<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=2006193252832260&amp;ev=PageView&amp;noscript=1">

56 Min Read

[The AI Show Episode 89]: A New In-Depth Sam Altman Interview, The “Inflection Point” for Enterprise Generative AI Adoption, and Inflection AI’s Big Shakeup

Featured Image

Wondering how to get started with AI? Take our on-demand Piloting AI for Marketers Series.

Learn More

In this week's episode of The Artificial Intelligence Show, our hosts discuss some tough news for generative AI companies from this past week. Paul Roetzer and Mike Kaput break down the biggest happenings and events, including Sam Altman's latest interview with Lex Fridman, Andreessen Horowitz's research on generative AI in the enterprise, major changes occurring at Inflection AI, and more.

Listen or watch below—and see below for show notes and the transcript.

Listen Now

Watch the Video

 

 

Timestamps

00:03:47 —  Sam Altman Interview with Lex Fridman

00:17:33 — Changes to the Way Enterprises Are Building and Buying AI

00:28:32 —Big Changes at Inflection

00:36:04 — Stability AI Might Be Crashing

00:40:14 —  Tepid Revenue at Cohere Shows OpenAI Competitors Face Uphill Battle

00:45:09 — Anthropic, AWS, and Accenture team up

00:48:57 — Claude 3 Opus + Prompt Library

00:56:23 — Gemini 1.5 API + Google AI Studio

01:03:09 — Chrome AI Rollout

01:06:53 — The Story Behind How 8 Google Employees Invented Modern AI 

Summary

Sam Altman’s interview with Lex Fridman

Sam Altman just gave another in-depth interview on the Lex Fridman podcast that covered everything from the OpenAI board saga to Elon Musk’s lawsuit to GPT-5.

The interview clocks in at nearly 2 hours long—and contains some very interesting comments from Altman that are worth unpacking.

Altman revealed some personal insights, describing his firing as "shockingly painful" and how it changed his trusting nature. He also mentioned potential safety concerns, alluding to the possibility of threats against him.

Altman also downplayed the current GPT-4, saying it will seem obsolete compared to GPT-5, which is expected to have major advances in reasoning capabilities, something that has been a key focus for OpenAI.

When asked about Elon Musk's lawsuit, Altman confirmed Musk wanted to merge OpenAI with Tesla, leading to the split, which Altman finds disappointing given his respect for Musk.

Generative AI in the enterprise inflection point

Famed venture capital firm Andreessen Horowitz just published new research that suggests we’re at “an inflection point” when it comes to generative AI in the enterprise.

In fact, they say that while generative AI took the consumer landscape by storm in 2023, reaching over a billion dollars of consumer spend in record time…

In 2024, they believe the revenue opportunity will be multiples larger in the enterprise. In the research, a16z, as the firm is nicknamed, spoke with dozens of Fortune 500 and top enterprise leaders, and surveyed 70 more, to understand how they’re using, buying, and budgeting for generative AI.

According to the firm: “We were shocked by how significantly the resourcing and attitudes towards genAI had changed over the last 6 months.”

It turns out these enterprise leaders, despite still having some reservations about genAI deployment, are overall planning to triple their budgets.

In terms of model providers, OpenAI and Google are dominating enterprise adoption so far, with smaller players like Anthropic and open-source models like LLamA lagging behind. This make it challenging for startups to compete with big tech companies that have proprietary data, computing power and the ability to attract top AI talent.      

Big Changes are Happening at Inflection AI

We’ve got some major AI news we’ll probably be hearing more about in coming weeks…

Mustafa Suleyman, co-founder and CEO of Inflection AI (and former co-founder of Google DeepMind), is leaving Inflection to start Microsoft AI, a new division at Microsoft that will bring together their consumer AI efforts, as well as Copilot, Bing and Edge.

Microsoft also appears to have hired most of Inflection’s staff as part of the move.

Inflection as a company will actually continue with a new CEO, but without its founder and much of its talent.

The company said in a blog post that its plan moving forward is: “...to lean into our AI studio business, where custom generative AI models are crafted, tested and fine tuned for commercial customers.”

As we unpack this event, it’s helpful to remember some potentially relevant context: Inflection was (is?) a giant in AI. It had raised $1.5 billion from investors that included Microsoft.

Inflection’s main consumer-facing product was Pi, which it billed as the first emotionally intelligent AI assistant. Reid Hoffman, of LinkedIn fame, was a co-founder of Inflection and sits on Microsoft’s board.

Links Referenced in the Show

This week’s episode is brought to you by our Marketing AI Conference (MAICON).

From September 10-12 this year, we’re excited to host our 5th annual MAICON at this pivotal point for our industry.

MAICON was created for marketing leaders and practitioners seeking to drive the next frontier of digital marketing transformation within their organizations. At MAICON, you’ll learn from top AI and marketing experts, while connecting with a passionate, motivated community of forward-thinking professionals.

Now is the best time to get your MAICON ticket. Ticket prices go up after Friday, March 22. Visit www.maicon.ai to learn more.

Read the Transcription

Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.

[00:00:00] Paul Roetzer: We come to trust these AI tools, these AI companies, but what happens to data? What happens when the company goes away and you don't have that tool or they do just get acquired and they shut down and sunset the product. again, it's just a bit of a cautionary tale of this is a very dynamic space we're living through and it's just starting,

[00:00:18] Paul Roetzer: Welcome to the Artificial Intelligence Show, the podcast that helps your business grow smarter by making AI approachable and actionable. My name is Paul Roetzer. I'm the founder and CEO of Marketing AI Institute, and I'm your host. Each week, I'm joined by my co host, and Marketing AI Institute Chief Content Officer, Mike Kaput, as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career.

