In short
Podcast Episode Notes: The Man Who Sold ChatGPT to the World - Zack Kass [Open AI]
Podcast Overview Title: Billions Host: Guillaume Moubeche Description: A deep dive into the power dynamics and mental frameworks behind building billion-dollar companies.
---
Episode Summary In this episode, Guillaume Moubeche interviews Zack Kass, the former Head of Go-To-Market at OpenAI, who played a pivotal role in transforming ChatGPT from a research project into a globally recognized product. Zack shares insights from his experiences at OpenAI and discusses the challenges and successes in marketing advanced AI technologies to large corporations.
Key Themes
- Introduction to OpenAI:
- Zack joined OpenAI when it was a small research lab (~100 people, $2M revenue).
- Initially focused on selling AI capabilities to Fortune 500 companies unaware of the technology.
- Development of ChatGPT:
- Discusses the early ecosystem around GPT-3 and the strategy behind ChatGPT's development.
- The chat interface decision was critical in making AI accessible and user-friendly.
- Rapid Growth and Market Response:
- ChatGPT's explosive growth and the vibrant atmosphere within OpenAI during its launch.
- The product became a household name through organic growth and word of mouth.
- Challenges of Scaling:
- Discusses the scaling challenges and personal burnout experienced during rapid growth.
- The need for a "bigger boat" metaphorically to manage the scale of demand and expectations.
- AI's Implications for Society:
- Explores the philosophical implications of AI as possibly humanity's last invention.
- Discusses the potential of AI in solving major societal issues, including healthcare and education.
- Global AI Competition:
- Discusses the geopolitical landscape of AI development, emphasizing the importance of equitable access to advancements in AI technology.
- Highlights the role of open-source initiatives in democratizing AI capabilities.
---
Detailed Timeline 00:00 - 06:34: Joining OpenAI as the First Sales Person
- Zack's initial impressions and the challenges of selling AI technology to non-technical audiences.
06:34 - 09:47: The Early GPT-3 Ecosystem
- Discussion on the initial product offerings and customer interactions.
09:47 - 11:03: Strategy Behind ChatGPT's Development
- The reasoning behind the decision to create a chat interface for accessibility.
11:03 - 16:45: Market Response to ChatGPT
- Analysis of the rapid adoption and user engagement with ChatGPT.
16:45 - 21:04: Lessons from Viral Growth
- Insights gained from the partnership with Microsoft and the factors contributing to viral growth.
21:04 - 26:45: Scaling Challenges
- The internal struggles of managing explosive growth and the implications on employee health.
26:45 - 30:21: Personal Health Crisis
- Zack's personal experiences with burnout and the importance of self-care.
30:21 - 33:44: AI as Humanity's Last Invention
- The philosophical implications and future of AI technology.
33:44 - 45:21: Technology, Inequality, and Policy Failures
- Discussing the intersection of technology and societal challenges, including healthcare and education.
45:21 - 48:35: Global AI Competition and Geopolitics
- The competitive landscape of AI development globally.
48:35 - 50:18: Potential of AI to Solve Major Problems
- The transformative potential of AI in various sectors.
50:18 - 54:33: The Next Renaissance and Education Transformation
- Anticipating changes in education and societal structures due to AI.
---
Key Takeaways
- Transformational Potential of AI: AI has the potential to solve many of humanity's pressing issues but requires thoughtful implementation and policy support.
- Importance of User Experience: The success of ChatGPT was heavily influenced by its accessible and user-friendly chat interface.
- Personal Well-Being in High-Stress Environments: Zack emphasizes the necessity of self-care and mental health awareness in fast-paced startup environments.
- AI’s Global Impact: There's a need for equitable distribution of AI advancements globally, with a focus on addressing structural societal inequalities.
- Future of Education: AI is likely to revolutionize educational practices and experiences, moving away from traditional models.
---
References
- Key Personalities Mentioned:
- [Sam Altman](https://fr.wikipedia.org/wiki/Sam_Altman)
- [Brad Lightcap](https://www.linkedin.com/in/bradlightcap/)
- [Lukas Biewald](https://www.linkedin.com/in/lbiewald)
- [Chris Van Pelt](https://www.linkedin.com/in/chrisvanpelt)
- Key Concepts:
- [Attention Is All You Need](https://papers.nips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf)
- [Next Renaissance: AI and the Expansion of Human Potential](https://www.amazon.fr/Next-Renaissance-Expansion-Human-Potential/dp/1394381085)
---
This detailed summary encapsulates the critical insights from the podcast episode, reflecting both the personal experiences of Zack Kass and broader discussions on the implications of AI technologies.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Chapters
Tap a time to open that second in VOZack Kass's Journey at OpenAI
0:45 to 2:10
Zack discusses his early days at OpenAI and the challenges he faced.
“He built a playbook for Microsoft, for Coca-Cola.”
From Lab to Superpower
2:10 to 4:25
Zack explains how OpenAI transitioned from a research lab to a major player.
“was a, uh, equal understanding of the size of the opportunity to, to change the world with, with, with unmetered intelligence, uh, and, uh, equal part sort of weight of the importance of getting it right.”
Building the Go-To-Market Strategy
4:25 to 6:40
Zack shares insights on how he developed the go-to-market strategy for OpenAI.
“It was very early neural nets, started also by two Stanford researchers.”
The Early Days of GPT-3
6:40 to 8:45
Zack reflects on the initial launch of GPT-3 and its market reception.
“And extra large in this case was 175 billion tokens.”
Lessons from Launching ChatGPT
8:45 to 11:15
Zack discusses the launch of ChatGPT and the key lessons learned.
“And that's sort of what happened with us.”
The Future of AI and User Experience
11:15 to 14:06
Zack elaborates on the importance of user experience in AI products.
“or even you could predate that 2020 when GPT-2 comes out when we start seeing a lot of progress with large language models and November 30th, 2022 when ChatGPT comes out.”
The Launch of ChatGPT
14:06 to 15:32
Learn about the unexpected success and rapid adoption of ChatGPT.
