OpenAI Co-Founder: Start Building With AI Before You Feel Ready | Greg Brockman — Transcript
Full transcript
- 0:00We're going to see this massive
- 0:01renaissance of entrepreneurship. Like I
- 0:02think it is coming. I think it's really
- 0:04going to kick in over the next year to
- 0:05two years.
- 0:06>> This is Greg, co-founder of Open AI, the
- 0:09company behind Chat GPT, which more than
- 0:11a billion users use every week.
- 0:14>> I'm pretty good at writing software. I
- 0:15don't write software anymore. I direct
- 0:17an AI for it to write software. I
- 0:19provide lots of [music] feedback and
- 0:20guidance.
- 0:20>> When people are building software, how
- 0:22do you strategize around building
- 0:24something that's not going to be made
- 0:25obsolete with the new models?
- 0:28I'm just about to show you something
- 0:29that I've built recently. Um, am I
- 0:32ambitious enough with AI?
- 0:33>> Well, I think the answer to being
- 0:35ambitious enough will always be no
- 0:37because there's always more we can do
- 0:39>> for someone who is willing to step up
- 0:41their agent game and they have 1 hour.
- 0:43Where do they start this weekend?
- 0:45>> I'd say first of all,
- 0:48>> this video is sponsored by HubSpot.
- 0:51>> I want to talk all things practical. I
- 0:53was recently talking to Nobel Prize
- 0:55laurate in economics and the one theme
- 0:58kept coming up that we have such capable
- 1:00AI models but we lack applications. What
- 1:04do you think that you're launching this
- 1:06year will help change that and help
- 1:07people realize what's the power of AI?
- 1:10>> Well, it's been very interesting to see
- 1:12first of all how many people use chat
- 1:15GBT. You know, we have 1.2 billion
- 1:17weekly active users. We have like 300
- 1:19million people who use chat every single
- 1:21week for health queries, right? to help
- 1:22them in their personal lives. At the
- 1:24same time, how many of those use cases
- 1:26just scratch the surface? And we see
- 1:28this in the metrics that if people use
- 1:31chat for three different use cases, so
- 1:33three different things that you realize
- 1:34chat could do for you, you become a
- 1:35power user. You become you retain you
- 1:37you keep using chat and you want to use
- 1:38it more. But getting to those three use
- 1:40cases, it's such a hard challenge.
- 1:43>> And in some ways, it kind of makes sense
- 1:45because okay, so here's this AI that can
- 1:47do anything, but what's one thing it can
- 1:48do, right? It's like sometimes
- 1:50>> what's the magic thing?
- 1:51>> Exactly. Right. And so really finding
- 1:53that magic that speaks to you, but also
- 1:54at the same time, I think it shows this
- 1:56massive opportunity, this massive space
- 1:58that's almost a gap where you have
- 2:00you're talking to an AI and the AI knows
- 2:02what it's capable of and it can suggest
- 2:03to you, hey, you asked me to do this
- 2:05thing for you. Actually, I could do even
- 2:06more, right? Here's an idea. I can make
- 2:08a whole presentation a spreadsheet. So
- 2:10that's actually been something we've
- 2:11been exploring. But I feel like there's
- 2:12something here about the fact that what
- 2:14you want out of AI is simplicity, right?
- 2:17You don't want an interface that has
- 2:18buttons and sliders and model pickers.
- 2:20Like none of that. That's not the AI we
- 2:21were promised.
- 2:22>> You want something you can talk to that
- 2:24can solve problems, that can help you,
- 2:25can help solve goals, but that can
- 2:27actually even be proactive, that can say
- 2:28to you, "Hey, you got this email that's
- 2:30in your inbox that you've been ignoring
- 2:32this person for the past couple days and
- 2:34that actually I did some research and it
- 2:36turns out that here's like the right
- 2:37answer to the question and I've drafted
- 2:39an answer. Do you approve?"
- 2:41>> And by the way, the use case I just
- 2:43mentioned, it's not theoretical with
- 2:44dots. I actually got that exactly uh
- 2:47this this this use case happening to me
- 2:49right before we we started this podcast.
- 2:51And so I think we're at a point now
- 2:53where AI is actually able to be much
- 2:57more proactive and to remove so much of
- 2:59the burden of figuring out like what is
- 3:01this system capable of and to really
- 3:04help you and then help you be able to
- 3:06achieve whatever it is that you want to
- 3:07do.
- 3:08>> Are there any other use cases that were
- 3:09unimaginable a year ago that are
- 3:11possible now and that people should be
- 3:13trying out? Well, I think so many,
- 3:15right? There's a whole spectrum of them
- 3:16from and I think the way that I always
- 3:18recommend going to to discovery is you
- 3:20got to play with the system, right? You
- 3:21got to see what it's capable of. And so,
- 3:23when it comes to coding, like these
- 3:24models, they're now better at writing
- 3:26software than basically anyone I know,
- 3:29>> right? I'm pretty good at writing
- 3:30software. Like, I don't write software
- 3:32anymore. I direct an AI for it to write
- 3:34software. I provide lots of feedback and
- 3:36guidance. So, it's always fun to like
- 3:37make a little game and just kind of get
- 3:39this this quick hit on just seeing like,
- 3:40okay, how long did it take to deliver on
- 3:42that? and then ask for something more
- 3:44complex. I I think that we're at a point
- 3:46now where Astra is so good at Blender,
- 3:48so good at 3D modeling. And so it's kind
- 3:51of fun to ask it to produce 3D models of
- 3:54a photo and those kinds of things. You
- 3:56can see it using software like using
- 3:58Microsoft Paint to to sketch out a
- 4:00photo. But you can ask for more real
- 4:02things, too. Like I know for example a
- 4:04friend who was designing or wanted to
- 4:07have a like guest house built and used
- 4:10Astra to actually produce CAD plans and
- 4:12then brought those to a contractor and
- 4:14now it's being built.