[00:00:48] Paul Roetzer: Join us as we accelerate AI literacy for all.

[00:00:55] Paul Roetzer: Welcome to Episode 89 of the Artificial Intelligence Show. I'm your host, Paul Roetzer, [00:01:00] along with my co host, Mike kaput.

[00:01:02] Paul Roetzer: we are both in cleveland. this week, right?

[00:01:06] Mike Kaput: We're back in Cleveland

[00:01:08] Paul Roetzer: are in Cleveland. We're recording this

[00:01:10] Paul Roetzer: a little early. It is Friday, March 22nd at 1 p. m. Eastern time. we usually record on Monday mornings. But, I've got a talk Monday. It was going to eat up of my day. So we're gonna, we're doing this on Friday, but it worked out well. Because

[00:01:25] Paul Roetzer: it was a really busy week in, in terms of like large language models and, lots of news to cover in that. we've got an interview with Sam Altman. We've got some changes at Inflection and Microsoft, bunch of stuff from Google.

[00:01:37] Paul Roetzer: So it just really worked out where we plenty to go on by Friday. So. It was okay to cut it a day short or an afternoon

[00:01:45] Paul Roetzer: So hopefully nothing crazy happens Friday afternoon or Monday morning. but like I we've got plenty to get through already.

[00:01:52] Paul Roetzer: So this episode brought to us by the Marketing AI Conference, or MAICON.

[00:01:57] Paul Roetzer: It'll be September 10th to 12th this [00:02:00] year. This is our 5th annual MAICON, so it's hosted by Marketing AI Institute. MAICON was was created back in 2019

[00:02:07] Paul Roetzer: for marketing leaders practitioners seeking to drive the next frontier of digital

[00:02:12] Paul Roetzer: marketing transformation within their organizations and in their own careers.

[00:02:17] Paul Roetzer: At MAICON, you'll learn top AI and marketing experts while connecting with passionate and motivated community of forward thinking professionals. The 2023 event drew more than 700 people to Cleveland, Ohio, the convention center in Cleveland, right across from the Rock and Roll hall of Fame. And we're expecting,

[00:02:34] Paul Roetzer: probably 1, 500 or more this year is of, we're, we're trending above, but, um, that's the goal was 1500 and right now it's looking pretty good.

[00:02:43] Paul Roetzer: And, inside track, we will actually. all goes well, be announcing the vast majority of the agenda, um, probably sometime in the next, 10 to

[00:02:55] Paul Roetzer: so days. I don't know, I don't want to put Tracy on the spot, but, I've spent the last, [00:03:00] personally spent the last three days this week going

[00:03:02] Paul Roetzer: through over 150 submissions for speakers, in addition to my own sandbox of about 50 more speakers.

[00:03:09] Paul Roetzer: And we, um, we are at the point where we're, we're gonna be able to announce the vast majority of the the main stage. We won't announce all those. There's some stuff in the works there, but we are going to announce, a, a good portion

[00:03:21] Paul Roetzer: of the speakers. So stay tuned for that. You can learn more at MAICON.ai

[00:03:26] Paul Roetzer: MAICON.ai And again, that's coming up September 10th to the 12th in Cleveland, Ohio. we would love to have you be a part

[00:03:36] Paul Roetzer: of it and join Mike and I and the rest of the Institute team there.

[00:03:40] Paul Roetzer: Okay, Mike, I know you and I both listened to the Lex fridman interview with Sam Altman. So seems like a pretty good place to start.

[00:03:47] Sam Altman Interview with Lex Fridman

[00:03:47] Mike Kaput: Alright, so, Sam Altman,

[00:03:49] Mike Kaput: like mentioned, just gave another in depth interview, and this time it was on the Lex Fridman Podcast, and They covered everything from the [00:04:00] OpenAI board saga, back when Sam got briefly fired, to Elon Musk's more current lawsuit, to GPT-5. Now, interview clocks in

[00:04:10] Mike Kaput: nearly two hours long, and contains some very interesting comments from Altman that we think are worth unpacking.

[00:04:19] Mike Kaput: So, paul, what jumped out to you most during the interview?

[00:04:24] Paul Roetzer: it's honestly like hard to narrow this down. It's, I mean, if you

[00:04:27] Paul Roetzer: look at the timestamps, I would definitely recommend people go listen to I think that. it's just so important. You know, I've said it many times in the show,

[00:04:35] Paul Roetzer: one of the ways I learn the most about AI is by listening to the people leading these research firms. And SAM is certainly

[00:04:43] Paul Roetzer: no and I think it's important that people understand not only

[00:04:47] Paul Roetzer: what they're trying to build, like follow the technological aspects of it, but you, you have to understand the human aspects of this. And this was a,

[00:04:54] Paul Roetzer: you know, a pretty raw interview with Sam,

[00:04:58] Paul Roetzer: I say. Like, it started [00:05:00] right off the first, like, ten minutes with his firing.

[00:05:02] Paul Roetzer: And you can tell that there are significant scars, sent for Sam

[00:05:08] Paul Roetzer: mentally. He used the phrase, shockingly painful, multiple times to explain what happened, and the thing I found kind of sad is he mentioned, I don't know if you caught this Mike,

[00:05:20] Paul Roetzer: like he mentioned how he was a very trusting person and that this sort of like changed him.

