OpenAI DEV DAY! It was seriously like Christmas morning. So many presents under the GPT4 tree!

Description

Chapters

Transcript

0:00 I need to set up my
0:11 screenshotting let me see I should save
0:14 that to desktop it is okay so now I can
0:17 screenshot if there's something good to
0:20 screenshot for for later big news uh I
0:23 don't know Source Camp we'll see this is
0:25 their this is the keynote for the
0:27 developer conference today Sam Alman
0:29 last night posted that he said it felt
0:32 like Christmas Eve which says to me
0:34 they're going to announce some big stuff
0:37 uh that he's excited
0:38 about um and then and then the leaks the
0:43 leaks that happened on Saturday Friday
0:46 or Saturday I think Friday um
0:51 were a bot maker things like that so
0:55 going to going to be interesting stuff
0:56 hey UT what's
0:58 happening de de
1:01 I'm not sure my brain can handle anymore
1:04 I'm right there with you I'm like I just
1:07 decided to do this I've never really
1:08 done like a live stream commentary
1:10 before so this could could be an
1:12 absolute [ __ ] show but you know welcome
1:15 welcome to AI learning lab happy open AI
1:19 day making Bots is AI news yep it is hey
1:23 bud hey Jim Taylor what's
1:26 happening good to see you good to see
1:29 you human
1:32 nice
1:34 name Raven
1:38 toown and as soon as they go live I'll
1:40 pop up let's see let's see their choice
1:42 of Music let's see their choice of hold
1:45 [Music]
1:48 music that sounds like it was ai ai
1:53 composed that's looks like sounds like
1:55 something you do on sound
1:58 raw good morning good morning Ricky
2:01 Ticky Fabby hair growth geek what's
2:03 shaking what's
2:06 shaking I'm all
2:08 excited oh wait is this a Daydream
2:11 you're on once
2:14 again
2:16 Ella Ella
2:18 elun Funk Ella the elephunk yes
2:23 yes yeah I I it does feel different
2:27 doing this in the daytime I feel like I
2:29 I can't quite be myself but that's all
2:32 right what are you going to do what are
2:35 you going to
2:37 do I have no body parts shaking and
2:39 that's a good
2:41 thing here for the [ __ ] show or as DOI
2:44 might say s s so
2:49 show
2:52 exactly what's happening Cindy good to
2:54 see you I'm excited for uh for what
2:58 you're going to present at the salon
2:59 tomorrow so Cindy [ __ ] who's a futurist
3:01 is speaking at tomorrow night's AI salon
3:05 so if you
3:07 um if you're not a part of the AI Salon
3:09 go to the
3:11 salon. and um sign up for the Meetup you
3:16 should also join the Discord even though
3:18 we're going to move off that soon you
3:19 should still join it and become part of
3:21 the
3:22 community but oh that music's driving me
3:26 crazy there's party favors yay
3:31 all right welcome and do me a favor if
3:34 you would share this live so we can get
3:36 as many people in here because I don't
3:37 normally go live this time of day and I
3:40 also have my countdown timer says this
3:42 isn't starting you this is the little
3:44 you're in the Green Room hanging out
3:45 before the thing actually
3:47 starts uh encourage Be not Afraid
3:49 encourage more viewers to share that
3:51 would be awesome
3:52 yes Genie and a bottle good to see
3:56 you like the
3:59 live
4:01 wait can I invite how do
4:05 I share is this
4:09 ituno
4:11 Lazelle yeah this is
4:16 good oh source Camp's already in
4:29 here
4:35 all right I just shared the live too to
4:38 some some folks AI learning lab shared
4:42 with more than 10 friends yes I
4:47 did there's 12,200 people watching the
4:51 uh the Youtube
4:54 stream and we have 30 here so it's close
4:58 we're close we're close to AI level or
5:01 open AI level
5:10 numbers request to
5:12 join and by the way this is if you go to
5:15 if you want to see this live on your own
5:17 computer this is at uh open
5:20 ai's YouTube channel and then click on
5:22 their live Tab and you'll see this so
5:25 you can listen to it there locally CU
5:27 I'll probably mute occasionally to talk
5:30 about [ __ ] is my I don't really know
5:32 what I'm doing here like well that's
5:36 that's the that's the whole point of the
5:38 channel isn't
5:40 it I'm leaning into not knowing what I'm
5:43 doing
5:45 [Laughter]
5:48 here oh man wait what happened with the
5:51 time
5:52 change I have the stream happening in a
5:55 day now no it should
5:58 be they start started streaming 25
6:01 minutes ago with this little animated
6:03 opening so this was a non-live stream
6:06 and and now it's a live stream
6:09 so this is good this is unlike the haen
6:13 uh thing that I went to
6:15 where they didn't get this going
6:18 until 5 minutes
6:23 after I'm listening while at work don't
6:25 tell my boss would never I would never
6:27 tell your boss Forever Young this is
6:29 listen this is work this is we're
6:31 finding out what's happening with uh
6:34 with generative AI stuff that's going to
6:35 completely transform your company so I
6:38 think this is an appropriate use of your
6:39 time I will support you on
6:43 that you like say to your boss go talk
6:45 to Kyle Shannon who's Kyle Shannon
6:47 that's not important right now but he'll
6:49 vouch for
6:50 me I'll do it too I don't give it I
6:54 don't give two
6:55 shits um let me make that how do I make
6:58 that straight
7:02 Kyle I
7:03 request to your
7:07 live not Dar just
7:11 resting oh dar dar dar dar dar the
7:16 internet Tick Tock
7:19 star
7:21 bam all right I will most likely I don't
7:25 know what I'm going to do it it depends
7:27 how I assume there's going to be a lot
7:30 of kind of Dead Space where they're you
7:32 know thanking people and things like
7:33 that so I'm going to try to just let it
7:36 play so we can all see it
7:38 together check my request oh your
7:52 request um do we have time what time is
7:55 it we have 2 minutes um we can do it can
7:59 you can you do it quick
8:01 Sherry here you
8:03 go hello
8:06 hello does this work what's happening oh
8:10 that's cool okay it just works I just
8:11 wanted to test if it works so next time
8:14 we could do live with a bunch of people
8:16 bye okay cool awesome okay bye awesome
8:19 bye see
8:21 you how do I get rid of it now oh my
8:24 goodness I don't know there you
8:28 go all right there's that oops oh man
8:33 what the hell is going on with the hair
8:34 time for a
8:36 [Laughter]
8:38 haircut I don't mind it at night but
8:41 live good Lord make a Kyle Shannon
8:43 certificate of attendance for Future
8:46 boss oh that's not bad Rudy wow Mr
8:49 Shannon you go live now yeah so this is
8:50 we're we're here to uh I'm going to live
8:54 stream the open AI Dev keynote so we're
8:56 going to be live for an hour and then I
8:58 have a meeting right after after this so
9:01 I'll be shutting down right at 11: my
9:03 time it's or 12 my time it's just about
9:06 11: my time
9:07 so it is just
9:11 about 11 Mountain time and I'll go for
9:16 an
9:17 hour is this the 24hour live no no
9:20 no I still haven't quite decided if I'm
9:23 G to do a 24-hour live but I'm going to
9:25 do something I'm going to do something
9:28 so we'll figure it out damn okay well I
9:31 I got excited when I saw you yeah no
9:33 this is definitely a non-standard time
9:35 when I'm going live I also feel like I
9:37 can't swear because people are watching
9:39 this at work although for those of you
9:41 on the other side of the world you're