- 4:15>> And so it's a very empowering thing.
- 4:16Normally you'd have to hire someone else
- 4:18like I don't know how to make CAD plans,
- 4:19but now we have this intelligence in our
- 4:21pocket that can can actually do these
- 4:23kinds of things for us. To me, one of
- 4:25the things I really like to do is to try
- 4:26to push the AI in different dimensions
- 4:29of of my life. And so I think that voice
- 4:31mode is something that makes AI so
- 4:33accessible, right? You just go into
- 4:34voice mode and you start talking to it.
- 4:35Voice mode now is hooked up to tools. So
- 4:38you can actually ask it to do anything
- 4:40that your chat can do. And if you're
- 4:41hooked up to your email, to your
- 4:43calendar, now you can ask it about like
- 4:46what's my schedule for today or like
- 4:48what should I, you know, I want to free
- 4:49up some time for some deep thought. How
- 4:50can I do that? And it can actually go
- 4:52and have context on what's going on in
- 4:53each part of your calendar. So, I think
- 4:55that in my mind that there's this
- 4:57activity of just stepping back and
- 4:59thinking about what's a slightly more
- 5:00ambitious thing I can ask my AI to do
- 5:02than it's done for me to date. Let's see
- 5:04if it works.
- 5:05>> Yeah. And I'm just about to show you
- 5:06something that I've built recently. I
- 5:08think we're in this era where everyone
- 5:10can build a personal productivity app.
- 5:12Multiple. And uh this one that I've
- 5:14built, so I basically asked it to create
- 5:17a an app that's going to prepare me for
- 5:19the podcast. I tell it who I'm
- 5:21interviewing for how long and what's my
- 5:24success metric and it's basically the
- 5:26views on YouTube and then it goes ahead
- 5:27and creates a podcast. It does the
- 5:29research uh suggests topic to discuss
- 5:32this is great but how do I make it learn
- 5:35from my experience right and is there
- 5:37anywhere I can push this further cuz I'm
- 5:40going to record this conversation it's
- 5:41going to come here and it's going to
- 5:42check now it's checking my previous
- 5:43conversations but what do you think is
- 5:45missing here and am I ambitious enough
- 5:49with AI
- 5:50>> well I think the answer to being
- 5:52ambitious enough will always be no
- 5:54because there's always more we can do
- 5:55right and that's the beauty of AI is
- 5:56that it's really about solving problems
- 5:58helping you achieve what you want to
- 6:00achieve. You know, there's no ceiling to
- 6:01what you might imagine, right? And so,
- 6:03it's really just about sort of tuning
- 6:05that you were able to achieve a thing
- 6:07means maybe you can dream a little bit
- 6:08bigger. And so, in my mind, yeah, I
- 6:10think this I think this is super super
- 6:11cool app. So, f first of all, I in terms
- 6:14of just doing the background research
- 6:16and maybe I when you when this was being
- 6:18built, they already went and looked at
- 6:20all my interviews that I've ever done
- 6:22and did.
- 6:23>> Ah, there we go. Did all the background
- 6:24research.
- 6:25>> It did. It did the background. The only
- 6:26thing is it still can't connect to
- 6:28social media, but I think it's it's a
- 6:29plug-in thing. Yep.
- 6:30>> But then yes, it went through your
- 6:33conversations.
- 6:34>> There we go. And then all of Openai's
- 6:37past press releases. Oh, there we got a
- 6:39questions tab. I like that.
- 6:40>> Yeah, we got a questions tab. That's
- 6:42right. And that's the title and
- 6:43thumbnail already here. And these are
- 6:45the questions that I can copy.
- 6:46>> Yes. I'd be curious how much it helps
- 6:48with the post-processing and the actual
- 6:49production of the final of the final
- 6:51video once it's done. Right. because
- 6:53this thing has all this context on me,
- 6:55on open AI, on the kinds of things that
- 6:58you've covered in your your past
- 6:59episodes, and then how do you want to
- 7:01cut it? Like, I've I've heard there's so
- 7:03many good use cases of Astra helping
- 7:05with post-production for for videos and
- 7:08proactively being able to do that. And
- 7:09so, I think actually be I'd be very
- 7:11curious if you could have an app where
- 7:13it's just like the only thing you have
- 7:14to do is basically look over the
- 7:17questions, make sure you like them, and
- 7:18if you don't, provide some feedback, and
- 7:20it can do better. You show up, you
- 7:21actually do the interview, and then
- 7:23everything else is taken care of. How
- 7:24much work do you put in right now from
- 7:26the interview part to something actually
- 7:27being live on YouTube?