“the technology, they'll want to build something similar.”
Company Culture and Internal Reception
15:32 to 19:28
Discover the internal dynamics and expectations during ChatGPT's launch.
“And that set me on my own journey, my own journey of sort of education and exploration, where I realized that so many of the conversations that were going to be had for a while, were going to be about, do you use AI?”
The Role of Word of Mouth in Growth
19:28 to 21:48
Explore how word of mouth contributed to ChatGPT's explosive growth.
“And basically, like, when you launch such product and the traction is pretty insane, like, do you have a strategy to make it just to increase the traction?”
Partnerships and Pricing Strategies with Microsoft
21:48 to 25:44
Understand the strategic partnership with Microsoft and its impact on pricing.
“This is one of the things we always joke about.”
Show all 22 chapters
Scaling Challenges and Industry Impact
25:44 to 28:01
Learn about the scaling challenges faced by OpenAI amidst rapid growth.
“I just remember being blown away I remember multiple times being blown away by the scale and there's this line I don't know Guillaume have you seen Jaws the American film?”
The Surge of AI Demand
28:01 to 29:09
Understand the rapid growth and demand for AI technologies.
“GBT5 starts coming down in price and improving in quality and GBT4 is coming.”
Personal Struggles and Realizations
29:10 to 33:08
Explore the speaker's personal challenges and journey to wellness.
“Because, I mean, you had to deal with a scale that eventually, like, you hadn't seen in the past.”
AI as the Last Technology?
33:09 to 36:48
Delve into the concept of AI being the final major technology invented by humans.
“I mean, it's nice to see the way you've come through.”
The Future of AI and Inequality
36:49 to 41:37
Discuss the potential of AI to elevate society and the risks of increasing inequality.
“And I actually think part of the problem is that we don't ground the debates.”
Policy Failures in Technology Adoption
41:38 to 42:01
Examine how policy decisions impact the accessibility of technology.
“And every generation's greatest privilege was to pass their luxuries to the next generation as commodities, and our ruling class broke the promise.”
The Reckoning of Inequality
42:01 to 42:44
Discover the challenges of inequality and the potential of technology as a remedy.
“They can have this exceptional healthcare.”
Technology's Role in Healthcare
42:45 to 44:35
Learn how technology can transform healthcare and address upstream issues.
“They're going to drive down the cost to actually build.”
AI Race and Global Dynamics
44:36 to 46:40
Explore the competitive landscape of AI development globally and its implications.
“has what they need and can spend time with friends and family.”
AGI: Current State and Future
46:41 to 48:19
Examine the current state of AGI and the pressing problems it might solve.
“I don't know many Americans who - As long as it's cured.”
The Future of Education
48:20 to 51:19
Understand how AI will revolutionize education and the expectations of future generations.
“and you feel like we are very close from solving it?”
Parenting in the Age of Technology
51:20 to 52:34
Gain insights into the values and lessons for children in a tech-driven world.
“at the cost of the soul of every child, right?”
Transcript
Automatic transcript. May contain errors.0:00Zack Kass:Most of my friends look to be a little weird, but I would do it again and again and again, throw a wrapper on GPT-3 and convince people that you had invented fire. Almost eight months. Anyone could have built, chat GPT, we did, and we'd be slightly aligned, slightly fine-tuned to dialogue, but not much. Finally, the policymakers won't have a single excuse for why everything is prohibitively expensive for everyone except them.
0:23Billions Host:Today on Billions, I'm sitting down with a man who had the hardest sales jobs in Silicon Valley, Zach Kass. Before ChadGPT was a household name, it was just a research lab and Zach was head of GoToMarket. He joined when OpenAI was around 100 people doing$2 million in revenue. His job? Sell human-level intelligence to Fortune 500 executives who didn't even know what was a token. He built a playbook for Microsoft, for Coca-Cola. he turned a non-profit lab into an 80 billion dollar superpower then he walked away dax thank a lot for being here thanks for having me young yeah i'm super excited to be honest uh to have you because i think you know when when you joined uh open ai back in the days how was it like you know like to to be in uh in this kind of like scientific lab and then becoming it like the such a such a
1:21Zack Kass:big name and a big launch well i think uh yeah not now of course it's it's there's a very romantic idea about joining a company that's just building science uh with the opportunity to monetize it but i'm not sure it was it was so uh i'm sure actually at the time that it was not so romantic most of my friends looked at me a little weird but but it i would i would i would do it again and again and again and I would recommend it to anyone I think there is uh there's a magic to joining a company that is building something so incredible with the understanding that it will eventually figure out how to how to um monetize it and in the case of open ai you know we we had just launched gbt3 and for a long time if you emailed sales or support at you you arrived at my inbox and there was a, uh, equal understanding of the size of the opportunity to, to change the world with, with, with unmetered intelligence, uh, and, uh, equal part sort of weight of the importance of getting it right.
2:32Zack Kass:The emphasis on safety and security. I'll tell you more about that in a second and, um, an equal part understanding that it would be very hard that like the challenge ahead was super intense. And those are feelings and experiences I will never forget, just in those early days in early 2021.
2:57Billions Host:And basically, when you joined, what was kind of the pitch from the company to hire you? What did they say they were working on? What was your knowledge about AI? Why did you think it was the right company and the right opportunity to join?
3:16Zack Kass:So I had spent my entire career to that point in machine learning. Okay. And when I talk about this, people always go, you're lying. I say, well, here, here's my resume. So my first job was at a company called Crowdflower, which became Figure 8, started by Lucas B. Walden, Chris Van Pelt. Now the CEOs of, were the founders of Weights and Biases, which was acquired by CoreWeave. and they were building the first scale ai long before scale ai there was figure eight and figure eight was helping machine learning engineers build human labeled data sets uh quality controlling for your massive scale human interaction and uh that's where i sort of under cut my teeth on on machine learning and we you know we had a small set of customers the company ended up getting inquired for about$400 million, but we had a small set of customers spending a lot even at the time because the number of companies that could actually do anything with machine learning was very narrow.