9:42 probably watching this you know at night
9:44 when the when when the East Coasters do
9:47 on on the normal lives
9:50 [Laughter]
9:53 so um it's good to know Sherry can do
9:55 that for November 30th oh yeah yeah yeah
9:57 exactly that's awesome terrific
10:01 all right
10:05 so all right let me to pull something
10:08 over here cuz I am at work oh this is
10:11 getting
10:13 louder there we go good morning thank
10:15 you for joining us today please welcome
10:18 to the stage Sam
10:20 [Music]
10:28 mman good morning Welcome to our first
10:32 ever open AI Dev day we're thrilled that
10:34 you're here and this energy is
10:36 [Applause]
10:40 awesome and welcome to San Francisco San
10:43 Francisco has been our home since day
10:45 one the city is important to us and to
10:48 the tech industry in general we're
10:49 looking forward to continuing to grow
10:51 here so we've got some great stuff to
10:54 announce today but first I'd like to
10:57 take a minute to talk about some of the
10:58 stuff that we've done over the past year
11:01 about a year ago November 30th we
11:04 shipped chat GPT as a low-key research
11:07 preview and that went pretty
11:10 well in March we followed that up with
11:13 the launch of GPT 4 still the most
11:15 capable model out in the world
11:19 yep that's not
11:20 [Applause]
11:23 [ __ ] and in the last few months we
11:25 launched voice and vision capabilities
11:28 so that chat GPT can now see hear and
11:31 speak and more Rec there's a lot you
11:35 don't have to clap each
11:36 time and more recently we launched Dolly
11:38 3 the world's most advanced image model
11:41 you can use it of course inside of chat
11:43 GPT for our Enterprise customers we
11:46 launched chat GPT Enterprise which
11:48 offers Enterprise grade security and
11:50 privacy higher speed gp4 access longer
11:53 context Windows a lot
11:55 more today we've got about 2 million
11:59 Developers
12:00 wow building on our API for a wide
12:01 variety of use cases doing amazing stuff
12:04 over 92% of Fortune 500 companies
12:07 building on our
12:08 products and we have about a 100 million
12:11 weekly active users now on
12:14 chbt Wow 100 million weekly incredible
12:19 on that is we got there entirely through
12:21 Word of Mouth people just find it useful
12:23 and tell their friends yeah open AI is
12:25 the most advanced and the most widely
12:28 used AI platform in the world
12:30 now but numbers never tell the whole
12:33 picture on something like this what's
12:35 really important is how people use the
12:37 products how people are using Ai and so
12:39 I'd like to show you a quick
12:41 video I actually wanted to write
12:44 something to my dad in talo I want a
12:48 non-romantic way to tell my parent that
12:51 I love him and I also want to tell him
12:54 that he can rely on me but in a way that
12:57 still has the respect
12:59 of like a child to parent relationship
13:02 that you should have in toino culture
13:03 and in G when it's translated into G I
13:06 love you very deeply and I will be with
13:09 you no matter where the path leads I see
13:11 somebody possibility I was like who
13:13 something I'm not sure
13:16 oops hey this I'm thinking about so kind
13:19 give that more confidence the first
13:21 thing that blew my mind was it levels
13:23 with you like that's something that a
13:25 lot of people struggle to do it opened
13:28 my mind to just what every creative
13:32 could do if they just had a person
13:35 helping them out who listens so this is
13:38 a to represent sickling hemoglobin and
13:40 you built that with chat GPT chat GPT
13:43 built it with me using it for daily
13:46 activities like hey here's a picture of
13:47 my fridge can you tell me what I'm
13:49 missing because I'm going grocery
13:50 shopping and I really need to do recipes
13:52 that are following My Vegan Diet as soon
13:55 as uh we got access to code interpreter
13:57 I was like wow this thing is awesome
13:59 they could go spreadsheets it could do
14:01 anything I discovered chatty about three
14:04 months
14:05 ago 100 birthday wow is very friendly
14:11 very patient very knowledgeable and very
14:16 it's been a wonderful thing I'm a 4.0
14:19 student but I also have four children
14:21 while I started using chat T I realized
14:24 I could ask chat GPT that question and
14:27 not only does it gives me an explanation
14:31 didn't need booing as much it gave me a
14:34 life back it gave me time for my family
14:37 and time for me I have chronic nerve
14:39 pain uh on my whole left half of my body
14:41 of nerve damage I had like a spine a
14:44 brain surgery and so I have like limited
14:46 use of my left hand now you can just
14:48 have like the integration of voice input
14:51 and the newest one where you can have
14:53 the back and forth dialogue that's just
14:55 like accessibility maximum huge best
14:58 interface for me it's
15:01 [Music]
15:06 [Applause]
15:09 here so we love hearing the stories of
15:12 of how people are using the technology
15:14 it's really why we do all of this okay
15:17 so now on to the new stuff and we have
15:19 got a lot oh yeah here we go first we're
15:24 going to talk about a bunch of
15:25 improvements we've made and then we'll
15:27 talk about where we're headed next
15:30 over the last year we spent a lot of
15:32 time talking to developers around the
15:34 world we've heard a lot of your feedback
15:37 it's really informed what we have to
15:38 show you
15:39 today today we are launching a new
15:43 model gp4 turbo
15:46 [Applause]
15:50 oh gp4 turbo will address many of the
15:53 things that you all have asked for so
15:56 let's go through what's new we've got
15:58 six major things to talk about for this
16:00 part number one context link yay a lot
16:04 of people have tasks that require a much
16:07 longer context length gp4 supported up
16:10 to 8K and in some cases up to 32k
16:13 context length but we know that isn't
16:15 enough for many of you and what you want
16:17 to do gp4 turbo supports up to 128,000
16:21 tokens of contact
16:23 oh going after
16:27 Claude that's 30 pages of a standard
16:30 book 16 times longer than our 8K context
16:32 nice and in addition to longer context
16:35 length you'll notice that the model is
16:37 much more accurate over a long context
16:40 great number two more
16:44 control we've heard loud and clear that
16:46 developers need more control over the
16:48 model's responses and outputs so we've
16:51 addressed that in a number of
16:53 ways we have a new feature called Json
16:55 mode oh nice which ensures that the
16:57 model will respond with valid Json
17:00 this has been a huge developer request
17:02 it'll make calling apis much easier big
17:05 deal the model is also much more
17:07 accurate over a long conversation many
17:09 functions at once dunkaroo and it'll do
17:11 better at following instructions in
17:14 general we're also introducing a new
17:16 feature called reproducible outputs you
17:18 can pass a SE parameter and it'll make
17:20 the model return consistent outputs o
17:22 this of course gives you a higher degree
17:24 of control over Model Behavior this
17:26 rolls out in beta today wow
17:33 and in the coming weeks we'll roll out a
17:34 feature to let you view log probs in the
17:38 API now likely this is just going to be
17:40 an A in chat GPT playground it's not