- 7:28>> Oh, a lot of work. I have a I have a big
- 7:30team, but it's also we've documented a
- 7:32lot and I've been doing this for 12
- 7:34years, so we have all the files that
- 7:35we're referencing, so it's easier to
- 7:37build with AI now, but it's still every
- 7:39episode is at least 40 hours.
- 7:41>> Wow. And if you could get back that
- 7:43time, right, so that kind of all of the
- 7:44the mechanics are taken care of. So you
- 7:46can spend so much time really thinking
- 7:47about what the right interview is and
- 7:50all of the the other things that go into
- 7:51this or really thinking about if you
- 7:53could scale you could do way more
- 7:54interviews like yeah what would you do
- 7:55with that additional time
- 7:56>> that's the question to you with small
- 7:58businesses who that are applying AI for
- 8:00their day-to-day jobs what is the role
- 8:02of the business owner what should they
- 8:04be thinking more of and where's actually
- 8:06the bottleneck for the business now
- 8:08>> I think that there's something
- 8:09fundamental so first of all we see so
- 8:10many small businesses that are being
- 8:12started because people feel secure with
- 8:15the fact that they have chatbt, right?
- 8:16That there's so many pieces of expertise
- 8:18that if you want to get a business off
- 8:19the ground, it's hard,
- 8:21>> right? But you actually have a PhD level
- 8:23intelligence, a world expert, business
- 8:26consultant. You have all these things in
- 8:28your pocket now for free. It's like
- 8:30pretty remarkable, right? And if you're
- 8:31willing, if you if you're able to pay,
- 8:32you can get even so much more depth and
- 8:34so much more compute towards any problem
- 8:36you want. So small businesses easier to
- 8:38start than ever and we're seeing it very
- 8:40concretely. And then people are able to
- 8:42I when you talk to many of these small
- 8:44business owners that there's just so
- 8:45many tasks that they outsource to an AI
- 8:47that you know certainly there's the ones
- 8:49that they knew that they had to do but
- 8:50there's so many tasks that they would
- 8:51just no one would have done it before
- 8:53and so little holes in their business
- 8:54strategy are thinking about how do you
- 8:56actually produce this document. The
- 8:57other thing too is is small business
- 8:59owners who don't even realize what they
- 9:00could be doing. like one of our our
- 9:02employees was telling me that he was um
- 9:05shopping for for watches in this like
- 9:06antique watch store and that he started
- 9:09talking to the owner about AI and that
- 9:11the owner was like, "Oh, AI, I can't use
- 9:12that for anything." Like, you know, I'm
- 9:14a I'm a watch store. And he was like,
- 9:16"No, no, AI can help everyone. Let me
- 9:17show you." And so he talked about some
- 9:19of the problems he has and he's like,
- 9:20"Well, yeah, he doesn't really have a
- 9:21website." And so you just Exactly.
- 9:25Right. And just like chatty
- 9:26[clears throat]
- 9:27voice being like, "Build me a website.
- 9:28Build an awesome website." And then the
- 9:30owner said, "Hey, I also have each watch
- 9:32has its own story, right?" And so it's
- 9:33just like then becomes very easy like,
- 9:34"Okay, talk to the AI about here's the
- 9:36story behind each watch." And you end up
- 9:37with this beautiful visualization of
- 9:38everything. And you just realize that
- 9:40there's this this way that technology
- 9:42becomes alive that normally it's almost
- 9:44inaccessible to so many people. And so I
- 9:46think that this really pushing on if you
- 9:48had like the world's best programmer on
- 9:51your payroll, what would you put them to
- 9:53work on? And and that now you can do I
- 9:56think you mentioned somewhere that it's
- 9:57actually ambition that small businesses
- 9:59have to start working on, right?
- 10:01>> Yes. Yes. And I I think that
- 10:03>> I think it's fascinating.
- 10:04>> I I agree and I think it is really about
- 10:06we can all raise a ceiling of ambition,
- 10:08right? We can all accomplish more people
- 10:10who really lean into that. I think we're
- 10:12already starting to see really
- 10:12remarkable things. I'll tell you another
- 10:14example by the way is um that that we
- 10:16have uh there are these two two brothers
- 10:18who are who are blind who use our
- 10:20technology and uh that they use it to uh
- 10:23both in their daily life and and for for
- 10:25their work and even for tasks like being
- 10:27able to plug uh the the cable into the
- 10:30right port on a computer, right? You can
- 10:32use chatbt to to actually basically
- 10:35provide that information when normally
- 10:36they would have to ask someone for for
- 10:38help. And so I think this kind of
- 10:39empowerment, you know, being able to
- 10:40help people be able to be more
- 10:42independent, be able to live their life
- 10:44the way that they want, all of that,
- 10:45it's happening. It's happening right now
- 10:46at massive scale.
- 10:48>> When someone is thinking of building
- 10:49something in software, when OpenAI is
- 10:51building so much around like work and
- 10:53productivity, what do you think makes an
- 10:55app worth paying for?