[00:05:26] Paul Roetzer: Like his default used be just to trust everyone and you can tell that that is just not

[00:05:32] Paul Roetzer: case. So, you know, I think moving forward just to understand his, his personal

[00:05:39] Paul Roetzer: mental state, his personal desires, he mentioned at one point, I mean, I almost said like, Do a double take on this one. He said it in passing and Lex almost like glossed over and then came back

[00:05:49] Paul Roetzer: to it. But he's like, yeah, I might get shot at some basically. Like that was so weird. Like, but you understood he was saying that he was talking about the significance of what

[00:05:59] Paul Roetzer: [00:06:00] doing and how there are a lot of people who don't like necessarily what OpenAI is doing, and they see him as sort of the figurehead of it.

[00:06:09] Paul Roetzer: And so he was basically saying like, yeah, I don't know, man. Like my life is just. Bizarre. so

[00:06:15] Paul Roetzer: it it was just interesting from a personal perspective.

[00:06:18] Paul Roetzer: Anyone who's listened to this show knows, Ilya Sutskever. We talked about Ilya a lot, one of the,

[00:06:23] Paul Roetzer: you know, prominent figures in modern AI and was, on the board the time sam was fired.

[00:06:29] Paul Roetzer: And so, you know, to Lex, to his credit, asked all the hard questions and he said.

[00:06:34] Paul Roetzer: You know, what's going on with Ilya, and Sam said he doesn't know what his plans are.

[00:06:39] Paul Roetzer: You'd have to ask him, but he hopes that they get work together for the rest of their career. he asked specifically what

[00:06:47] Paul Roetzer: Ilya see, and Sam said Ilya did not see AGI.

[00:06:50] Paul Roetzer: None of us have seen AGI. And there was a lot more about Ilya. So that was one of the main characters in the interview. The other one that was, you know,

[00:06:58] Paul Roetzer: Very prominent was [00:07:00] Elon musk, and again, on the personal side, you tell Sam is just disappointed in how

[00:07:07] Paul Roetzer: this has played like Elon is someone that Sam has up to his whole career. He sees him as a very important person in human history and he just doesn't understand. His literal quote was, it makes me sad. Like Lex said,

[00:07:22] Paul Roetzer: so the podcast, Elon chose to leave. Okay. So, he asked, Lex, who is friends with asked Sam about, you know, the whole lawsuit and, why it happened.

[00:07:35] Paul Roetzer: And Altman confirmed exactly what we said on the podcast, was that Elon chose leave OpenAI.

[00:07:41] Paul Roetzer: Due to his own desires for AGI and that he wanted to merge OpenAI with Tesla, which exactly what we had, you know, cited from the New York Times

[00:07:47] Paul Roetzer: article. and then sam said, like, it just makes me sad. I miss the old Elon. and then later on in the interview, toward the end, he said, the amazing about Elon is amazing and I super [00:08:00] respect him.

[00:08:01] Paul Roetzer: I think we need him. All of us should be rooting him and need him to step up as a leader through this next phase. So.

[00:08:09] Paul Roetzer: I don't know. I found, I found that, kind fascinating.

[00:08:12] Paul Roetzer: A couple of the other ones that jumped out to me, and then Mike, I'd be interested to see if there's any

[00:08:16] Paul Roetzer: other ones that caught your attention.

[00:08:18] Paul Roetzer: He did ask about fair use and copyright, which we talked about last week on the show with Mira, the CTO. And Sora in

[00:08:25] Paul Roetzer: particular, lex said, do you think training AI be or is fair use under copyright law? And Sam completely dodged it. He said, I think question behind that question is, do people

[00:08:36] Paul Roetzer: create valuable, who create valuable data deserve to have some way they get compensated for it? And that I think

[00:08:43] Paul Roetzer: answer is yes. And he kind of says it's sort of hard.

[00:08:46] Paul Roetzer: And then Lex specifically asks about Sora. And says, but artists and creators are worried they see Sora, they're like, holy shit.

[00:08:53] Paul Roetzer: And Sam said, sure, artists were also super worried about photography, when it came out. And then [00:09:00] photography became a new art form and people made a lot of money taking pictures.

[00:09:03] Paul Roetzer: I think things that will, like that will

[00:09:05] Paul Roetzer: keep happening. People will use the new tools in new ways.

[00:09:08] Paul Roetzer: On AI and and jobs, sort on that path, obviously the last two episodes of this podcast, we've talked a lot about Sam's quote,

[00:09:16] Paul Roetzer: about 95 percent of what creatives do, you know, and so lex asked him about jobs and it was an interesting answer, and it does

[00:09:23] Paul Roetzer: align with what he's said previously. He said, people talk about how many jobs is AI going to. To do in five years, the framework that people have is what percentage

[00:09:33] Paul Roetzer: current jobs going to be totally replaced by some AI doing the job.

[00:09:37] Paul Roetzer: The way I think about it is not what percent of jobs AI will do, but what percent of tasks will AI do over a time horizon?

[00:09:45] Paul Roetzer: So you kind of like talk more about the level and then that sort of spun into the Sora conversation. And the one that

[00:09:52] Paul Roetzer: like, I was fascinated about here is. Alex asked him about this idea of is Sora actually representing [00:10:00] the physical world in some way? Is it understanding the physical world?