17:42 going to necessarily be in chat GPT to
17:45 be able to access better knowledge
17:47 that's my
17:48 guess so we're launching retrieval in
17:50 the platform you can bring Knowledge
17:52 from outside documents or databases into
17:54 whatever you're building o that's huge
17:56 we're also updating the knowledge cuto
17:58 off we we are just as annoyed as all of
18:00 you probably more the gp4s knowledge
18:02 about the world ended in 2021 we will
18:04 try to never let it get that out of date
18:06 again gp4 turbo has knowledge about the
18:09 world up to April of
18:11 2023 and we will continue to improve
18:13 that over
18:15 time number four new
18:19 modalities surprising no one Dolly 3 gp4
18:24 Turbo with vision and the new text to
18:26 speech model are all going into the API
18:29 today oh
18:32 yes so if you're doing zapier
18:34 automations you can now call all these
18:36 things is what that means a handful of
18:37 customers that have just started using
18:39 Dolly 3 to program programmatically
18:41 generate images and Designs today Koke
18:45 is launching a campaign that lets his
18:46 customers generate devali cards using
18:48 Dolly 3 wow and of course our safety
18:51 systems help developers protect their
18:52 applications against misuse those tools
18:54 are available in the API wow gb4 turbo
18:58 can now accept images as inputs via the
19:00 API wow generate captions
19:03 classifications and Analysis for example
19:06 the eyes uses this technology to help
19:09 people who are blind or have low vision
19:11 with their daily tasks like identifying
19:13 products in front of them API stands for
19:15 application programming interface it's
19:17 how programmers can take data from their
19:20 app send it to open AI run it through
19:23 gp4 and get really cool stuff Backes to
19:26 choose from I'll play an example did you
19:29 know that Alexander Graham Bell the
19:31 eminent inventor was enchanted by the
19:33 world of sounds his ingenious mind led
19:36 to the creation of the graphophone which
19:38 etched sounds onto wax making voices
19:40 whisper through
19:42 time this is much more natural than
19:44 anything else we've heard out there
19:46 voice can make apps more natural to
19:48 interact with and more accessible it
19:51 also unlocks a lot of use cases like
19:52 language learning and voice
19:55 assistance speaking of new
19:57 modalities we're also releasing the next
20:00 version of our open- source speech
20:01 recognition model whisper V3 today it'll
20:04 be coming soon to the API Whisper 2 is
20:06 amazing features improve performance
20:07 across many languages and we think
20:09 you're really going to like
20:10 it okay number five
20:13 customization fine tuning has been
20:15 working really well for gbt 3.5 since we
20:18 launched it a few months ago starting
20:20 today we're going to expand that to the
20:22 16k version of the
20:24 model also starting today we're inviting
20:28 active fin tuning users to apply for the
20:30 gp4 fine tuning experimental Access
20:33 program the fine-tuning API is great for
20:36 adapting our models to achieve better
20:37 performance in a wide variety of
20:39 applications with a relatively small
20:41 amount of data but you may want to model
20:44 to learn a completely new knowledge
20:45 domain or to use a lot of proprietary
20:48 data so today we're launching a new
20:50 program called custom models oo with
20:53 custom models our researchers will work
20:55 closely with a company to help them make
20:57 a great custom model especially for them
21:01 and their use case using our
21:02 tools think anyone with proprietary data
21:06 they're going to be able to build custom
21:08 applications with their data using chat
21:12 gp4 processing whatever else we won't be
21:15 able to do this with many companies to
21:17 start it'll take a lot of work and in
21:19 the interest of expectations at least
21:20 initially it won't be cheap but if
21:22 you're excited to push things as far as
21:24 they can currently go please get in
21:26 touch with us and we think we can do
21:28 something pretty
21:29 great okay and then number six higher
21:32 rate limits we're doubling the tokens
21:34 per minute for all of our established
21:36 gp4 customers so that it's easier to do
21:38 more and you'll be able to request
21:40 changes to further rate limits and
21:42 quotas directly in your API account
21:44 settings in addition to these rate
21:46 limits it's important to do everything
21:49 we can do to make it new successful
21:51 building on our
21:53 platform so we're introducing copyright
21:55 Shield oh interesting copyright Shield
21:57 means that we will step step in and
21:59 defend our customers and pay the cost
22:01 incurred if you face legal claims around
22:03 copyright infringement and this applies
22:05 both to chat GPT Enterprise and the API
22:08 they're following Microsoft there that's
22:10 probably because their Enterprise
22:12 customers are saying they need that on
22:14 data from the API or chbt Enterprise
22:16 ever ever yeah so this is this is in
22:19 response to customers there's actually
22:21 one more developer request that's been
22:23 even bigger than all of these uh so I'd
22:26 like to talk about that now
22:29 and that's pricing oh this is
22:32 Big gp4 Turbo is the industry-leading
22:36 model it delivers a lot of improvements
22:39 that we just covered and it's a smarter
22:41 model than
22:43 gp4 we've heard from developers that
22:46 there are a lot of things that they want
22:47 to build but gp4 just cost too much
22:51 they've told us that if we can decrease
22:53 the cost by 20 25% that would be great
22:56 it's going to be 10x cheaper I bet
22:58 I'm super excited to announce that we
23:00 worked really hard on this and gbt 4
23:03 Turbo a better model is considerably
23:06 cheaper than
23:07 gb4 bu a of 3x for prompt tokens that's
23:12 [Applause]
23:14 [Music]
23:17 huge and two expert completion tokens
23:20 starting today great
23:24 awesome so the new pricing is 1 cent per
23:27 th prompt tokens and 3 cents per th000
23:30 completion tokens for most customers
23:32 that will lead to a blended rate more
23:34 than 2.75 times cheaper to use for gp4
23:37 Turbo than gp4 great we worked super
23:39 hard to make this happen we hope you're
23:41 as excited about it as we
23:47 are so we decided to prioritize price
23:50 first because we had to choose one or
23:52 the other but we're going to work on
23:53 speed next we know that speed is
23:55 important too soon you will notice GPT 4
23:58 Turbo becoming a lot
24:00 faster we're also decreasing the cost of
24:03 GPT 3.5 turbo
24:05 16k also input tokens are 3x less and
24:08 output tokens are 2x less which means
24:10 the GPT
24:12 3.56k is now cheaper than the previous
24:14 GPT 3.5 4K
24:18 model running a fine-tune GPT 3.5 turbo
24:21 16k version is also cheaper than the old
24:23 fine-tuned 4K
24:25 version okay so we just covered a lot
24:28 about the model itself we hope that
24:30 these changes address your feedback
24:32 we're really excited to bring all of
24:33 these improvements to everybody
24:36 now in all of this we're lucky to have a
24:39 part does this affect subscription rate
24:42 I doubt it so I'd like to bring out a
24:43 special guest s Nella the CEO of
24:46 Microsoft oh here we go this could be
24:48 good this is going to be connect