- 10:56>> There's definitely parts of what app
- 10:59building used to be that's becoming
- 11:00commoditized. It's becoming accessible
- 11:02to everyone. It's becoming massively
- 11:04democratized. And I think that's a
- 11:05really important thing, a beautiful
- 11:07thing, right? that there can just be
- 11:08more software and so then the question
- 11:10of yeah what is differentiation becomes
- 11:12even more important. I think that there
- 11:14is going to be more room than ever for
- 11:18software that adds value in unique ways
- 11:20and some of this is really about deeply
- 11:22understanding the problem. Some of this
- 11:23is deeply understanding your users,
- 11:25right? You think about something like if
- 11:26you want to build something for
- 11:28education, there's teachers, there's
- 11:30parents, there's students, right? That
- 11:31there's so many stakeholders that you
- 11:34have to really think about. There's
- 11:35administrators. How do you build a
- 11:37product that actually gives the right
- 11:38information, has the right guard rails,
- 11:40that is useful in the right ways and who
- 11:42are you selling to? All these questions.
- 11:43So, I think that these like human
- 11:45factors of what is the core of the
- 11:47business? What is it that you have
- 11:49discovered? What is it that's your
- 11:50insight? What is your unique way of
- 11:51looking at the world? And I think that
- 11:53the mechanics of how you instantiate
- 11:56that, how does you know do you have to
- 11:58write in C++ or Swift or something like
- 12:00who cares about that? Like that was
- 12:01never what the barrier should have been.
- 12:03>> Yeah, I've seen that YC startups recent
- 12:05batches are 60% B2B and my intuition is
- 12:09telling me with these tools available to
- 12:11us if I want to become an entrepreneur.
- 12:13I've seen so many problems in consumer.
- 12:15Do you think we're going to enter this
- 12:17era when more and more people are
- 12:18building consumer businesses? Well, I
- 12:20think that the line between what
- 12:22enterprise and consumer are is going to
- 12:24blur.
- 12:25>> I think that what a company is is going
- 12:27to become just less well-defined in some
- 12:31ways or it's going to be less like,
- 12:32okay, a company is like these like big
- 12:34masses of humans that are organized in
- 12:37certain ways. I think people are going
- 12:38to be so empowered, right? We're going
- 12:39to see this massive renaissance of
- 12:41entrepreneurship. Like, I think it is
- 12:42coming. I think we're seeing the leading
- 12:44edges of it, but it's really, I think,
- 12:46really going to kick in over the next
- 12:47year to two years. And I think that the
- 12:50question of are you selling to consumers
- 12:52or enterprises, there's some things that
- 12:55will remain true in terms of the split.
- 12:57Like I think that when I think about
- 12:58what enterprise sales is, it's fun
- 13:00fundamentally about relationships. But
- 13:01if you think about it as relationships,
- 13:03it's like well the reason that you can't
- 13:05really do enterprise sales to consumers
- 13:07is because there's just too many
- 13:08consumers, right? It's just too high of
- 13:10a scale. It's too expensive. But if you
- 13:12have AI that really amplifies your
- 13:13ability to have relationships, really
- 13:15meaningful ones and and really take care
- 13:16of people at massive scale, then
- 13:18actually maybe the enterprise sales
- 13:19motion becomes something that makes
- 13:20sense even for more consumer businesses.
- 13:22But the mechanics of what a business is,
- 13:24I think, will evolve. I think that the
- 13:26how you relate to your users will
- 13:29evolve. But the fundamentals of that you
- 13:32want to be delivering value and you want
- 13:34to build a relationship with the people
- 13:36who are receiving that value and have it
- 13:37be a birectional exchange. All of that
- 13:40is going to become much more front and
- 13:41center regardless of which domain.
- 13:43>> Greg is talking about how much more you
- 13:45can do now even with a tiny team. If
- 13:47that makes you think maybe I can start
- 13:50that business, I want you to think about
- 13:52your first customer. You want to listen,
- 13:55understand what they need and do a great
- 13:57job. And you also have notes,
- 13:58follow-ups, and a spreadsheet to update.
- 14:00Our sponsor, HubSpot, helps with that.
- 14:02Its new smart CRM is a CRM that updates
- 14:05itself, keeping your customer
- 14:07information and conversations together.
- 14:09Its AI notetaker captures customer
- 14:11calls, drafts follow-ups, and suggests
- 14:13next steps and updates to customer
- 14:15records for you to approve, so more of
- 14:17your attention can go to the people
- 14:19you're building this business for. Check
- 14:21out HubSpot Smart CRM through the link
- 14:23in the description. And thanks HubSpot
- 14:25for sponsoring this episode. Now, back
- 14:27to Greg. when people are building um
- 14:30software, how do you strategize around
- 14:33building something that's not going to
- 14:34be made obsolete with the new models?
- 14:36>> There's some activities I think scale
- 14:38well as the models get smarter and some
- 14:40activities that are kind of working
- 14:41around limitations in the model. The
- 14:43second one is pretty hard to be durable
- 14:45and so I think that there are at various
- 14:47points been startups that really focus
- 14:48on it. take the model, we add some
- 14:50prompts, we have some very handcrafted
- 14:52whatever. And it's not to say that
- 14:54there's there's a point in time where
- 14:55that is valuable. And sometimes it's
- 14:57even about the interface, right? That at
- 14:59each mo level of model capability,
- 15:01there's almost a new interface that
- 15:02suggests itself. In 2022, we had an AI
- 15:06that was personalized enough that it was
- 15:08worthwhile reading the responses GBD4.