4:19Zack Kass:And then I went to a company called Lilt, and Lilt was building machine translation services using large language models. It was very early neural nets, started also by two Stanford researchers. Worked there for four years building the early, or one of the early, uh, AI enabled services companies. So long before there was Harvey or another, there was, there was Lilt doing machine translation, uh, or empowering human translators with machine translation. And so when it came time for opening, I had to hire a go-to-market person. They didn't call it that by the way. I think that, I think the job description was sales manager or might've been salesperson.
4:59Zack Kass:It might've just been accounting executive, but there, I was the first person. There was no one there selling the thing. And so they cast a net and a few people were like, well, this is the only other guy that's ever sold large language models. This is the only other guy who's ever sold machine learning products the way you probably want to. And you may want to chat to him. Now, I had heard of OpenAI for quite some time. Attention is All You Need had been out for a few years. Transformer architecture was gaining momentum. It was clear this stuff was going to work. It just wasn't clear on what timeline and who was going to do it.
5:32Zack Kass:And the prevailing sentiment at the time was still, oh, it's Google's game. It's, you know, it's, it's Amazon's game. And, uh, when I went to the office and sort of met people, I was like, okay, this is, this is very real. This is very exciting. Although all my interviews I'll never forget were, were over the camera because it was at the height of COVID and I had just torn my Achilles. So, um, so it was, it was pretty clear it was going to work. because, you know, placing the exact bet was going to be hard, but you have to place some bets. So I, so I took this one and the pinch from the company at the time was help people who are struggling using the API get support and sell companies when they show up looking for other things.
6:14Zack Kass:So we built the first set of support products and we built the first set of, you know, tiers that you could purchase. And we built the first, you know, marketing site and we did all the things that most SaaS companies do and then we scaled it and we you know created a partnership team and a sales team and a solutions team and and the rest is history because basically when
6:35Billions Host:you joined they were like just a hundred people and uh and I think it was at like two million in revenue what was the thing that they were selling where they're just selling like the the access to
6:45Zack Kass:the API and uh selling tokens or so there were four model types at the time you had uh GBT small, medium, large, all the way up to extra large. And extra large in this case was 175 billion tokens. And small was ADA, medium was Babbage, large was Curie, and extra large was Da Vinci. And these were, of course, named after really important scientists. And of course, there's no model routing. We didn't really, you know, we, we, we made a lot of assumptions. We didn't do a ton of really powerful product marketing. And so we sort of let you, we said like, look, just sort of figure it out. Um, you'll figure out which model works best, works best for you.
7:35Zack Kass:And, and, um, when GPT-3 had been launched, the idea was it would, you know, developers would use this to do, again, you know, human intelligence tasks, but it required an enormous amount of quality control. So the original set of customers were actually people building wrappers, as you of course remember, on top of GP53 to then sell a better product marketing product to the market. And so most of the early winners, winners is a sort of loose statement, but most of the early big customers were Jasper and other GBT3 wrappers who had figured out that they could build marketing sites and sell people this magic while sort of obfuscating all the messiness that was happening in the background, principally by them picking the right model and doing some quality control.
8:30Zack Kass:And it's really incredible to think that that was only five years ago, that it was only five years ago that you could just sort of throw a wrapper on gbt3 and convince people that you had you know invented fire uh and they were arbitraging an incredible amount of revenue and we were capturing very little and i think we all were like well why but you know early markets have tons of weird information asymmetry and they're super messy and this was just sort of how it had to happen i in the same way that the early ebay days most great ebay sellers were just buying goods that that they could find at low prices and selling them for market clearing rates.
9:08Zack Kass:And that's sort of what happened with us.
9:12Billions Host:And whenever you were looking, let's take Jasper for an example, because I think it grew extremely fast and it was like pretty popular at the time. When you looked at it internally, what was kind of the whole pitch? Were you saying like, yeah, we're going to come up with ChatGPT and a new interface that will be like so simple that people will be able to use it without having a wrapper or was it part of the strategy to kind of like uh just remove this kind of wrappers and this industry or how did you know yeah no i i don't think
9:47Zack Kass:there was never discussion about sort of uh doing something because a a customer of ours was was doing so well with it, the strategy was always, and I hope continues to be, producing the best, safest intelligence products that allow people the most convenient and sort of understandable access to them. How do you make exceptional technology very safe and very accessible? that was always the north star
10:28Billions Host:and back then I don't know if you remember but at the time of ChatGPT launch I remember on Twitter you had like kind of two visions one people one type of person would say like like having a chat interface is really not like the future of anything we've seen chatbot it doesn't work like this sucks etc etc and then we had people who were like okay this is amazing this is the best thing that's that's ever happened so internally prior to launching chat gpt within the company how did you decide about the interface and what were kind of the exchange you know within the company where everyone really pro the chat interface or were they other like
11:14Zack Kass:approaches well so a lot you have to remember a lot happened between may 2021 when gpt3 comes out or even you could predate that 2020 when GPT-2 comes out when we start seeing a lot of progress with large language models and November 30th, 2022 when ChatGPT comes out. I mean there's a lot in between those two events and I say that because I think it's there's a lot of things that you have to remember about that period to make sense of why the chat interface was so important. And the principal reason, I can explain in a pretty simple way. When we launched GPT-3, there was a real sense that we had sort of changed the game.
12:10Zack Kass:And we were so excited to see people start to really use this technology. And 175 billion parameters was a lot. And the model was super performance and um not many people showed up and in fact jasper was capturing a ton of the attention and and other and other writing tools um and and the market was sort of finding a lot of joy in a very specific sort of rapper instance which we were fine with but nothing else really and it took a while for us to sort of figure out it you know is there even a product here? Are we selling, should we be selling solutions? What should we be selling? When we launched GBD 3.5, there was a sense, there was this true palpable sense of excitement just before we did so in May, a year later, May 2022, that we were this time was going to be different, that we were really going to change the game.