24:51 connections to Office 365 I bet thank
24:54 you
24:55 huge thanks so much for coming here it's
24:58 fantastic to be here and uh Sam congrats
25:01 I mean I'm really looking forward to
25:02 Turbo and everything else that you have
25:04 coming it's been just fantastic
25:06 partnering with you guys I love this guy
25:07 he's great two questions won't take too
25:09 much of your time how are how is
25:10 Microsoft thinking about the partnership
25:12 currently
25:15 first we love you
25:18 guys look it's that's gonna change dunar
25:21 fantastic for us in fact I remember the
25:23 first time I think you reached out and
25:24 said hey do you have some Azure credits
25:26 we've come a long way from there uh
25:28 thank you for
25:30 that you guys have built something
25:32 magical I mean quite frankly there are
25:33 two things uh for us when it comes to
25:35 the partnership the first is these
25:37 workloads and even when I was listening
25:39 backstage to how you're describing
25:41 what's coming even it's just so
25:43 different and new I've been in this
25:44 infrastructure business for you know
25:46 three decades no one has ever seen
25:47 infrastructure yeah and and the workload
25:50 the pattern of the workload these you
25:52 know these training jobs are so
25:53 synchronous and so large and so data
25:56 parallel um and so the first thing that
25:59 we have been doing is building in
26:00 partnership with you the system all the
26:02 way from thinking from power to the DC
26:05 to the rack to the accelerators to the
26:07 network uh and just you know really the
26:10 a the shape of azure is drastically
26:13 changed uh and is changing rapidly in
26:16 support of these models that you're
26:17 building and so our job number one is to
26:20 build the best system so that you can
26:22 build the best models and then make that
26:25 all available to developers and so the
26:27 other thing is we ourselves are a
26:28 developer so we're building products in
26:30 fact my own conviction of this entire
26:33 generation tired Runner I don't I don't
26:35 know that Microsoft will ever buy open
26:37 AI but I I I doubt it unless they get in
26:41 trouble so we want to build our co-pilot
26:44 get up co-pilot all as developers uh on
26:46 top of open Ai apis and so we are very
26:49 committed to that and what does that
26:51 mean to developers you look I Rudy Miss
26:54 I just spent my 20 bucks a developer
26:56 company and a partner company and so we
26:58 want to make you for example we want to
27:00 make GitHub available GitHub co-pilot
27:02 available as the Enterprise Edition
27:04 available to all the attendees here so
27:06 they can try it out that's awesome yeah
27:08 we're very excited about that that's
27:10 good when you going to give it to the
27:12 rest of us buddy and right now count on
27:14 us to build the best infrastructure in
27:17 Azure with your API support uh and bring
27:20 it to all of you and then even things
27:22 like the Azure Marketplace so for
27:24 developers who are building products out
27:25 here to get to Market uh rapidly so
27:28 that's sort of really our intent here
27:30 great and how do you think about the
27:31 future future of the partnership or
27:33 future of AI or whatever yeah
27:36 there that's you know like there are
27:40 couple of things for me that I think are
27:42 going to be very very key for us right
27:43 one is I just described how the systems
27:48 that are needed as you aggressively push
27:51 forward on your road map um requires us
27:54 to be on the top of our game and we
27:56 intend fully to commit ourselves deeply
27:59 to making sure you all as Builders of
28:02 these Foundation models have not only
28:04 the best systems for training and
28:07 inference but the most compute so that
28:09 do me Savor favor and please share the
28:11 live if you can so we can get some more
28:13 people in here way we want to make
28:15 progress the second thing I think both
28:17 of us care about in fact quite frankly
28:18 the thing that excited both sides to
28:21 come together is your mission our
28:23 mission our mission is to empower every
28:24 person and every organization on the
28:26 planet to achieve
28:29 AI is only going to be useful if it
28:31 truly does empower right I mean I saw
28:33 the video you played early I mean that
28:34 was fantastic to see thank you thanks
28:37 for sharing the live describe what AI
28:40 meant for them and what they were able
28:41 to achieve so ultimately it's about
28:43 being able to get the benefits of AI
28:45 broadly decimated to everyone I think is
28:48 going to be the goal for us and then the
28:49 last thing is of course we're very
28:51 grounded in the fact that safety matters
28:52 and safety is not something that you
28:54 care about later but it's something we
28:55 do shift left on and we very very
28:58 focused on that with you all great well
29:00 I think we have the best partnership in
29:01 Tech I'm excited for us to build AGI
29:03 together no I'm really excited have a f
29:04 thank you very much for coming
29:07 nice just reiterating it's an important
29:10 partnership they're not going to let
29:12 them
29:14 flail okay make it safe for corporations
29:17 a lot of great updates for developers
29:19 already and we got a lot more to come
29:21 but even though this is a developer
29:23 conference we can't resist making some
29:25 improvements to Chachi PT here we go
29:28 so a small one chat GPT now uses GPT 4
29:31 Turbo with all the latest improvements
29:33 including the latest knowledge cut off
29:35 which will continue update that means we
29:36 get 128,000 tokens it can out browse the
29:39 web when it needs to write and run code
29:41 analyze data take and generate images
29:44 and much more and we heard your feedback
29:45 that model picker extremely annoying
29:47 that is gone starting today might have
29:50 to click around the drop down menu
29:51 starting today all of this will just
29:52 work together chat GPT
29:56 yeah
29:59 I'm going to look now chat gbt will just
30:01 know what to use and when you need
30:03 it but that's not the main
30:06 thing uh and neither was price I still
30:08 have the old version but we'll see
30:10 developer request there was one that was
30:11 even bigger than that and I want to talk
30:14 about where we're headed and the main
30:16 thing we're here to talk about today so
30:19 we believe that if you give people
30:21 better tools they will do amazing things
30:23 we know that people want AI that is
30:24 smarter more personal more customizable
30:27 can do more on your behalf eventually
30:30 you'll just ask the computer for what
30:32 you need and it'll do all of these tasks
30:34 for you these capabilities are often
30:37 talked in the AI field about as agents
30:40 oh here we go here's the bots upsides of
30:42 this are going to be tremendous at open
30:45 AI we really believe that gradual
30:47 iterative deployment is the best way to
30:50 address the safety issues the safety
30:51 challenges with AI we think it's
30:54 especially important to move carefully
30:55 towards this future of Agents it's going
30:58 to require a lot of technical work and a
31:00 lot of thoughtful consideration by
31:02 Society So today we're taking our first
31:06 small step that moves us towards this
31:08 future we're thrilled to uh we're
