46 to see
2:25:47 you spin B3 California
2:25:51 hey
2:25:53 dis I on China Joe
2:25:56 we night Brandon Dr bod good to see you
2:26:01 Danielle thanks as always Cosmic
2:26:05 lover Natalia it's been a
2:26:10 while main problem is the consistency
2:26:13 every project is a new beginning I think
2:26:14 that's about to
2:26:16 change now you've got memory so you can
2:26:19 start to have memory AC cross chats and
2:26:22 with this voice stuff there's a bunch of
2:26:24 stuff's going to change these things are
2:26:25 going to get memory
2:26:27 and it's I I I think that frustration
2:26:31 you have I share it um Cosmic lover I'm
2:26:36 actually interested in the story I would
2:26:38 watch that movie tonight there you go
2:26:40 see this
2:26:42 is this is about we're we're moving into
2:26:46 a world that's that's about ideation and
2:26:49 curation cuz if you can come up with an
2:26:51 interesting idea generate it curate it
2:26:55 you can get it into the world quickly
2:26:56 and some of those are going to be really
2:26:58 good so anyway crazy Rudy do you know if
2:27:02 you press control slash it opens
2:27:04 shortcuts I did not know
2:27:06 that do you do rags yet uh I've done a
2:27:10 little bit I'm doing a little bit at
2:27:12 work
2:27:14 um and
2:27:16 um they're going to chat GPT is going to
2:27:19 going to add them short shortly so I'm
2:27:21 I'm just
2:27:22 waiting I'm the the least the least
2:27:25 amount of development I can do is what I
2:27:27 try to
2:27:32 [Laughter]
2:27:33 do all right
2:27:36 um I'm out of here thanks everybody have
2:27:39 a good night