- 15:11So it's like, of course, you want to
- 15:12turn make a chat GBT unit at 3.5. and
- 15:14GB4, you know, end of of 2025, I'd say,
- 15:17is when the coding agent era really
- 15:19started. And that was like you had AI
- 15:21that kind of shifted from being 20% of
- 15:23your software production to 80% of your
- 15:24software production. And so, you just
- 15:26want a very different product that
- 15:28captures that thing. This is the thing
- 15:30everyone's going to utilize. Now, I
- 15:32think we're starting to enter this era
- 15:34of the personalized agent. So, you're
- 15:36going to have a cloud-based AI and this
- 15:37is being raified number of different
- 15:39form factors. Openi, we just released
- 15:41Dots. I'm super excited about I think
- 15:43it's got the smartest model backing up
- 15:45this this kind of of technology and it's
- 15:48so useful. It's useful to me. I think
- 15:49it's be useful for lots of people. And
- 15:51so I think that the core thing is
- 15:52thinking about where will you scale as
- 15:56the models get smarter. And again, some
- 15:58of this is about understanding the
- 15:59industry. Some of this about building
- 16:00relationships with back to if you think
- 16:02about hospitals, right? There's so many
- 16:04different stakeholders there, right?
- 16:05That there's the patient, the doctor,
- 16:07the hospital admin, there's the uh
- 16:08insurance provider. There's so many
- 16:10different parties. And so how do you
- 16:12build trust across each of those and
- 16:15have a product that's useful to them and
- 16:17then actually be able to even if you're
- 16:19yeah when you think about if you have a
- 16:20model that's just even better if you are
- 16:22already in that industry you have these
- 16:24customers that actually your product
- 16:26gets better and more defensible as a
- 16:28result.
- 16:28>> Is there anything you particularly
- 16:30looking forward to um seeing startups uh
- 16:33use models to build something? I think
- 16:34the domain expertise is also really key
- 16:36because one thing that's easy to miss is
- 16:38that as something like open AI we have
- 16:40this technology that's so broad right it
- 16:42kind of touches every single aspect of
- 16:45economic growth and development people
- 16:47in their personal lives work lives
- 16:49there's something for everyone something
- 16:50that can really help them something to
- 16:52help them achieve their goals but often
- 16:54in any specific vertical we're so far
- 16:56from optimal right that if you you
- 16:58almost think of the economy as this
- 16:59fractal thing right where it's like you
- 17:01zoom in you're like okay here's a sector
- 17:03of like healthcare and then you zoom and
- 17:04you like really look at the number of
- 17:05different companies and the number of
- 17:07different ways of doing things and no
- 17:08matter what sector you pick, there's
- 17:10just so much low hanging fruit. That's
- 17:12actually been one of the most
- 17:13interesting and surprising things for me
- 17:15in building OpenAI is to realize that
- 17:17there's just so much suboptimality in
- 17:19the economy in terms of how you can
- 17:22actually have much more efficient
- 17:23process that gives people their time
- 17:24back that actually delivers the value
- 17:26that being able to connect across
- 17:28different fields is not something we do
- 17:29very much, right? It's like people
- 17:31specialize
- 17:31>> and now it makes economic sense to build
- 17:33them out because 5 years ago they would
- 17:35be like a feature of a big company now
- 17:37you can build a small company around
- 17:38that feature.
- 17:39>> Exactly. And I think that even these
- 17:41like classic sectors and domains are
- 17:43really because of the limitation of our
- 17:46tools. Again you think about how do you
- 17:48become good like an expert in any field
- 17:50is like you go to school you like
- 17:51specialize and as humanity has learn
- 17:53more things the what it means to
- 17:55specialize becomes narrower and
- 17:57narrower. Like if you're a biologist
- 17:58now, you get a PhD, there's like one
- 18:00specific reaction that you got your PhD
- 18:02on, right? And it's like it's not
- 18:03connected to the whole system. And I
- 18:05think that when it comes to to building
- 18:07a business, a question of can you have
- 18:09these much more hybrid things. So you
- 18:11think about things like edtech as an
- 18:12example of like you take people who
- 18:15understand about education, they
- 18:16understand technology, they mash them up
- 18:17and that they're able to innovate and do
- 18:19new things. I wonder if we're going to
- 18:21find that there's going to be other
- 18:22domains that we've never thought about
- 18:24that actually mesh together super well.
- 18:25>> Totally. That's the future of
- 18:26entrepreneurship, matching those domains
- 18:28and finding the right solution. Can you
- 18:29talk to me about a setup that you have
- 18:31that is sophisticated enough for me to
- 18:33compare this because sometimes people
- 18:36bash me in the comments for asking
- 18:38people like you about morning briefing
- 18:39agents. So, let's not do morning
- 18:41briefing agent. Let's do something that
- 18:42helps you run a company and be
- 18:43efficient. It's something that people
- 18:45see, they will think it's an aha moment
- 18:47for their productivity.
- 18:48>> I would not poo poo the morning briefing
- 18:50agent too much [laughter] because I
- 18:52think the morning briefing agent has so
- 18:53much depth to it, right? There's the
- 18:55like, okay, yeah, it's like it kind of
- 18:56gave me some info where it's like,
- 18:58honestly, I could have just scrolled the
- 18:59notification myself. There's the version
- 19:01of it where it says like, here are the
- 19:04things that are pending, right? Like
- 19:06here are these slacks that you have not
- 19:07answered. Um, and you really should get
- 19:09back to these people with this
- 19:10prioritization order. There's the next
- 19:12level of I have like done the research.