31:10 thrilled to
31:11 introduce
31:13 gpts G gpts are tailored versions of
31:16 chat GPT for a specific purpose these
31:19 are Bots these are Bots you can build a
31:20 GP just like customized version of chat
31:23 GPT for almost anything yeah
31:26 instructions expanded knowledge and
31:28 actions and then you can publish it for
31:30 others to use calling gpts is genius com
31:33 instructions expanded knowledge and
31:35 actions they can be more helpful to you
31:38 they can work better in any context and
31:40 they can give you better
31:41 control they'll make it easier for you
31:43 to accomplish all sorts of tasks or just
31:45 have more fun Jim Ross it's like
31:47 Christmas it is chat
31:49 GT you can in effect program a GPT with
31:53 language former president Obama from
31:56 Becky R is in charge of
31:59 regulations very access and it gives
32:01 agency to
32:03 everyone so we're going to show you what
32:05 gpts are how to use them how to build
32:08 them and then we're going to talk about
32:10 how they'll be distributed and
32:11 discovered Jennifer L happy birthday
32:14 developers we're going to show you how
32:15 to build these agent like experiences
32:17 into your own apps so first let's look
32:20 at a few
32:21 examples our partners at code.org are
32:24 working hard to expand computer science
32:27 in schools lesson planner they've got a
32:29 curriculum millions of students
32:31 worldwide this is huge that you can
32:33 publish these plan GPT to help teachers
32:36 provide a more engaging experience for
32:38 middle
32:39 schoolers if a teacher asks it to
32:41 explain for Loops in a creative way it
32:43 does just that in this case it'll do it
32:46 in terms of a video game character
32:48 repeatedly picking up coins super easy
32:50 to understand for an eighth
32:51 grader as you can see this GPT brings
32:55 together code.org extensive curriculum
32:57 and expert
32:58 and lets teachers adapt it to their
32:59 needs quickly and
33:01 easily next Cana has built a GPT that
33:06 lets you start designing by describing
33:08 what you want in natural language if you
33:10 say make for a Dev date reception this
33:13 afternoon this evening just in a school
33:15 board meeting thinking about chat gbt
33:17 it'll generate a few options to start
33:18 with by hitting canvas apis now this
33:21 concept may be familiar to some of you
33:23 we've evolved our plugins to be custom
33:25 actions for
33:26 gpts you can keep chatting with this to
33:29 see different iterations when you see
33:31 one you like you can click through to
33:33 canva for the full design experience wow
33:36 so now we'd like to show you a GPT live
33:39 oh that's so cool here we go here has
33:41 built a GPT that that lets you perform
33:44 actions across 6,000 applications to
33:46 unlock all kinds of integration possib
33:48 my God I'd like to introduce Jessica one
33:50 of our Solutions no sleep for us going
33:52 to drive this demo this is done this is
33:55 we this is
33:59 thank you all holy
34:02 crap thank you all for being here my
34:04 name is Jessica Shay I work with
34:06 partners and customers to bring their
34:07 product alive and today I I can't wait
34:10 to show you how hard we've been working
34:11 on this so let's get started so to start
34:15 where your gbt will live is on this
34:17 upper left corner I'm going to start
34:19 with clicking on the zappier your AI
34:22 actions and on the right hand side you
34:24 can see that's my calendar for today so
34:26 it's quite a day I've going to use this
34:29 before so it's actually already
34:31 connected to my calendar to start I can
34:34 ask what's on my schedule for today we
34:37 build gpts with security in mind so
34:40 before it performs any action or share
34:42 data it will ask for your permission so
34:45 right here I'm going to say allowed so
34:49 gbt is designed to take in your
34:52 instructions make the decision on which
34:54 capability to call to perform that
34:56 action and then execute that for you so
34:58 you can see right here it's already
35:00 connected to my calendar it pulls into
35:03 my my
35:04 information I've also prompted it to
35:07 identify conflicts on my calendar so you
35:09 can see right here actually was able to
35:12 identify that so it looks like I have
35:15 something coming up so what if I want to
35:17 let Sam know that I have to leave early
35:19 so right here I say Let Sam no I got to
35:23 go um chasing
35:26 gpus
35:27 so with that I'm going to swap to my
35:32 conversation with Sam and then I'm going
35:34 to say yes please run
35:39 that Sam did you get that I
35:43 did
35:46 awesome
35:48 so this is only a glimpse of what is
35:51 possible and I cannot wait to see Papa J
35:54 yes it looks like Z zappier and cber
35:57 integrated but it sounds like we're
35:58 going to be able to
35:59 connect these Bots to other services so
36:03 we'll be able to use a consumer version
36:05 of an API basically in addition to these
36:08 there are many more kinds of gpts that
36:10 people are creating and many many more
36:12 that will be [ __ ] genius that they're
36:14 calling them gpts they're making it like
36:16 Xerox a term that everyone will just use
36:19 always associate it with them brilliant
36:20 you can program the GPT just by having a
36:24 conversation we believe that natural
36:26 language is going to be a big part of
36:27 how people use computers in the future
36:29 and we think this is an interesting
36:30 early example so I'd like to show you
36:33 how to build
36:37 one he's gonna build one right now so I
36:39 want to create a GPT uh that helps give
36:42 Founders and developers advice when
36:43 starting new
36:44 projects
36:46 um I'm going to go to create a GPT here
36:49 and this drops me into the GPT
36:52 builder getting emotional this is
36:55 exciting devel
36:57 always about how do I you know think
36:59 about a business idea can you give me
37:00 some advice uh I'm going to see if I can
37:02 build a GPT to help with that so to
37:05 start GPT Builder asks me what I want to
37:08 make and I'm GNA say I want to help GPT
37:11 Builder Founders this is it's like do
37:13 it's this is app development like do you
37:15 just ask it for what you want it's GNA
37:17 code it this is staggering after the
37:21 founder has gotten can neural link
37:23 connect with gbt probably not that far
37:26 off crypto Joe
37:27 them on why they are not growing
37:33 faster all right so to start off I just
37:35 tell the GPT a little bit about about
37:37 what I want here and it's going to go
37:39 off and start thinking about that and
37:41 it's going to write some detailed
37:42 instructions for the GPT um it's also
37:45 gonna let's see ask me about a name how
37:47 do I feel about startup Mentor that's
37:49 fine that's good so if I didn't like the
37:53 name of course I could call it's even
37:54 naming the the thing for you brilliant
37:57 and start there and this is just like
38:01 the right in preview mode that it's
38:03 starting to fill out the GPT um where it
38:06 says what it does it has some like ideas
38:09 of additional questions that I could ask
38:11 um and made an icon using DOL so it just
38:15 generated a candidate of course I could
38:17 regenerate that or change it but I sort
38:19 of like that so I will say that's
38:24 great and you see now that the GP is