[00:10:03] Paul Roetzer: and so he asked like, you know, about its ability to represent and understand like the 3D model of the world and asked whether

[00:10:12] Paul Roetzer: or not like current path OpenAI is taking could get to the point where it does seem to really do that and I thought Sam,

[00:10:20] Paul Roetzer: you know, he, he was very thoughtful in his response, but he said, I think this approach is going to go surprisingly far. So again, like we've talked many times, not all the leading AI researchers agree with other. Like Yann LeCun

[00:10:32] Paul Roetzer: LeCun would not. agree with this. Like he does not think this is the path, but Sam seems fairly confident. GPT GPT

[00:10:41] Paul Roetzer: 4, think you had made note

[00:10:43] Paul Roetzer: of this one, like, um, lex said, for me, looking back, GPT-4, ChatGPT pretty damn impressive, historically impressive. and Sam said,

[00:10:52] Paul Roetzer: I think it kind of sucks. So, know, I think the key there is he's, Sam is three to five [00:11:00] years out on he is fairly confident will be happening.

[00:11:03] Paul Roetzer: And then he looks at the current stuff and says, this is obsolete. Like what we have. Yeah, sure. Okay. You might

[00:11:08] Paul Roetzer: impressed by it, but we won't be in time. and then that led GPT-5 5 conversation, which Sam was like, I don't even know what we're going to call it, but like, call it

[00:11:19] Paul Roetzer: GPT-5 this. Like, when's it coming out? And sam said, You know, we're not like given

[00:11:25] Paul Roetzer: a timeline, but I think we had talked previously, I was at episode 87, Mike, the timeline, like May to july, which seems viable, but then he talked about, how

[00:11:34] Paul Roetzer: they're going to be releasing a lot of things in the process.

[00:11:37] Paul Roetzer: Like there's going to be pieces they're going to kind of release that won't be

[00:11:40] Paul Roetzer: necessarily GPT-5, but of build up to it.

[00:11:42] Paul Roetzer: That led the Q* conversation. So if people have followed along on the podcast

[00:11:48] Paul Roetzer: since last November, When Sam got fired, there was this belief that they had had this scientific breakthrough they were calling Q Star.

[00:11:55] Paul Roetzer: So episode 74 from November 28th is when we talked about that in [00:12:00] detail. But this, I thought, was maybe the most interesting part of the whole interview.

[00:12:04] Paul Roetzer: Because lech said, this does make me think about the mysterious behind Q*. What's this Q* project?

[00:12:11] Paul Roetzer: and then a few things about nuclear facilities, secret facilities, and some jokes kind of happen.

[00:12:17] Paul Roetzer: And then Lex comes back and says, can you speak what Q* is? And Sam very abruptly said, We are not ready to talk about that. And Lex of Pushed them a little bit. He said, yeah, but that's an answer,

[00:12:29] Paul Roetzer: or, but an answer like that means there's something to talk about. It's very mysterious, Sam.

[00:12:33] Paul Roetzer: And Sam said, I mean, we work on all kinds of research. have said for

[00:12:37] Paul Roetzer: a while that we think better reasoning in these systems is important direction that we'd like to pursue. We haven't cracked the code yet. We're very interested in it. And that leads to

[00:12:49] Paul Roetzer: What Q* believed to be is some mathematical, some ability for it to do mathematics in new ways that enhances its reasoning.

[00:12:58] Paul Roetzer: now, why is [00:13:00] reasoning important? Like, why does that matter? So reasoning is basically like the cognitive process of of our ability to think in a logical way, form a conclusion, a judgment.

[00:13:09] Paul Roetzer: So, reasoning is fundamental to our human intelligence. It lets us analyze situations, figure out new information, apply knowledge under contexts. So if

[00:13:19] Paul Roetzer: you at all the things that enables, like in the scientific process, our ability to a hypothesis about something is because of reasoning. Our problem solving,

[00:13:27] Paul Roetzer: ability to reflect on our past experiences and thoughts, like all of that

[00:13:31] Paul Roetzer: is reasoning and that is fundamental. to system 2 thinking, which like the slow thinking. We're going to talk more about that throughout. today's

[00:13:38] Paul Roetzer: episode. this idea of giving these models time to think and analyze and reflect. And so I think that's really so I

[00:13:47] Paul Roetzer: mean, I thought language models and search was interesting. Regulatory capture conversation interesting. Are there any other ones jumped out to you, mike, that I didn't touch on already? Million that Gemini 1. 5 is enabling.

[00:13:56] Mike Kaput: Instead of

[00:13:56] Mike Kaput: the million that we have that already seems crazy.

[00:13:59] Mike Kaput: [00:14:00] So very far flung speculation, but I think it's worth taking seriously that it's not on that big a timeline. So we're talking like five to 10 right, Max? Yeah, and I

[00:14:09] Paul Roetzer: I think he has a very, very strong history of being directionally correct in what he thinks the world will look like and

[00:14:16] Paul Roetzer: timelines that will happen. Like, the quote I often lead my keynotes off with is from his moore's Law for Everything blog post from March of 2021, in which says, the coming will focus you know, focus on our most impressive, our

[00:14:29] Paul Roetzer: human abilities think, to reason, to understand language. And people didn't listen at that time. So March 2020 gave us a year and a half

[00:14:37] Paul Roetzer: start, for when they released ChatGPT and most business leaders weren't paying attention. And I think this

[00:14:44] Paul Roetzer: a similar moment. Like they're telling GPT-4 will look impressive in retrospect. It will not seem like a thing. And keep in

[00:14:52] Paul Roetzer: mind, they didn't release GPT 2 in 2021 because of concerns of [00:15:00] misuse and all the things that could go wrong. So back in 2021, GPT 2

[00:15:05] Paul Roetzer: was so advanced for society that they didn't release it right away.

[00:15:09] Paul Roetzer: And he's already saying GPT-4, most people are still trying to understand and apply to their business, that it will feel equally as obsolete. he said,

[00:15:19] Paul Roetzer: you know, I think the leap from four to five will feel similar to three to four, it it was like, holy cow, us, the outsiders, it was mind blowing.