13:06Zack Kass:And I want to make sure I get that right. Yeah, in May 2022. it. And, or March, sorry, when we launched GBT 3.5 in March, 2022, there was a real sense that this time was going to be different and that now the models were going to be better, faster, and somehow also cheaper. And then everyone was going to do interesting stuff. Well, then revenue ticked up from like 20 to 25 million over the next few months. And again, there was the sense of like, man, we have to make it, all the stuff that these folks are doing to make it easy for people to see how how much you can do maybe we need to do that and so and i tell this story because the the purpose of chat gbt was principally to help people see the value of the models in this in the pursuit of getting people to use the api and it was launched as a research preview the idea being that if you get someone to see how cool it is to interact with the technology, they'll want to build something similar.
14:08Zack Kass:And then we'll have this Cambrian event where everyone will be finally building with our APIs. And what you discover when you launch ChatGPT, what we discovered when ChatGPT was launched, is one, that the science was never the bottleneck. The research was never the bottleneck. ChatGPT exploded in popularity, trained on GPT 3.5, a model that had been available for almost eight months. For almost eight months, anyone could have built ChatGPT. We did, and we were slightly aligned, slightly fine-tuned to dialogue, but not much. Second thing we discovered, that you really had to bring people along for the journey, that an API was too complex, that what people needed more than anything was to see and feel and touch the technology, which is why the chat interface ended up being the perfect thing.
15:04Zack Kass:A magical AOL instant messenger was the thing. And the third thing we realized is that forevermore, or at least for a while, the chat interface was going to be AI. That the discussion about AI as intelligence was actually then suddenly going to shift to AI as a magical AOL instant messenger, an app, a consumer app, or eventually, of course, Chacham BD Enterprise, an enterprise app. And that set me on my own journey, my own journey of sort of education and exploration, where I realized that so many of the conversations that were going to be had for a while, were going to be about, do you use AI?
15:43Zack Kass:Implying, do you download an app and use it versus all of the incredible ways that unmetered intelligence was going to change our life. And that for better or worse, there was going to be a myopia around what this technology could do because of the correlation between ChatGPT and the underlying technology, which of course was going to have positive and very negative implications on how we actually evolved.
16:10Billions Host:And whenever you tested for the first time internally ChatGPT before anyone else, how was it within the company? Did everyone felt like it was going to be a success or were people doubting?
16:27Zack Kass:I think the company had become accustomed to doing things very well. And Brad Lightcap, the CEO, tells a story about the bets that were taken at the executive leadership team on the popularity of the product. and um i i can't remember exactly where people landed but you know most people you know say sam famously adds a zero to to most people's uh ideas of what of what success should be and i'm not sure this was any different but i still don't think people there was no real sense that this was going to be as big as it was i mean how could how could you believe that and if we believe that we certainly would have assigned a lot more compute to it because it crashed many times in the first in the first couple weeks um and of course we might have built an app beforehand it was just a web app but but it was uh no i mean it was there was an understanding that there had been an understanding since we launched gbt 3.5 yeah like there there was no understanding that what we had built was was incredible it was just a matter of figuring out how to get people to make sense of it.
17:38Billions Host:And when the traction was insane in the first couple of weeks, and then it kept growing insanely, what was the atmosphere at the office?
17:52Zack Kass:Like a sober excitement. I think people again sorry again I think so much of so much of what open ai was working on at the time felt important so it wasn't it wasn't like this was a consumer application company that had finally found it's not like this was zynga and we finally found the game that everyone wanted to play like we were you know we were at the casino rolling dice until we until we got the until we got the the the right combination we had been building exceptionally important research for a while and the question wasn't will this research matter it was how do we get people to care so yes i think there was a real sense of excitement among the employees that suddenly we were going to matter to to people but there was no more sense of importance or weight because what was really exciting that we knew was coming because this is this is the thing by the time chat gbt came out everyone knew gbt4 was coming and the timing could not have been better so we knew that if people were excited about this thing now they were going to be blown away in a few months and of course a few literally a few months later, it was four months later or four and a half months later that GPT-4 would come out.
19:23Zack Kass:And we were already, you know, private previewing GPT-4 at the time internally.
19:28Billions Host:And basically, like, when you launch such product and the traction is pretty insane, like, do you have a strategy to make it just to increase the traction? Or was it just pure word of mouth because you have found like the the perfect market fit and you know like that that just worked by itself or were you also on the on the gintm team you know like engineering a few things so basically you can capitalize on the huge growth and the huge pmf the the story here is is a brilliantly designed product made publicly available to anyone who had access to
20:15Zack Kass:the internet for free and a um back-end team that kept the product alive for most of the time i mean i don't i just don't think that i don't think any other story could or should be told okay and And I would love to say, look at what I did. This is a function of truly lightning in a bottle. And when people are like, what lessons do you have? I'm like, forget my lessons from OpenAI. I have lessons to share. Some are personal, some are professional. But my go-to-market advice will almost never apply again. because what we did, what the company did, what the products did were so unique, such moment in time, sort of stroke of genius on many basis that I think it's actually quite hard to give anyone specific advice from that period.
21:16Billions Host:I think what's interesting with what you say, I feel like for the last maybe five to ten years, everyone especially like in the SaaS world everyone has been telling like you should focus on distribution if you own distribution like you need to own your channels etc etc and then you come out with chat GPT and then you know it's like none of the advice actually works it's just an amazing product that is accessible to everyone and then word of mouth make it used by hundreds of millions within like just a few months
21:53Zack Kass:yeah yeah and not i think about that all the time not only do i think about the fact that it is there is um you almost are breaking the rules in this one it was a web app it was a web app that had no instructions it was a um web app that had no instructions that went that went the constantly was rate limiting people or crashing, it had no virality. There was nothing. This is one of the things we always joke about. There was nothing in the product that encouraged you to share the product with someone else. I mean, it was truly word of mouth. It was one of these incredible moments. But this goes back to the fact that if you want to point at why it was successful, it was successful because GB3.5 was always going to be successful.