38:27 being built out a little bit more as we
38:29 go this is unbelievable what I want this
38:31 to do um how it can interact with users
38:33 I could talk about style here but what
38:35 I'm going to say uh is I am going to
38:40 upload transcripts of some lectures
38:44 about startups I have given oh man give
38:48 advice based off of okay he's uploading
38:51 his documents to inform this bot now uh
38:55 it's going to go figure out how to do
38:56 that and I would like i' like to show
38:57 you the configure tab so you can see
39:00 were built out got your name written all
39:01 over it by by the Builder itself and you
39:04 can see that there's capabilities here
39:05 that I can enable um I could add custom
39:08 actions these are all fine to leave um
39:10 I'm going to upload a file uh so here is
39:14 a lecture that I picked that I used to
39:16 that I gave with some startup advice um
39:18 and I'm going to add that here in terms
39:20 of these questions uh this is a dumb
39:23 one the rest of those are reasonable uh
39:27 and like very much things Founders often
39:29 ask um I'm going to add one more thing
39:31 to the instructions here which is be
39:33 concise and constructive with
39:36 feedback right so it sets it up and now
39:39 you just go in and edit it again if I
39:40 had more time i' show bunch of other
39:41 things brilant but this is uh this is
39:44 like a decent start and now uh we can
39:47 try it out over on this preview tab so I
39:49 will say um that's a common question
39:52 what are three things to
39:55 look what are three things to look for
39:58 when hiring
39:59 employees at an early stage
40:05 startup now it's going to look at that
40:07 document I uploaded um it'll also have
40:09 of course all of the background
40:10 knowledge of
40:12 gp4 look how fast it is too that's the
40:16 turbo that's pretty good those are three
40:17 things that I definitely have said many
40:19 times um now we could go on and it would
40:21 start following the other instructions
40:23 and you know you can add your own data
40:25 to your Bot you can add own data to your
40:27 Bot you can add your own data to your
40:29 bot then you can publish it I can work
40:32 on it later I can add more content I can
40:34 add a few actions that I think would be
40:35 useful we're gonna make money with chat
40:37 GPT so that's what it looks like to
40:39 create a
40:41 GPT with
40:43 thank that's
40:47 amazing I think I need this rolled out a
40:50 little slower I always
40:52 wanted God aming man someday I'll be
40:55 able to make a bot that will do this and
40:58 awesome so with gpts we're letting
41:01 people easily share and discover all the
41:04 fun ways that they use chat GPT with the
41:07 world you can make private gpts like I
41:09 just
41:10 did or you can share your Creations
41:13 publicly with a link for anyone to use
41:15 awesome or if you're on chat GPT
41:18 Enterprise you can make gpts just for
41:21 your company
41:22 beautiful and later this month people
41:24 have been asking about that in their
41:25 lives GPT store
41:27 later this month GPT store thank
41:31 you one person
41:35 CLA democracy of AI yep indeed you can
41:38 list a GPT there and we be able to
41:40 feature the best and the most popular
41:42 gpts of course we'll make sure the gpts
41:45 in the store follow our policies before
41:47 they're accessible so just like the App
41:50 Store Revenue sharing is important to us
41:51 here we go you can make money with C GPT
41:54 the most useful and the most used GPT
41:57 a portion of our Revenue wow we're
41:59 excited to foster a vibrant ecosystem
42:01 with the GPT store just from what we've
42:03 been building ourselves over the weekend
42:05 we're confident there's going to be a
42:06 lot of great stuff who said that tired
42:08 Runners seems like they learned a lot
42:09 from the plugins garbage so those are GS
42:12 exactly and we can't wait to see what
42:13 you build holy crap that's huge people
42:15 but this is a developer conference the
42:17 coolest thing about this is that we're
42:18 bringing the same concept to apis
42:24 amazing how do I sign up
42:27 well go to platform of you have already
42:29 been building if you want to get on the
42:31 developer side the playground go to
42:32 platform.
42:34 open.com playground and just get an
42:36 account there that's then you're on the
42:38 developer side and definitely pay your
42:40 20 bucks on the consumer side on chat
42:42 GPT and snaps my AI a customized chat
42:45 bot that can be added to group chats and
42:47 make recommendations these experiences
42:50 are it is isn't it jrc it is it's
42:52 emotional sometimes holy [ __ ] teams of
42:55 dozens of Engineers there's a lot to
42:57 handle to make this custom assistant
43:00 experience So today we're making that a
43:02 lot easier with our new assistance
43:09 API I was gonna say something API
43:12 includes persistent threads so they
43:14 don't have to figure out how to deal
43:15 with long conversation history oh
43:17 built-in
43:18 retrieval code interpreter a working
43:21 python interpreter in a sandbox
43:23 environment and of course the improved
43:26 function calling that we talked about
43:28 earlier so we'd like to show you a demo
43:31 of how this works and here is Raman our
43:33 head of developer experience
43:37 [Music]
43:37 [Applause]
43:38 welcome yeah Robert Stone it's with with
43:41 it being able to do python the math
43:43 thing is going to be is going to be
43:45 handled it's been so inspiring to see so
43:48 many of you infusing AI into your apps
43:51 today co go go to YouTube go to the open
43:54 AI Channel and click on their live
43:58 Tab and you'll see this video AG so
44:01 let's Dive Right
44:02 In imagine I'm building one theas a
44:05 travel app for Global explorers and this
44:07 is the landing page I've actually used
44:09 gp4 to come up with these destination
44:11 ideas and for those of you with the Keen
44:13 Eye these illustrations are generated
44:16 programmatically using the new dly 3 API
44:19 available to all of you today so it's
44:20 pretty
44:22 remarkable but let's enhance this app by
44:25 adding a very simple assistant to it
44:28 this is the screen we're g to come back
44:29 to it in a second first I'm G to switch
44:31 over to the new assistant playground
44:34 creating an assistant is easy you just
44:36 give it a name some initial instructions
44:39 a model in this case I'll pick dpt4
44:41 turbo and here I'll also go ahead and
44:43 select some tools I'll turn on code
44:45 interpreter and retrieval and save and
44:48 that's it our assistant is ready to go
44:51 wow next I can integrate with two New
44:54 Primitives of this assistance API
44:56 threads and messages let's take a quick
44:59 look at the
45:00 code the process here is very simple for
45:03 each new user I will create a new thread
45:06 and as these users engage with their
45:08 assistant I will add their messages to
45:10 the threads very simple and then I can
45:14 simply run the assistant at any time to
45:16 stream the responses back to the app so
45:19 we can return to the app and try that in
45:22 action if I say hey let's go to Paris
45:28 all right that's it with just a few
45:30 lines of code users can now have a very
45:33 specialized assistant right inside the
45:35 app wow and I'd like to highlight one of