- 19:15you know, there's this question about
- 19:16like what our strategy should be in some
- 19:18particular market and like I've actually
- 19:19gone and I even like, you know, sort of
- 19:21looked at all the presentations and like
- 19:23here's like the things you need to like
- 19:24be paying attention to to make this
- 19:26decision. That third thing is like where
- 19:27I think we're at, right, is like the AI
- 19:29that really can do very sophisticated
- 19:31research across a bunch of different
- 19:32things and provide a very compressed
- 19:33version of of what I should be paying
- 19:35attention to. thing I'm actually very
- 19:36excited about as a next step that we're
- 19:37not quite there yet is the morning
- 19:40briefing agent so to [clears throat]
- 19:41speak that is like hey I actually called
- 19:43like you know I called up these people
- 19:45like I talked to you know these these
- 19:47co-workers of yours I talked to this
- 19:48external partner and like I got
- 19:50everything into perfect order like I
- 19:52just need you to approve this purchase
- 19:53order and like we're ready to go on the
- 19:55new whatever AI that really is proactive
- 19:57and able to do things on your behalf
- 19:59like that's coming and I think with DOTS
- 20:01we actually have the technology for it
- 20:03so I think we're going to start seeing
- 20:04people really start to utilize that. I
- 20:06think we have [clears throat]
- 20:06improvements to make and things like
- 20:07that. But I'd say in my mind, any of
- 20:09these use cases, there's so much depth
- 20:11to them. And and that is to me the key
- 20:13thing about AI. If you just look at the
- 20:15label of an AI that writes software,
- 20:18okay, sure, we've had an AI that writes
- 20:19software for 5 years now, but AI that
- 20:22writes software today, totally different
- 20:24from the AI that wrote software 5 years
- 20:25ago. So I have a lot of appreciation for
- 20:28the depth of application that's possible
- 20:31even if the like label on the tin is
- 20:33like the same thing and kind of
- 20:34everyone's heard it 100 times.
- 20:35>> When you look at these use cases inside
- 20:36OpenAI, do you have like an internal
- 20:38metric how advances in models is
- 20:40actually are actually translating into
- 20:42people's productivity?
- 20:43>> We published blog posts on this as well.
- 20:44So we've done a decent amount of of work
- 20:46looking at token usage and looking how
- 20:48it correlates with like lines of code or
- 20:50commits and PRs and things like that.
- 20:52And again, all these are proxy metrics.
- 20:53They're not perfect, but I think that
- 20:56another way that you can tell is AI
- 20:58really changing productivity and how
- 20:59much is if you ever have an internal
- 21:01outage, like if internally people can't
- 21:03get access to a model, um how much they
- 21:05feel upset. Everyone's like, "I can't do
- 21:07my work." And you're like, "Okay, 6
- 21:08months ago you had nothing like this and
- 21:10now you're saying you can't do your work
- 21:11when you don't have it for two hours."
- 21:12Like, how is that possible? But we see
- 21:14this all the time. It's so so cool. And
- 21:16I feel for myself like the kinds of
- 21:18things on the morning briefing agent
- 21:20access that I do a lot of is I like
- 21:22anytime I have a question about open AI
- 21:24I just ask my agent like I don't know
- 21:27why I would do anything else to start
- 21:28because it's just so efficient. It's
- 21:30like often and sometimes I even feel
- 21:31like I'm like being a little bit lazy of
- 21:33like there's like some message that
- 21:35someone has sent me with a doc in it.
- 21:36I'm just like trying to vaguely remember
- 21:38where was it and
- 21:39>> so just connect it to everything. It can
- 21:40it's going to do a mail slack whatever
- 21:42you're using and it has all the
- 21:43information. So whenever you need
- 21:44something, you just ask it and it has
- 21:46all this.
- 21:46>> Yes. And with with computer use, it
- 21:48really closes the last mile of being
- 21:49able to connect to to all my enterprise
- 21:51context. And so then that makes it so
- 21:53useful because I really have this chief
- 21:54of staff and I actually have a human
- 21:56chief of staff is also incredible,
- 21:57right? And that I kind of am able to
- 21:59like give her so much more leverage by
- 22:01not having to ask these basic questions
- 22:03that now my AI can answer. And some of
- 22:05these, you know, call call basic
- 22:07questions are also quite sophisticated
- 22:08like for example asking questions about
- 22:10different business metrics, right? So it
- 22:12actually has to then go and issue
- 22:13queries into our data warehouse and come
- 22:15up with like you know you're looking at
- 22:16like the growth in some market or the
- 22:18you know the retention curves or
- 22:19something like that. And I remember in
- 22:22you know my my last job when I was at
- 22:24Stripe that we had a data team that you
- 22:26would send like an email to and then a
- 22:28week later they would come back with an
- 22:29answer to that that query and now I just
- 22:31ask the AI and it just does it. It's
- 22:33it's incredible.