22:42Zack Kass:It was just a matter of figuring out the means to help it reach people. And this was it. This was sort of in retrospect, obviously it. Yeah.
22:53Billions Host:And you also started to work on more of the enterprise aspect. I think you worked on a Microsoft deal. So can you maybe share a bit the insight of that deal and how you kind of structured it?
23:11Zack Kass:Well, no, I can't. It wasn't my deal and it's not mine to share. But what is public, what is fairly public domain now is that Microsoft made a sizable investment in OpenAI in late 2020 and that the investment was sort of contingent on some success criteria, some of which was revenue growth, some of which was product success. And that the idea is that Microsoft would be one of the distribution channels for OpenAI, which I got to work on quite a bit and figuring out how Microsoft salespeople would sell this, figuring out how we could get this embedded in Microsoft products, or figuring out actually how we would sort of co-brand and co-market.
23:56Zack Kass:And that was a ton of fun because you have one of, and that I learned a lot from, because you have a true titan of industry, one of the great behemoths of all of all software and maybe the greatest enterprise sales organization in the world probably um and there's a lot to learn from when you spend time with with those with those people in those teams about their expectations in particular their expectations for scale and what i remember learning a lot from microsoft in the early days and especially after chat gpt launch was what they were preparing for in terms of scale and i remember going back after a day with Microsoft and being like, they, they want to, they want to train a hundred thousand, they want to train 70 ,000 salespeople.
24:43Zack Kass:And I was like, what does that even look like? Like, how does, like, what does that even, how, you know, and, and Microsoft put pressure on us to keep bringing the price down, which was a really important pressure that I think we were like, whoa, whoa, whoa, why would we do this? And Microsoft was going to like drive the price to the bottom, which put a bunch of pressure on, on compression and all the research. to actually make the thing cheaper. I mean, long before DeepSeek drove the market prices down, lots of researchers were trying to do it too, principally because there was this belief in Jevin's paradox, which Microsoft had always known.
25:20Zack Kass:They had basically known the less expensive you make stuff, the more our salespeople are going to sell it. And this ended up being very true, like incredibly true. I mean, you know, GBT3 was$10 per million tokens, which is incredible to consider, right? GBD 5.2 isn't even, I think it's$5 per million tokens today. So it's like our expectation for how good this technology has gotten doesn't even compare to our expectation for how cheap it's gotten. And I think that to me is one of the incredible reminders in all this, which is that distribution is that the rules around go-to-market are now constantly changing because everyone's expectations for how good something is are matched often by how inexpensive it will become or how customizable it would become, which is just kind of incredible.
26:11Billions Host:And how quickly do you think, I mean, how quickly did you start implementing like ChatGPT within Microsoft and how quickly did it take for the 70 ,000 sales rep to be trained on what they needed to solve with this new partnership? Well, I mean, I assume it.
Read the full transcript
26:33Zack Kass:I just remember being blown away I remember multiple times being blown away by the scale and there's this line I don't know Guillaume have you seen Jaws the American film? No I don't think so Okay well you should see it it's a good American film and it's one of our great films and there's a line in Jaws they're hunting a great white shark the premise of the film is they're trying to get a great white shark
26:54Billions Host:Jaws okay sorry yeah yeah yeah yeah
26:57Zack Kass:they've been paralyzing this coastal town and they don't know how they haven't seen the shark yet and so finally did this one of the great what truly one of the great scenes in in american film they see the shark and they you know they're wrestling with it and the shark just rips the rips the lime or maybe take drags the boat down i can't remember exactly and then the protagonist of the show turns to to his sort of you know tough but but but sort of stoic friend and goes we're gonna need a bigger boat and that line that i like delivering like that line has become synonymous with sort of anytime someone realizes the scale of the challenge they face and you know it's it's used probably thousands of times a day in the American social lexicon.
27:53Zack Kass:And I said it to myself and people around me so many times because once ChadGBT arrived and once we knew that GBT4 was coming, the confluence was just wild. ChadGBT, November 30th, 2022. GBT5 starts coming down in price and improving in quality and GBT4 is coming. once it's clear this thing is going to collide sometime in March we don't know exactly when everyone's like okay we're going to need a much bigger boat and this is when you know I sprinted out ahead to to to form the Bain and OpenAI alliance um which was the first of its kind in the industry and sort of changed uh you know that that changed Bain's commercial outlook for for the next uh 48 months right rebooted their their their their sales started calling up every um uh you know premium services vendor getting them ready and and just sort of saying to everyone like get yourself ready batten down the hatches because pretty soon everyone and their mom is going to want to be building with this and sure enough i mean sure enough it really did feel like that like it became like the phones didn't stop ringing there wasn't there weren't enough people around and i think um there was uh uh yeah just this the zone flooded it was it was a low pressure zone and everyone entered um competitors partners customers and what was like the
29:31Billions Host:most stressful time for you at the time? Because, I mean, you had to deal with a scale that eventually, like, you hadn't seen in the past. So how do you approach it?
29:44Zack Kass:Apart from buying a bigger boat, obviously. Yeah, I will tell a personal story here because I think this is, it would be an easy moment for me to say, well, the thing that was most stressful was the pressure that Sam and Brad had put on me. I don't think that's true. eyeing the thing that was most stressful is that i hadn't actually taken care of myself for the last i had been sprinting at startups in silicon valley for at the time 15 years and i had not taken care of myself i was very overweight i was sleeping badly i had terrible hypertension and was like basically nearing a stroke. And I had just lost all sense of habit and time.
30:31Zack Kass:And when you're in this cycle, it's easy to sort of perpetuate. I didn't even drink that much. I can't even attribute a lot of this dysfunction to addiction. But this is my reminder to the folks listening right now, you have one life and you have one body and you have one soul and one spirit and there is no cost that you are willing to pay to neglect these things. And I learned this in some ways in a beautiful way and in some ways a hard way, which is I ran so hard and so fast at this incredible moment. I probably was manic in some respect and abandoned all the things that that mattered most. Um, and that to me, that to me was the greatest source of stress that I could, that I would wake up in the morning truly with a splitting hypertension headache, you know, at 6am every morning and just start working because I couldn't, uh, you know, I, I, there was, I had not created space to actually, um, make sense of it or, or to, or to, or to be, or to treat myself well.