[00:15:29] Paul Roetzer: He is thinking that that will be the case, that we will see five and think, wow, we had no idea it was going to be capable of these things.

[00:15:37] Mike Kaput: And kind of the last point that you hit on perfectly, but I just think it's just so weird and just has my Spidey sense tingling, is like,

[00:15:47] Mike Kaput: there are memes going around that have Ilya on the side of a milk carton, like where is he in all of this? Like, the Q Star stuff, which was highly related

[00:15:58] Mike Kaput: your point, to what you said. He was one of the creators

[00:15:59] Paul Roetzer: [00:16:00] of it, supposedly.

[00:16:00] Mike Kaput: It's very easy, obviously, to fall into theories,

[00:16:04] Mike Kaput: which I'm not gonna get into, but it is so fascinating, that human element of there is something really important about whatever went down there, and the silence is deafening, I would say.

[00:16:15] Paul Roetzer: I agree. I, as we said back in November, there's something still much.

[00:16:21] Paul Roetzer: more to this story. And I do think that as they start releasing things throughout this year,

[00:16:26] Paul Roetzer: we will start understand a little bit more, but I do think that this system to thinking the reasoning, and there's some other topics we'll cover today

[00:16:35] Paul Roetzer: that sort of further validate that this is likely going to to be very important context to have. So again, I think as a

[00:16:42] Paul Roetzer: listener to this show, if you listen regularly, you're going to hear some recurring themes. And I just, I mean, like back in 2022, I,

[00:16:51] Paul Roetzer: I interviewed, interviewed Vedant Mishra at, at mAICON. I've mentioned Vedant recently, and he's working on reasoning at DeepMind.

[00:16:59] Paul Roetzer: And back in [00:17:00] when I interviewed him, he was talking about the coming advancements in reasoning capabilities and what that would unlock

[00:17:06] Paul Roetzer: these models. So there, I mean, many of the smartest AI researchers in the world are, are working on variations of whatever Q* is and whatever, kind of these advancements are.

[00:17:19] Paul Roetzer: It

[00:17:19] Paul Roetzer: It was even a hot topic this week with, um, Andrew Ng, Yann LeCun, other people from the FAIR research lab,

[00:17:27] Paul Roetzer: It's directionally correct. Whatever happens next is going to be in things

[00:17:31] Paul Roetzer: similar to this.

[00:17:33] Changes to the Way Enterprises Are Building and Buying AI

[00:17:33] Mike Kaput: All right. So in our second big topic this week, the famed venture capital firm Andreessen Horowitz just published some new research that

[00:17:42] Mike Kaput: we're quote, at an point when generative AI in the enterprise. They say that while generative AI took. The consumer landscape by storm in 2023, it actually over

[00:17:56] Mike Kaput: a billion dollars of consumer spend in record time. [00:18:00] In 2024, the revenue opportunity will be multiples larger in the enterprise. So in this research, A16Z, which is kind of their nickname, I'll refer to him as that, spoke with dozens of fortune 500 and top enterprise leaders, and then surveyed 70 more of them to understand how they're using buying and budgeting for Gen AI.

[00:18:23] Mike Kaput: They said, quote, we were shocked by how significantly the resourcing and attitudes towards Gen AI had changed over the last six months. So it turns out that these enterprise leaders, at least the ones they talked to, despite still having reservations about Gen AI deployment are overall planning to triple their budgets with Gen AI.

[00:18:45] Mike Kaput: Now this full Rundown is really well worth reading, but a couple highlights, Paul jumped out at me. They said that nearly every single enterprise we spoke with saw promising early results of Gen AI experiments and plan to [00:19:00] increase their spend anywhere from 2x to

[00:19:01] Mike Kaput: 5x in 2024 to support deploying more workloads to production.

[00:19:07] Mike Kaput: This budget also appears to be shifting from innovation budgets to more permanent software line items. They also said that enterprise leaders are currently mostly measuring ROI by increased productivity generated by AI. Most of them currently do not have the in house technical talent to implement and scale gen AI.

[00:19:28] Mike Kaput: They're spending a ton on implementation and foundation model providers are kind of filling that gap by providing professional services. And then when it comes to which model and model providers people are going with, if I'm interpreting their. Chart correctly, 100 percent of the people I surveyed are either testing OpenAI's models or have them in production, 66 percent say they have them in production, 34 percent are testing.

[00:19:55] Mike Kaput: The next most popular models being used come from Google, which 63 percent [00:20:00] are either testing or have in production. Llama

[00:20:04] Mike Kaput: is the third highest one, they have six total, Anthropic is fourth, and then Mistral and Cohere are way, way at the back of the pack. So it sounds like Also, based on this research, enterprises are overwhelmingly focused on building in house right now versus buying, and they're still quite cautious about external facing use cases, given issues they're trying to overcome with hallucinations and PR.

[00:20:30] Mike Kaput: possible disasters, which makes sense. The most popular use cases that these

[00:20:35] Mike Kaput: reported were either focused internal productivity right now or stuff that gets routed through a human, which marketing falls into. Stuff that a human marketer or business person is taking a look at, making sure it's ready for primetime.

[00:20:50] Mike Kaput: So I'll stop there. Paul, you have Tons of on the ground insight into enterprises in the work you're doing with them, your talks, your [00:21:00] conversations you're having basically daily with enterprise leaders. Like, how accurate do you find A16Z's assessment here?