- 22:34>> Do you ever see areas where people are
- 22:36underutilizing tokens? Do you want to
- 22:38tell people like use more tokens for
- 22:40this? Oh, absolutely. And look, I think
- 22:42there's two dimensions to it. So, one is
- 22:44that there was this era of token maxing,
- 22:46right? Where it's just like people were
- 22:47just blindly saying throw more tokens at
- 22:48it, which is clearly not a primary
- 22:49objective, but it is a good proxy
- 22:51metric. If you're not trying to optimize
- 22:53tokens, you're probably going to then
- 22:55see improve productivity correlated with
- 22:58increased number of tokens. And we
- 22:59really do see it. We have leaderboards.
- 23:01We kind of look at that. And I think
- 23:02that we see like the people who tend to
- 23:04be the most sophisticated in their usage
- 23:06are just able to do really incredible
- 23:07things. Again, it all depends on the
- 23:09domain and things like that, but I love
- 23:11seeing, for example, in our finance team
- 23:13like that they've really gone full AI AI
- 23:16first. And I think that there's just
- 23:18like so many things where like workflows
- 23:20that used to take a long time for
- 23:21closing the books that are now just so
- 23:23much faster, right? And none of those
- 23:24are like everyone just feels that
- 23:26increased productivity and how how
- 23:27helpful it is. Uh we have an early some
- 23:30early adopters from our comm's team. I
- 23:31think that they're just like being able
- 23:33to manage like you have an event with a
- 23:35bunch of journalists coming and you're
- 23:36putting together a seating chart and
- 23:37trying to figure out everyone's dietary
- 23:39preferences and being able to manage all
- 23:40of that as one person because you have
- 23:42codeex that's actually taking care of so
- 23:44much of the back end for you. Like
- 23:45that's all been happening since early
- 23:47this year. So we see these kinds of use
- 23:49cases for knowledge work productivity
- 23:51everywhere. And then of course the thing
- 23:53that I think is also very easy to to
- 23:56miss and I think is actually coming soon
- 23:57for everyone is the scientific discovery
- 24:00side of things, right? that we put a ton
- 24:02of compute into solving Navier Stokes,
- 24:04you know, this this big millennium
- 24:06problem and that was like 10,000 agents
- 24:09for some number of days like it's a lot
- 24:11of compute and we're able to create new
- 24:13knowledge for humanity and you think
- 24:16about actually at much smaller scale I
- 24:18think it still is the case that we're
- 24:19seeing lots of breakthroughs happening
- 24:21and so I just have this feeling of like
- 24:23I think that people who really point
- 24:25compute at the right problem right have
- 24:28some judgment and some real thought on
- 24:29like what really matters where is the
- 24:31compute going to be utilized and then
- 24:33are willing to push that that token
- 24:34budget high. We're really seeing this
- 24:36like sort of outsiz impact.
- 24:38>> I think it's a big question that a lot
- 24:39of people have who are starting using
- 24:41agents and automating for the first
- 24:43time. What do you start with? How do you
- 24:45think about that?
- 24:46>> This white page problem is like the
- 24:48hardest problem in AI in some ways and
- 24:49utilizing AI is this question of where
- 24:51do you get started? What do you use? I
- 24:52think my answer is always start small.
- 24:54Like just try something, type something
- 24:56random, push enter, see what happens,
- 24:57right? And like I think that then as
- 25:00soon as you get that feedback loop then
- 25:02you can iterate, right? You can improve.
- 25:03And so I think that find like a little
- 25:06pain point in your life, right? Whether
- 25:07it's maybe it's an email in your inbox
- 25:09that you you know that you have been
- 25:11treading replying to or you know that
- 25:13you know it's going to take a lot of
- 25:14work to do to to get there. Um and so
- 25:17then the first question is well how do I
- 25:18get the email to the AI? Right? And so
- 25:20there's one answer which you copy paste
- 25:21it. There's another answer where you
- 25:23have your Gmail connector connected. And
- 25:25so I think that you can start very
- 25:27mechanically, very much going through
- 25:29the the effort yourself and then
- 25:31realize, okay, I've done this three
- 25:33times, I've done this five times, and
- 25:34now I understand how this process works.
- 25:37Let's automate it. And that, by the way,
- 25:38is like a general software engineering
- 25:40maxim. Like that's very much how I've
- 25:41always approached when you build
- 25:42software versus when you just kind of do
- 25:44it mechanically is after you've done it
- 25:45five times, it's probably a good time to
- 25:47to start automating. And so I think
- 25:48>> I like that the five times metric.
- 25:51>> It's a pretty good one.
- 25:52>> Exactly. Yeah. And I think I think that
- 25:53doing that for AI because that way you
- 25:55kind of keep your your finger on the
- 25:57pulse of the problem uh which I think is
- 25:58very important like at the end of the
- 26:00day you know back to the question of
- 26:02what are small business owners doing
- 26:03with AI accountability is something that
- 26:06remains with them right it's still the
- 26:07business is operating a certain way and
- 26:09the customer either signs up or doesn't
- 26:11and you know either pays or doesn't and
- 26:13so they should really care about that
- 26:14outcome and be deep on the details
- 26:16>> another side of this problem is how do
- 26:17you decide what not to automate and I
- 26:19wanted to ask you about with you know
- 26:22your mission in mind. How do you at
- 26:23OpenAI decide what not to build next?
- 26:26>> I think it's very important that humans
- 26:28remain in control and in charge, right?