31:38Zack Kass:And everyone learns this. This is what I've learned. When I share this story, everyone shares a similar one. Mine happened to come at this like exceptional moment at this exceptional company, but everyone has something similar and you either learn it late in life or you learn it early in life or somewhere in between. But everyone has an epiphany or some moment in time where they go, this is actually what I need to care about. And in my case, it came at this moment where both my parents were sick and I was sick and I basically woke up one day and was like, I'm going to die. I'm going to die. I mean, I fainted.
32:17Zack Kass:I had a blood pressure off the charts. I was like, I'm going to have a stroke. I need to go home. And it wasn't even clear at the time that how much time I was going to have left with my parents. So I packed my bag, packed my car, drove my car home to Santa Barbara and I never left. I moved in with my parents at the time and i started doing advisory work um against this backdrop of of everything going incredible and rebuilt my life around the thing that mattered most to me and i'm now you know here in santa barbara with uh with my with my now wife and daughter in the other room you know two miles from my parents two miles from my sister and her husband in you know it and having comfortably said i've you know i've been i've gone through it to know that you can but
33:04Billions Host:everyone pays this price. You just, you know, when you pay it, you don't know. Congrats. I mean, it's nice to see the way you've come through. And I think you're right. Like, most important thing down the line is always like your life, your health, and the rest is, of course, interesting and important. But still, I mean, yeah, now you're advising like tons of amazing companies. So I think you did well and that was maybe the path you needed. And talking about your advisory also, I remember I think you said AI could be the last technology humans ever invent. Can you explain a bit why you were saying this?
33:52If you believe, which I think is a reasonable belief.
33:56Zack Kass:And what I, what the line is, I think technology, AI may be the last technology that humans invent on our own. And it's, it's already looking like that will be true in many respects. I mean, if, if, if, if GPT 5.2 can solve a lot of reduced problems, these, these complex open-ended math problems, what technology are we ever going to invent again on our own? I mean, it would have to be, it would be an anomaly at this point for us to discover a new technology or to have some scientific breakthrough that wasn't aided by or done by artificial intelligence at this point. And that, that feels like a reason that, that used to feel like a hot take.
34:43Zack Kass:That feels like quite a reasonable take now. And that's super exciting because that puts us on a new trajectory of, of scientific discovery. And my part of, part of the reason I wrote the book and part of the reason that I live this life and part of the values that I, that I stand for is this idea that the purpose of technology is to allow humans to live a more full life, whatever that may be. right if i like if you and i were having this conversation a thousand years ago let's say we could have it a thousand years ago but you know if we were and i would like yeah do you just try to be happy you would be like well against the backdrop of of a lot of despair that's that's pretty callous of you to say right i mean searching for happiness in a world of dysentery and and and sort of you know starvation is is much is much harder and i think that we are approaching a a future where scientific discoveries and technological automation, principally, drive down the cost of goods and services and expand what we are capable of doing.
35:45Zack Kass:And then we will have to be forced to ask ourselves, why are we actually here? And the book, The Next Renaissance, the book that just came out, is an homage to John Maynard Keynes, the father of modern macroeconomics, who wrote in 1930, The Economic Possibilities for Our Grandchildren. and in this paper he writes and i quote i must now disembarrass myself to take flight into a future that i will certainly not live to see one in which humans may have solved the economic problem and be faced with something more profound so the modern the father of modern macroeconomics is proposing that technology will eventually accelerate to the point where we can house feed and care for every person on earth and then what and then we have to figure out why we're here and for me my own journey through ai my own journey in writing this book and now my journey today actually is sort of a microcosm for much of this which is once you've once you solve for what you need then then you actually have to find something greater and i think it's this i think it's a spiritual battle
36:51Billions Host:that human and me now faces and do you feel like this is uh the case with ai like do you feel like we are actually gonna you know solve from most of the people like the basic needs so they can actually like search for a higher level of consciousness if you want to call it that way and or do you do you feel like it's actually the opposite and that ai is creating like a gap that is getting bigger and bigger between the ones who are actually using technology and the one who actually don't?
37:26Zack Kass:Well, those are two different things. And I actually think part of the problem is that we don't ground the debates. Is the debate that the floor is rising? Because that shouldn't really be a debate. The floor globally is rising all the time. So being poor in the world is way less harsh than it was to be poor a hundred years ago or a thousand years ago. And this just sort of continues to be true that the Indian economic miracle, the Chinese economic miracle. I mean, billions of people are pulled out of poverty over the last 30 years because of automation and industrial farming and industrial agriculture.
37:59Zack Kass:And you don't have to like that we slaughter 80 billion animals a day, but you have to acknowledge that in order for us to feed everyone, we needed to arrive here. And the only way we can do that is by automating these slaughterhouses and doing all these other things. And so there is a, there's a cost, but you it's very hard to argue that it is that it has never been better to be to be poor or or on earth that being said and and and by the way people are even these statements can really bother people and so i try to talk about them economically but i also appreciate that like part of the reason that no one wants to not not know it that some people don't see this as a great moment is that we are overexposed to negative information that's the other thing i remind people constantly is that part of the reason that people have lost hope in the future is not because we don't have lots of reason for it.
38:47Zack Kass:It's because we never talk about it anymore. It's very unpopular to talk about how good things are getting to the point that I get a lot of vitriol online for being aloof because I talk about things that are getting better without acknowledging all the things that are breaking. And my point is there are plenty of things that are breaking. They're just often less broken than they used to be. Now, the other problem is, are we creating a bigger gap between the rich and everyone else. And that is true. Like the amount of influence that an individual can have on elections, et cetera, continues to be a huge problem.