[00:21:08] Paul Roetzer: Yeah, I thought

[00:21:09] Paul Roetzer: it was, it was really good. I mean, I really like these reports that are just hitting.

[00:21:13] Paul Roetzer: Like they have what, there's like 16, I think, findings and just like a paragraph or two about each, like a quick read. I definitely recommend people go check out the full thing.

[00:21:22] Paul Roetzer: The budget numbers really jumped out to me. I mean, those are not insignificant numbers

[00:21:27] Paul Roetzer: they were throwing around. Uh, uh, increase of to five X, tripling of budget.

[00:21:32] Paul Roetzer: It's like that's,

[00:21:34] Paul Roetzer: That's wild, but it makes sense as you move from the innovation budget to the operations budget, that this would happen. So I

[00:21:40] Paul Roetzer: wouldn't say I've heard these kinds of numbers being thrown around, that those were a bit eye opening, but everything else they said definitely jives what we're hearing.

[00:21:49] Paul Roetzer: I would imagine they talked to a lot of people. tech companies. But you know, when we talk to Fortune 500 companies, sit down with their leaders, like these are similar things we're hearing, you know, it's that the obvious [00:22:00] ROI play

[00:22:00] Paul Roetzer: is efficiency and productivity, but people want to kind of get beyond that and out ways to do

[00:22:06] Paul Roetzer: it. And what we always advise companies is you have to benchmark performance now and

[00:22:10] Paul Roetzer: then you have have KPIs that you'll measure to determine if what you're doing is working. So whether this is a pilot project with a large language model. You know, if you're just applying it to email writing or article writing or video script writing or podcasting, whatever it is,

[00:22:24] Paul Roetzer: like, whatever you're applying it to know how you're going to measure

[00:22:27] Paul Roetzer: existing performance and then how you're going to measure against and what success looks like.

[00:22:31] Paul Roetzer: I thought the multi model future definitely jives with what we're hearing.

[00:22:35] Paul Roetzer: Yeah. so they're all testing a of models and may actually scale with collection of models because some are, and we'll talk a

[00:22:43] Paul Roetzer: little bit about this with your experience with Claude in a minute. some are just better at other, others than at certain use cases.

[00:22:50] Paul Roetzer: So like, we have ChatGPT, we have we have Gemini internally. And I don't know that I would bet on

[00:22:56] Paul Roetzer: any one of them. Like sometimes the other one is just better at a [00:23:00] use case. And if that use case is critical to us, I don't want to lock us in say, Hey, Mike, sorry, you have

[00:23:04] Paul Roetzer: to chatGPT because I'm not going to pay the other 20 bucks a month for you to have a Claude it's like, no, right. Like we're going to, the amount of value you can create with that 20 is massive. the

[00:23:14] Paul Roetzer: the other thing that people want to avoid getting locked in, and then they want to quickly tap into advancements as we, and I hear this All the time.

[00:23:21] Paul Roetzer: we don't want to lock into a 12 month agreement with this company. Cause what happens GPT-5 comes out and it obsoletes that company? we're going to hear some examples of that in a couple of minutes. the other one is desire for control stems sensitive use cases enterprise data security concerns.

[00:23:37] Paul Roetzer: Hear that all the time. I a bunch of time with, a law firm recently, and this was like 1A. was the number one thing they were asking about was how, how do we protect our data and our clients data?

[00:23:49] Paul Roetzer: which of these models can we actually trust with that kind of stuff? And then the other one I thought that definitely makes a ton of sense is Cloud service provider driven decisions.

[00:23:59] Paul Roetzer: [00:24:00] If you're already in, again, big enterprises all the time. you already are a Microsoft customer, going to likely use azure.

[00:24:06] Paul Roetzer: you're a Google cloud customer, you're going to use some model through Google, probably Gemini. And then if you're AWS, you're probably going to go with Anthropic, Amazon's own models or Cohere or something like that.

[00:24:17] Paul Roetzer: So. In a big company where you already have a trusted provider who already has access to all of your data, you are way more likely

[00:24:26] Paul Roetzer: trust them to infuse this stuff in and you don't have to get them through procurement again.

[00:24:31] Paul Roetzer: That's the challenge we're seeing with a lot of these kind of more startup,

[00:24:35] Paul Roetzer: models in the Gen AI space, the companies in that space. Is getting through procurement really hard, especially if they want access your data.

[00:24:43] Paul Roetzer: So overall, again, a quick read, it's

[00:24:46] Paul Roetzer: like a seven minute read, definitely aligns with what we're seeing and hearing in the marketplace.

[00:24:53] Mike Kaput: So I found the chart really interesting where they talked about this specific models [00:25:00] from specific companies that enterprises are using. So like OpenAI and Google, according to the chart, which you can go look at are basically ahead by a mile and the open source move is like pretty clearly Llama. Are we moving towards a reality in the enterprise where.

[00:25:16] Mike Kaput: Even if they are using multiple models, it's like a very small number of them.

[00:25:21] Paul Roetzer: I would so. I think it's going be

[00:25:24] Paul Roetzer: like the cloud space, you know, it's like AWS, Google cloud and Microsoft Azure. That's pretty much it. And I could see

[00:25:32] Paul Roetzer: language models largely falling into that realm where there's three major players.

[00:25:36] Paul Roetzer: One of them, OpenAI slash probably is dominant and less Google gemini really, Powers ahead, which, know, certainly possible.

[00:25:46] Paul Roetzer: And then you've got, you know, Anthropic out there, which obviously they're getting a lot of love for their updated cloud models, but

[00:25:54] Paul Roetzer: I think it seems very viable. And again, going to talk about a couple other topics today that sort of

[00:25:59] Paul Roetzer: [00:26:00] point in the direction of there's going to be a small collection of dominant players, and then there's to to be some other players.