45:38 my favorite features here function coing
45:40 if you have not used it yet function
45:42 calling is really powerful and as some
45:45 mentioned we're taking it a step further
45:46 today it now guarantees the Json output
45:49 with no edit latency and you can invoke
45:52 multiple functions at once for the first
45:54 time wow the the Jon thing was big
45:57 people were having to write around that
45:58 so the fact that they fixed that's huge
46:00 for
46:02 developers have the assistant respond to
46:04 that
46:05 again and here what's interesting is
46:07 that the assistant knows about functions
46:09 including those to annotate the map that
46:11 you see on the right and so now all of
46:13 these pins are dropping oh my God that's
46:21 remarkable and that if you didn't see
46:23 what just happened there the the thing
46:25 made a list of things to do in Paris and
46:27 then the function calling drop those
46:29 pins on the map the harmony you can
46:32 build between Ai and UI where the
46:35 assistant is actually taking action oh
46:37 my God but next next let's talk about
46:39 retrieval and retrieval is about giving
46:42 our assistant more knowledge Beyond this
46:44 iMed this is huge too this retrieval
46:46 thing this is what Lang chain does tiet
46:49 this basically puts Lang chain out of
46:50 business can drop here this PDF what
46:53 it's including I can just sneak peek at
46:56 it very typical United flight ticket and
46:59 behind the scene here what's happening
47:01 is that retrieval is reading these files
47:04 and boom the information about this PDF
47:07 appeared on the screen
47:09 wow wow and this is of course a very
47:12 tiny PDF but assistants can par longm
47:15 documents I need some problems to solve
47:17 I think we all do Lun deping what you're
47:19 building in fact I also booked an lbnb
47:21 so I'm just going to drag that over to
47:23 the conversation as well and by the way
47:26 we heard from so many of you developers
47:28 how hard that is to build yourself you
47:30 typically need to compute your un
47:31 biddings you need to set up chunking
47:33 algorithm now all of that is taken care
47:35 of wow and there's the chunking
47:38 algorithm means if you put in big um
47:42 documents it will auto chunk them which
47:44 means you know setting up a key value
47:46 store that means like handling the
47:47 context window serializing messages and
47:49 so forth that complexity now completely
47:52 goes away with this new stateful API so
47:54 so the question is this going to be
47:56 included with our paid subscription so
47:58 there's two different things here chat
48:00 GPT is the 20 bucks a month what they're
48:02 talking about now is the developer side
48:04 of this so if you're over on the
48:06 playground working with their API that's
48:08 what all this stuff is
48:10 so a lot of what developers have access
48:14 to over here is looks like it's
48:16 Incorporated in chat GPT but these are
48:18 two distinct worlds let's move on to a
48:21 new capability that many of for a while
48:25 wait let me interpreter is now available
48:27 today in the API as well let's watch
48:28 this code interpreter in the API the
48:30 ability to write and execute code on the
48:32 fly but even generate fights so let's
48:35 see that in action if I say here hey
48:38 will'll be four friends staying at DC
48:45 rbnb what's my
48:48 share of it plus my flights wow all
48:55 right here what's happening is that code
48:58 interpreter noticed that it should write
49:00 some code to answer this query so now
49:02 it's Computing you know the number of
49:04 days in Paris the number of friends it's
49:06 also doing some exchange rate
49:07 calculation behind the scene to get the
49:09 sensor for us yeah Papa J not the most
49:12 like math but you get the picture
49:14 imagine you're building a very complex
49:16 like finance app that's crunching
49:17 countless numbers plotting charts here's
49:20 here's your math tired Runner or whoever
49:22 talked about the math will work great
49:24 for you Yep this is now in the API I
49:27 think my trip to Paris is solid so to
49:29 recap here we've just seen how you can
49:31 quickly create an assistant that manages
49:33 state for your user conversations
49:36 leverages external tools like no leg and
49:38 retri yeah they're not they're they're
49:39 not going to show the answer
49:41 donaroo because it's probably
49:43 wrong how does Microsoft feel about the
49:46 demo on a MacBook they don't give a [ __ ]
49:49 they've got they've got Mac products
49:51 combined with our new modalities that we
49:53 launching most Engineers I know use Mac
49:56 I built a small custom assistant that
49:58 knows everything about this event wow
50:00 instead of having a chat interface while
50:02 running around all day today I thought
50:04 why not use voice instead oh here we go
50:08 voice for the dev conference here so you
50:09 can see it on the right awesome so on
50:12 the right you can see a very simple
50:13 Swift app that takes microphone input
50:16 and on the left I'm actually going to
50:18 bring up my terminal log so you can see
50:20 what's happening behind the scenes so
50:22 let's give it a
50:24 shot hey there I'm on the keynote stage
50:27 right now can you greet our attendees
50:29 here at Dev
50:33 day hey everyone welcome to Dev day it's
50:36 awesome to have you all here let's make
50:38 it an incredible
50:44 day isn't that impressive you have six
50:47 unique and Rich voices to choose from in
50:50 the API each speaking multiple languages
50:52 so you can really find the perfect fit
50:54 for your app and on my laptop here on
50:56 the left you can see the logs of what's
50:58 Happening behind the scenes too so I'm
50:59 using whisper to convert The Voice
51:01 inputs into text an assistant with gp4
51:04 tubo and finally the new TTS API to make
51:07 it
51:08 speak but thanks to function calling
51:10 things get even more interesting when
51:12 the assistant can connect to the
51:14 internet and take real actions for users
51:17 so we've been talking about this here uh
51:20 together Bots that can take
51:22 actions P assistant can you randomly
51:25 select
51:26 five de attendees here and give them
51:29 $500 in open eye
51:32 credits yes checking the list of
51:36 attendees
51:40 wow five de attendees and added $500 of
51:43 AP credits to their account congrats to
51:45 Christine M Jonathan C Steven G Lis K
51:49 and surj S all right if you recognize
51:52 yourself awesome congrats um yeah and
51:56 that's it a quick overview today of the
51:58 new assistant CI combined with some of
52:00 the new tools and modalities that we
52:02 launched we still have 20 minutes the
52:04 Simplicity of a rich text or voice
52:06 conversation for you and users we really
52:09 can't wait to see what you build and
52:11 congrats to our lucky winners this is
52:13 mindblowing you know what you're all
52:15 part of this amazing a community here
52:18 so one last time before I step off the
52:22 stage hey assistant can you actually
52:24 give everyone here in the audience $500
52:28 in open
52:29 credits sounds great let me go through
52:33 everyone look at
52:35 [Applause]
52:38 that all
52:41 right that function will keep running
52:43 but I don't have time so thank you so
52:45 much everyone have a great day back to
52:47 you