39:22Zack Kass:And even though we have democracies, it does feel like a small percentage of the population actually control the outcomes of those democracies. So that is true. And that sort of has been true for quite some time. And it's something we should continue to point at things like campaign finance reform, etc i don't really care how wealthy elon musk is as long as he can't buy elections right it like the number of yachts he has doesn't bother me what bothers me is that he can buy elections or control media narratives or or jeff bezos or or the coke brothers or or anyone for that matter with with that much with that much money and i would also say that like when you start to look at this when you start to peel back why it is that people believe that we aren't going to have everything we need, that we aren't going to give everyone everything we need.
40:09Zack Kass:They can recognize that we have discretionary income beyond our great grandparents' comprehension. They can recognize that we've advanced so many things so far that even the idea that we would have, you know, the thing, you and I would have the things on our desk is sort of mind blowing to most people 50 years ago. Where people get really upset is when we talk about cost of living, standard cost of living. and that principally comes back to housing, healthcare, and education. And this is where I have to remind people that housing, healthcare, and education are not expensive and they are prohibitively expensive for too many people.
40:44Zack Kass:Let's be very clear. I want to make that so clear. But they are not prohibitively expensive for too many people because we don't have the technology to make them cheap. They are prohibitively expensive for many people because we choose not to. And one of the crusades I'm on is to help people see that we are not far from a self-driving Waymo taking you across Los Angeles, maybe in a flying car for$5. But a trip to the emergency room in the ambulance five blocks away might actually bankrupt you. And you look at these problems and you suddenly realize that it is a policy failure, not a technology one.
41:24Zack Kass:that the technology exists to give people the housing need. Technology exists to provide everyone an exceptional education. The technology exists to give everyone exceptional healthcare. We just aren't doing it. We've lost our way. And every generation's greatest privilege was to pass their luxuries to the next generation as commodities, and our ruling class broke the promise. And instead of pointing at technology and saying, this is only going to make matters worse, we should point at technology and say, finally, they won't have any excuse. Finally, the policymakers won't have a single excuse for why everything is prohibitively expensive for everyone except them.
42:02Zack Kass:They can live in these mansions. They can have this exceptional healthcare. They can send their kids to the best colleges and everyone else has to pick up scraps. And finally, people can say, well, now there's a reckoning and it's coming. I mean, it's clearly coming. People like Mondani are winning exclusively on cost of living and housing. He and I don't necessarily agree with the solution, but it is clearly the problem. And when I talk about technology, when I spouse technology as a means to make the world better, my big asterisk is as long as the policymakers see it as a tool to do so, as long as we leverage unmetered intelligence to unlock these gates and people like, well, how would that happen?
42:42Zack Kass:I'm like, well, look, robotics are going to drive down the cost of construction. They're going to drive down the cost to actually build. We've already seen this, right? You can pre-fab homes are like, I think like$100 a square foot now. So why don't we put them on every corner where we need to be built? Well, because permitting takes three years. Why did we stop building universities in the United States? And I went to Berkeley, acceptance rate was 10%. Today it's like 3%. I would not have gotten in today. Why? Well, we stopped building seats for students and we stopped building great American universities despite the fact that the United States kept growing.
43:17Zack Kass:so now we have a problem now we have to figure out how we how we can offer this to everyone and technology affords us that and and you and you look at the problem like you know the the health care problem it's like well what's going on here well we require that everyone come to a box we require that everyone pay for overpriced sort of like grossly um uh uh overwrought care when actually the problem sits so further upstream we just don't let companies actually provide good preventative care unless they're in the box. And this is why you see companies like Function Health taking off because people are finally like, oh, I can figure out or Pronovo.
43:53Zack Kass:Oh, I can finally figure out if there's something wrong with me and then I can actually start to solve my problems much further upstream. And even by the way, breakthroughs like GLP-1s are going to save hundreds of millions of lives, right? We are just, we are discounting what will happen to a population when it is no longer obese. And that is technology. Like we get to celebrate so many of these things, let's not forget that so much of our lives continue to get better, not worse, because of technology. And that this is probably not going to be an exception. As long as we put pressure on policymakers to do the right things, and as long as we avoid our sort of traditional traps, the addiction to the device, the addiction to gambling or porn or social media, these are the things that we can avoid to actually discover a panacea where everyone has what they need and can spend time with friends and family.
44:43Zack Kass:I mean, this is well within our grasp.
44:46Billions Host:And since you talk about policy and a bit politics, I'm curious to have your take on AI from a global standpoint. So typically, at first, I think the US were extremely dominant when it came to AI, and they are still definitely. But we're seeing models like DeepSeek being also like really strong competitors. So how do you see like the world and the different like models globally?
45:22Zack Kass:So if AI is a race, so first I think that the open source pressure to drive down cost has been a great thing. Like I think DeepSeek, I celebrated DeepSeek in January last year and I would celebrate it again and again. Like it is, regardless of how it was trained, we should always celebrate when a commodity gets less expensive and we should celebrate the people who make it so. Driving down the cost of inference is such an important task that will save a bunch of energy in the future and also guarantee access. I mean, you can run GBT 4.5 locally on your device now. That's thanks to all the work that the model compression and hardware updates have done.
46:18Zack Kass:If you, on a geopolitical basis, we talk about this in the book, is this North Star problem. Not everyone believes in the same sense of alignment. Not everyone wants the same thing with superintelligence. But my argument is that actually, if this is a race, it is a prize that will be shared equally. and whereas there are other prizes that are harder to be shared equally super intelligence leads to an exceptional new understanding of the known universe which almost certainly unlocks a bunch of shared value and what isn't necessarily shared is things like energy supply you know or or or resources so i sort of point at the ai race is actually something that like we all like i don't care if China comes up with a cure for cancer.
47:05Zack Kass:And I don't think you do either. I don't know many Americans who - As long as it's cured. I want these problems solved. What I think we need to do in the United States, if we acknowledge that we have political adversaries, if we acknowledge that we have a North Star problem, what the United States wants or sees as a panacea is not necessarily what other nations see as a panacea, then what we need to do is actually reinforce ourselves with other infrastructure, energy principally. We are losing the race to create more energy, natural resources, education, healthcare, housing, the tenants of a society that are good, that produce a really strong sense of national pride.