- 26:29That the goals come from people. Point
- 26:31of this technology is to help people,
- 26:34right? To help empower everyone to help
- 26:36people achieve more of what they want to
- 26:38do. And so I think that that human first
- 26:41team humanity kind of of vibe is is very
- 26:43important, very core to our mission. And
- 26:45that's something that really bleeds
- 26:46through to how we think about things.
- 26:48And so when we think about what to
- 26:51build, we always think about how do you
- 26:52put in the right guard rails? How do you
- 26:54build trust? How do you ensure that
- 26:56there's good oversight, moniability, and
- 26:58that that is something that has been a
- 27:00very core focus of building all that
- 27:01technology for us for really since
- 27:04inception?
- 27:04>> For someone who is willing to step up
- 27:07their agent game and they have 1 hour,
- 27:09where do they start this weekend?
- 27:10>> I'd say first of all, sign up for chat
- 27:12to these dots. It's I I think just like
- 27:14a really cool new form factor. Again,
- 27:16you can talk, you know, chat a little
- 27:17bit, get some sense of what it does. Um,
- 27:19but connecting it to the appropriate
- 27:21context can really do work, right? It
- 27:22has its own cloud computer and it can
- 27:24optionally hook to your local computer.
- 27:26And I think that this picture that we've
- 27:28all had of like the way the coding
- 27:29agents work is this like task based
- 27:31system that you kind of micromanage that
- 27:32lives on your laptop. You close your
- 27:34laptop, it stops working. Like clearly
- 27:36that was never the AI that was promised,
- 27:37right? That was never the AI we pictured
- 27:395 years ago. And so I think we now have
- 27:40something that really looks like the AI
- 27:42that we were always picturing. And so
- 27:44again, start with a little task like ask
- 27:46it to go buy something for you or try to
- 27:50like I I really like texting mine,
- 27:51right? Just like go and and and really
- 27:53set up some connector so that it feels
- 27:55like okay, I'm I've got a communication
- 27:56channel to this now and ask it to do
- 27:59some recurring task, right? Or ask it
- 28:01for to do some research. And the thing
- 28:03that's very interesting is that DOTS can
- 28:06really, it's quite sophisticated, right?
- 28:07It's powered by GP6 Astra. So, it's
- 28:09actually got basically this incredible
- 28:10brain, like this smart, incredible, most
- 28:13intelligent model for any of these kinds
- 28:14of assistants powering it. And that
- 28:18means that it makes some connections
- 28:19between things like you ask it a
- 28:20question about, hey, like what's going
- 28:22on in this particular geography or
- 28:24something? And it will realize that
- 28:26that's something you care about. And
- 28:27then it might proactive within surface
- 28:29if there's a new news report about that
- 28:30that region, then maybe it will say,
- 28:32"Hey, you're doing
- 28:33>> some extra thinking for you, which I
- 28:34really appreciate." Do you think that's
- 28:36the future of work or do you think in 5
- 28:38years it's going to be something we
- 28:39can't even imagine now?
- 28:40>> Well, I think AI is always surprising.
- 28:42So, I think that there's going to be
- 28:43somehow that the fundamentals will be
- 28:45we'll have smarter AI that's like more
- 28:47accessible and more empowering and all
- 28:48those things. Um, but that the way in
- 28:50which you'll utilize it will be
- 28:52surprising relative to what we expect
- 28:54today. So, I do think that the form
- 28:55factors will continue to change. And I
- 28:57think that the like if you think about
- 28:59really we're gonna have this this AI
- 29:01that's able to solve these unsolved math
- 29:04problems in everyone's hands.
- 29:06>> And so what are people going to do with
- 29:08that? Like what are the kinds of
- 29:09breakthroughs that are going to become
- 29:10possible? And I think about that for
- 29:12medicine, right? For drug discovery is
- 29:14going to be incredible. Material
- 29:15science, we're just going to live in a
- 29:17world where people are going to be able
- 29:19to solve problems that before required a
- 29:21very small set of specialized experts to
- 29:23be mobilized behind them. Like you will
- 29:25be able to do that. you will be able to
- 29:26solve these challenges that no one else
- 29:29has ever thought about and there going
- 29:30to be some challenges that no one's ever
- 29:32faced before that you're going to be
- 29:33able to just like be the first person to
- 29:36to solve and I think that we're going to
- 29:38do this as this broad community and so
- 29:40the tools for that I think there's going
- 29:41to be so much more collaboration and
- 29:43these AIs are going to uh be something
- 29:45that will represent you um that you are
- 29:48going to have an AI that that is kind of
- 29:50there working for your interest
- 29:51companies will have their own AI so it
- 29:53will be I think there's like a lot of
- 29:55infrastructure to be built and building
- 29:56this in a way that's trustworthy and
- 29:58that is observable and that really
- 30:00uplifts humanity like that is I think
- 30:01very core to what we should collectively
- 30:03be doing.
- 30:04>> I think that's an amazing answer and I
- 30:05think we'll have homework to work on our
- 30:08ambition this weekend not just agents.
- 30:10>> All right.
- 30:10>> Thank you so much Greg. That was
- 30:11amazing.
- 30:12>> Thank you. Thank you.
About this transcript
This page contains the full transcript of OpenAI Co-Founder: Start Building With AI Before You Feel Ready | Greg Brockman by Silicon Valley Girl, generated from the public captions YouTube serves with the video. The transcript has 7,129 words across 1,010 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.