52:49 son the Oprah of AI That's
52:54 funny when is the remed class Kyle soon
52:57 as I figure this [ __ ]
52:59 out should I make a Dev account with the
53:02 same email yeah I use the same goes into
53:04 beta today and we are super excited to
53:07 see what you all do with it
53:09 any over time gpts and assistants are
53:13 precursors to agents are going to be
53:16 able to do much much more they'll
53:18 gradually be able to plan and to perform
53:21 more complex actions on your behalf as I
53:24 mentioned before we really believe in
53:26 the importance of gradual iterative
53:28 deployment unless he does it one more
53:29 thing important for people to start
53:31 building with and using these agents now
53:33 to get a feel for what the world is
53:35 going to be like as they become more
53:37 capable and as we've always done we'll
53:40 continue to update our systems based off
53:41 of your
53:43 feedback so we're super excited that we
53:46 got to share all of this with you
53:47 today we introduced
53:49 gpts custom versions of chat GPT that
53:52 combine instructions extended knowledge
53:55 and actions amazing we launched the
53:57 assistance API to make it easier to
53:59 build assisted experiences with your own
54:01 apps these are our first steps towards
54:04 AI agents and will be increasing their
54:06 capabilities over time we introduced a
54:09 new gp4 turbo model that delivers
54:12 improved function calling knowledge
54:14 lowered pricing new modalities and more
54:18 and we're deepening our partnership with
54:20 Microsoft in closing I wanted to take a
54:23 minute to thank the team that creates
54:25 all of this open ey has got remarkable
54:28 Talent density but still it takes a huge
54:30 amount of hard work and coordination to
54:33 make all this dunkaroo some of those
54:34 tools are for developers but the gpts
54:37 are for
54:39 consumers we do all of this because we
54:41 believe that AI is going to be a
54:43 technological and societal Revolution it
54:46 will change the world in many ways and
54:48 we're happy to get to work on something
54:50 that will Empower all of you to build so
54:52 much for all of
54:53 us we talked about earlier how if you
54:56 give people better tools they can change
54:58 the world we believe that AI will be
55:00 about individual empowerment and agency
55:02 at a scale that we've never seen before
55:05 and that will Elevate Humanity to a
55:06 scale we've never seen before either
55:08 we'll be able to do more to create more
55:11 and to have more as intelligence gets
55:14 integrated everywhere we will all have
55:16 superpowers on
55:17 demand we're excited to see what you all
55:19 will do with this technology and to
55:21 discover the new future that we're all
55:23 going to architect together we hope that
55:26 you'll come back next year what we
55:28 launch today is going to look very
55:29 quaint relative to what we're busy
55:31 creating for you now wow thank you for
55:33 all that you do thank you for coming
55:34 here
55:35 today there you
55:36 [Applause]
55:39 go holy crap there you have it
55:45 unbelievable
55:47 unbelievable all right what do you think
55:49 Queen yeah what we're going to see a
55:50 year from now is going to be quaint is
55:53 going to make this stuff look quaint
55:56 um the fact that that we're going to be
55:59 able to to make our own
56:02 [Music]
56:05 um let me get that
56:08 going make our own Bots with our own
56:11 data is huge so more bots so Kyle what's
56:15 up
56:16 um well the
56:19 so
56:22 Claude
56:24 um
56:26 the main reason I used Claude was for
56:28 long documents that's going to be no
56:30 longer necessary they so what they've
56:33 told us for all of this year is that gp4
56:36 was going to have a 32,000 token context
56:39 window one of the things he announced
56:40 today was 128,000 tokens Claude has
56:44 100,000 tokens so they just they just
56:47 flushed Claude down the toilet in terms
56:49 of a requirement if you have long
56:51 documents um one of the things people
56:54 ask for my lives all the time is but how
56:57 how do I get it to know [ __ ] about my
56:59 data they just made that easy for
57:01 consumers also easy for developers they
57:04 took all the stuff that Lang Chain's
57:06 doing not maybe not all the stuff Lang
57:07 Chain's doing a lot of the stuff Lang
57:09 chain is doing and they're making that
57:11 available to developers right within
57:13 chat GPT so that's a big thing um naming
57:17 these things gpts and not bot Bots is
57:20 brilliant right this is similar to when
57:23 when Apple launched their Vision Pro
57:25 goggles they never used the term
57:27 metaverse once right why because they
57:30 wanted to redefine it as spatial
57:33 Computing which they successfully did as
57:36 evidenced by the fact that meta referred
57:39 to what their goggles do is spatial
57:42 Computing like two months after Apple
57:44 did that so chat GPT by calling these
57:47 things
57:48 gpts that's all anyone is ever going to
57:50 call these things are you going to GPT
57:52 that oh give me that GPT hey do you have
57:54 a GPT that does this it [ __ ]
57:56 brilliant people don't want Bots we
57:59 already have a Hollywood narrative that
58:01 the robots are going to kill us so if
58:02 you call them Bots everyone will be
58:04 afraid of them but now it's just a GPT
58:06 go GPT that yeah yeah so that's
58:10 huge um I don't even really remember any
58:13 any of the other details those are the
58:14 biggies uh thanks for going live I think
58:16 that was great awesome thank you yeah
58:18 thanks for thanks for hanging out with
58:20 me how do how much time do I have I have
58:21 a few more minutes here um party next
58:23 year in San Francisco just got off the
58:25 open AI live stream and mind blown I
58:27 know Nexus Nexus the a AI Hub it's I I
58:31 mean the the amount the
58:34 am the amount of announcements that just
58:37 happened is like it's it's really hard
58:39 to unpack 128,000 token context window
58:43 the collapsing of everything into a
58:45 single tool that just does
58:47 everything gpts a GPT Marketplace a GPT
58:51 marketplace with Revenue sharing so if
58:53 you publish a publicly usable GPT that
58:58 probably ranks right in their different
59:01 categories they're sharing Revenue with
59:03 you um that's what we say to ourselves
59:06 when we think we can leave your live
59:09 unless he does one hour or unless he
59:12 does one more one more what oh one more
59:15 thing oh yeah
59:17 exactly that's good yeah although my my
59:20 one more things are never planned
59:21 they're always just you know chat add
59:24 brain fart
59:25 [Laughter]
59:27 oh my God con actions can take actions
59:30 on our behalf that was the other one
59:31 River can take take actions is huge oh
59:34 and then the fact that you can upload
59:36 just your documents to it amazing I've
59:38 never been so glad for a Monday morning
59:40 meeting was postponed thank you I know I
59:42 postponed a meeting I'm so glad I did
59:45 because it was only like an hour ago I
59:46 thought why don't I go live and just
59:47 talk about this stuff and though it was
59:50 it was you know Sherry was asking about
59:52 you know why am I crying and someone
59:54 else said I'm finding myself getting
59:55 really emotional I'm headed over to pie
59:57 for some therapy like this was oddd