47:46Zack Kass:These are the things that we really need to focus on investing in so that we can actually make the most use of AGI when it arrives.
47:58Billions Host:When do you think it will arrive?
48:00Zack Kass:I think AGI is already here. I think we're sort of at AGI. But again, AGI is such a marketing term. There's no shared definition. So we're probably at AGI. We're a ways off from super intelligence, but you could argue we're at AGI.
48:19Billions Host:What's one problem you really want to see being solved and you feel like we are very close from solving it?
48:28Zack Kass:well when people are like why should i care about ai you know i get a lot of vitriol uh online from people who are like it's ai slop your ai slop this is all destroying the earth and i can remind them that we discovered our first antibiotic in 60 years recently because of ai i can remind them we split hiv out of dna recently because of ai i can remind them that baby KJ received the first custom gene therapy thanks to CRISPR and AI last year in May. And, but what I think actually has to happen is AI has to solve such a big pressing problem that no one can call it irrelevant. And I think it's either going to be something like fusion energy, where people actually see what unmetered energy looks like, or quantum computing, or ideally something even more tangible like cancer.
49:18Zack Kass:Like we cure a cancer with AI and the heads of the oncology research lab go on TV and say, we couldn't have done this without this technology. And then people realize it's not the app. It's not chat GBT that they are yelling at. It's actually unmetered intelligence. It's something way, way, way more profound.
49:37Billions Host:Okay. And in your book called The Next Renaissance, if you compare it to the… sounds way better it sounds way better when you say I can say the French way Renaissance so in the original like Renaissance the people who controlled like printing press like they got extremely rich and in the AI Renaissance like who do you think will win
50:08Zack Kass:I think everyone who so this is going to be a proliferation and diffusion of, of intelligence. And if you are, if your business gains when intelligence gets less expensive, then you're going to do really well. And for many people, this is novel science research. This is, um, you know, this is businesses that would do really well if they better understood the known universe. And, uh, and then businesses that will do less well are businesses that are used to sort of hoarding intelligence.
50:41Billions Host:And we'll sort of see how that shakes out. So when you look at different industries, where do you feel like there's going to be the biggest shift, if you had an example? Life sciences, healthcare, education.
50:55Zack Kass:I mean, it's just not close. We are going to see a renaissance of education over the next 10 years that is going to fundamentally change how parents, what parents expect the system to provide, how children are raised. And we are going to probably see the end of the industrial education system that we have practiced for 150 years at the cost of the soul of every child, right? Sitting for eight hours a day in a classroom, staring at a chalkboard is not what children are supposed to do. and we've we forgot we've totally forgotten and there are lots of solutions we write about tons of them and um and we are we are very close to i think a really magical moment in how we actually um and how we actually educate the next generation and to say nothing of how we how we care for them
51:51Billions Host:you have like a young daughter so in a in an ideal world how would you like her to to learn new things.
51:59Zack Kass:I want only for my daughter to love her friends and family and love nature. And my belief is if my wife and I can instill the importance of those two things, some of which are genetically predisposed and some of which are learned, then she will do really well. And my belief is that she is going to have tools and technology beyond our comprehension to explore the world. And I just want to make sure she cares about the right things and that she believes that she can't, that she has agency and personal responsibility to explore beyond our imagination today, a world that I think is going to continue to unfold in front of us.
52:37Billions Host:We're almost running out of time. So before saying goodbye, where can people follow you and where can people buy your book?
52:46Zack Kass:My website, ZachCast.com. I have a sub stack. That's where I really like to engage with people. I send weekly to bi-weekly updates on everything that's happening in AI my book The Next Renaissance AI and the Expansion of Human Potential I'm very proud of that and I have an Instagram the team will post clips from me social media makes me anxious so I'm not on there but it's the best way I know to engage with everyone who doesn't want to necessarily read my sub stack
53:12Billions Host:Awesome, thanks a lot Zach have an amazing day Thanks Gil
53:22All right.
From the publisher
Today on BILLIONS, I'm sitting down with the man who had the hardest sales job in Silicon Valley history: Zack Kass.
Before ChatGPT was a household name, it was just a research lab.
Zack was Head of Go-To-Market. He joined when OpenAI was around a hundred people doing two million in revenue.
His job? Sell human-level intelligence to Fortune 500 executives who didn't even know what a token was.
He built the playbook for Microsoft. For Coca-Cola. He turned a nonprofit lab into an eighty billion dollar superpower.
Then he walked away.
Zack, thanks a lot for being here!
TIMELINE
00:00:00 - 00:06:34 : Joining OpenAI as the first sales person
00:06:34 - 00:09:47 : The early GPT-3 wrapper ecosystem
00:09:47 - 00:11:03 : Strategy behind ChatGPT's development
00:11:03 - 00:16:45 : The chat interface decision and market response
00:16:45 - 00:21:04 : ChatGPT's explosive growth and company atmosphere
00:21:04 - 00:26:45 : Lessons from viral growth and Microsoft partnership
00:26:45 - 00:30:21 : Scaling challenges and the "bigger boat" moment
00:30:21 - 00:33:44 : Personal burnout and health crisis
00:33:44 - 00:37:29 : AI as humanity's last invention
00:37:29 - 00:45:21 : Technology, inequality, and policy failures
00:45:21 - 00:48:35 : Global AI competition and geopolitics
00:48:35 - 00:50:18 : AI's potential to solve major problems
00:50:18 - 00:54:33 : The next renaissance and education transformation
REFERENCES
- "The Next Renaissance: AI and the Expansion of Human Potential" by Zach Cass
- Lilt
- Scale AI
- Jasper
- Harvey
- Waymo
- Prenuvo
- DeepSeek
- OpenAI
- RLHF
- CRISPR
- Jaws