[00:26:07] Mike Kaput: So some final thoughts here. I just had a question on kind of, do you have any take on what this might mean for like enterprise focused startups, not these big companies, the ones trying to break into enterprises with AI solutions?

[00:26:21] Paul Roetzer: I just, I mean, each of the points sort of stack against those startups. You know, I think it's. It's indicating it's going to be lot

[00:26:33] Paul Roetzer: harder for these companies that are trying to build their own models

[00:26:40] Paul Roetzer: or that are, you know, wrappers for the existing models. I think it's just a, it's a continually steepening uphill battle to

[00:26:51] Paul Roetzer: compete with companies that can spend billions building these models and acquiring the talent. And yeah, [00:27:00] I think the next main

[00:27:00] Paul Roetzer: topic, going to see this. Played out like even the ones who are raising billions still have uphill battles. So yeah I think it's a challenging landscape for the companies that don't have tens of thousands

[00:27:13] Paul Roetzer: of Nvidia chips and can pay a million dollar bonuses to get the top AI researchers to switch over to their company.

[00:27:20] Paul Roetzer: it's a very, very, Competitive environment now.

[00:27:25] Paul Roetzer: And you got some of that, even listening to Sam talk, like they asked about the competition and, it's, it's just wildly competitive

[00:27:32] Paul Roetzer: for, for the customers, for the talent, for compute power, for the data. and you don't have the data.

[00:27:40] Paul Roetzer: Then it becomes even harder, which is again, what we've always said it's what, it's what what Google has.

[00:27:47] Paul Roetzer: got lots and lots of proprietary tools and proprietary data. OpenAI has it sort of through Microsoft, through their partnership with Microsoft, but didn't inherently

[00:27:57] Paul Roetzer: have their own proprietary data to start with. [00:28:00] it's where Amazon has, it. you know, potentially a say in this and it's where XAI slash Twitter

[00:28:06] Paul Roetzer: X slash Groq, slash Elon, like they, they have

[00:28:10] Paul Roetzer: The X Twitter feed other people don't. Meta, certainly a player. Like it's really hard to compete with the companies

[00:28:19] Paul Roetzer: have the money and the resources to do the compute and get the people. and

[00:28:24] Paul Roetzer: they have proprietary data. if don't have the proprietary data and you have the money, it may not matter.

[00:28:32] Big Changes at Inflection

[00:28:32] Mike Kaput: Alright, so we're kind of seeing, to your point, this playing out a little bit in our third big topic this week. We've got some major AI news that I'm sure we'll be hearing more about in coming weeks. Mustafa Suleyman, the co founder and CEO of Inflection AI and a former co founder of Google DeepMind, is leaving Inflection to start Microsoft AI, which is a new division at Microsoft that will bring together their consumer AI efforts [00:29:00] as well as co pilot.

[00:29:01] Mike Kaput: Bing an edge as part of this move. Microsoft has also hired most of inflection staff inflection as a company says it will continue with a new CEO, but it will not have its founder and much of its talent. Now, the company set in a blog post that its plan moving forward is to quote, lean into our AI studio business where custom generative AI models are crafted, tested, and fine tuned for commercial customers.

[00:29:31] Mike Kaput: Now, this is a huge deal, and I want to, as we unpack this, it's kind of helpful to remember some of the context here. Inflection is a giant in

[00:29:41] Mike Kaput: ai. It had raised 1.5 billion from investors and those investors included Microsoft. Their

[00:29:49] Mike Kaput: main consumer facing product was Pi, which they billed as the first emotionally intelligent AI assistant.

[00:29:56] Mike Kaput: They're trying to essentially compete in a [00:30:00] consumer context with the open AIs of the world.

[00:30:03] Mike Kaput: Also relevant here is that Reed Hoffman of LinkedIn fame was a co founder of Inflection and sits on Microsoft's board. So Paul, there is a lot to unpack here, but let's start with what

[00:30:15] Mike Kaput: does

[00:30:15] Mike Kaput: this kind of mean in context, and what does this mean for the overall AI ecosystem?

[00:30:20] Mike Kaput: Yeah,

[00:30:22] Paul Roetzer: ironic because we just talked about Pi last week and I was, think it was last week before I was sharing, like I was, I was actually kind of enjoying the new model. I'd only played with it the

[00:30:29] Paul Roetzer: one time, but it was pretty cool.

[00:30:32] Paul Roetzer: I think it like, as to build on the previous topic, it is a tale of caution. um, for sure, as a, as a user, as a or an individual, personally, professionally, um, if you.

[00:30:44] Paul Roetzer: We're enjoying Pi, and it had become like a daily tool for you, whether it was a, you coach, advisor, therapist, friend, mentor, whatever, however you were using Pi. What happens now?

[00:30:55] Paul Roetzer: Like, are they going to at some point realize this studio thing is just, like, it's just [00:31:00] PR and this is not going to work.

[00:31:01] Paul Roetzer: And microsoft just eventually absorbs them? Is there some deal where they get the data in this? Non acquisition acquisition. I mean, in essence, it's an acquisition without actually buying the company. Um, so if, let's say you've been

[00:31:15] Paul Roetzer: having, experimenting pi for the last six months as a friend, like just talking to it, seeing how it works. did Mycroft get that data? Like what happens to all that personal

[00:31:27] Paul Roetzer: information you were talking to Pi about? And so again, I'm not saying Pi in particular, I'm saying like.

[00:31:31] Paul Roetzer: This is what happens. Like we come to trust these AI tools, these AI companies, but what happens to data? What happens when the company goes away and you don't have that tool or they do just get acquired and they shut down and sunset the product. so I

[00:31:44] Paul Roetzer: I think, again, it's just a bit of a cautionary tale of this is a very dynamic space we're living through and it's just starting,

[00:31:52] Paul Roetzer: a couple of other things that I found interesting was, [00:32:00] Again, building on the previous topic, further evidence that we will likely end up having a small collection of dominant

[00:32:05] Paul Roetzer: Companies, dominant models. they had, as you said, 1. 5 billion. And they raised that like in the last 12 months and it's just gone basically.

[00:32:16] Paul Roetzer: Um, one interesting note was they are rumored to have 22, the going rate for those GPUs is, give take, around 30, per gPU. 30, 000 times 22, 000

[00:32:36] Paul Roetzer: There was an article in the Information yesterday that said in the, this exchange, Microsoft was paying 650 million. To inflection. Kind of, kind of real

[00:32:45] Paul Roetzer: close numbers that maybe Microsoft basically is just getting the talent the GPUs NVIDIA, in exchange for whatever. So, yeah. And then like you said, there's just so much intrigue here. The,

[00:32:58] Paul Roetzer: there's a potential conflict with [00:33:00] OpenAI. I mean, they're, they're competitors and Microsoft's the biggest investor in open AI. So what happens there? Microsoft now has a Microsoft AI division, which

[00:33:09] Paul Roetzer: I thought they previously had, but apparently they didn't. So now they have one, and Mustafa is the CEO of it, which is a shake up to all the other AI talent.

[00:33:18] Paul Roetzer: At Microsoft, I would assume. there's that whole thing. And then the one that I just immediately of when I saw this was, isn't this what Sam and

[00:33:26] Paul Roetzer: Greg were going to go do?

[00:33:27] Paul Roetzer: So when Sam got fired that weekend, it was what, a Thursday or Friday.

[00:33:31] Paul Roetzer: By Saturday night, Satya Nadella said, Sam and Greg are going to come work with us. We're going to take, you know, their, their team and we're going to build an AI division with Sam and

[00:33:38] Paul Roetzer: Greg at the core of it. So in essence, Satya got the consolation prize. Like, Oh, okay. Sam went back to aI. All right, fine. We'll, We'll take

[00:33:46] Paul Roetzer: Mustafa and we'll build division around him. So I don't it's so fascinating. This stuff is, it is like a soap opera. I know people comment

[00:33:54] Paul Roetzer: all the time about the podcast. like, in essence, reporting on this soap operas. And this is definitely one of there's, [00:34:00] and I'm sure there's way more to this story than, than has currently been reported.

[00:34:03] Mike Kaput: Well, as you noted too, it's pretty interesting. And potentially repeatable playbook where someone,

[00:34:10] Mike Kaput: instead of having to acquire the full company, is essentially getting all the parts of the machine and leaving the shell of it behind, right? So, as part of all this commentary around a guy named Gavin Baker, who's a really sharp AI commentator at Atreides Capital, he posted a thread on X about the long term implications of this, and he started it out by saying, quote, inflection will likely be the first of many VC backed foundation model companies to fail.

[00:34:44] Mike Kaput: Foundation models without proprietary real time data exist. and Massive Distribution for Reinforcement Learning with Human Feedback,

[00:34:52] Mike Kaput: RLHF, rLHF, are the fastest depreciating assets in history. I think we have discussed some of his [00:35:00] commentary before on this subject. Do you find any truth in that?

[00:35:03] Paul Roetzer: Yeah, I mean, it's, what we just said on the previous topic.

[00:35:06] Paul Roetzer: Like if you don't have data and all have is a model and computing power and some talent and some VC money,

[00:35:15] Paul Roetzer: it's only going to get so far, unless find some massive differentiator, some market that no one else is in. Like, I don't know what it is.

[00:35:23] Paul Roetzer: I haven't sat down and thought deeply enough about it, but, um, Yeah, it's, it's a very challenging environment to invest.

[00:35:31] Paul Roetzer: Like as someone who does some angel investing, some personal investing, like I would have a real hard time

[00:35:39] Paul Roetzer: making personal bets on, on model companies right now. It's, it's just such a unknown and it's, it's really, really hard to bet against the big players that they aren't going to just dominate everything.

[00:35:53] Paul Roetzer: So yeah, I have a a hard time disputing what Gavin's saying there.

[00:35:59] Mike Kaput: [00:36:00] Alright, so in our rapid fire topics today, the hits keep on coming, stability AI,

[00:36:04] Stability AI Might Be Crashing

[00:36:05] Paul Roetzer: Tough week for GenAI companies, man.

[00:36:06] Mike Kaput: yeah. Stability AI, which is a leading AI image generation company, they're the ones behind commercializing the stable diffusion image generation model. They appear to be having some serious issues. Now, we just learned that three of the five original researchers, three of the five people who are integral to the company's founding technology, they have just resigned.

[00:36:30] Mike Kaput: And this is not the first set of high profile departures. Over the last year or so, Stability AI has lost a number of product research and engineering executives. One of these was someone we talked about at the time named Ed Newton Rex, who resigned in November of 2023 to basically just protest the fact that he thought the company was training its models on copyrighted data. This is also not the only drama. to engulf the company. Bloomberg has [00:37:00] reported that at one point the company was burning

[00:37:02] Mike Kaput: eight million dollars a month

[00:37:04] Mike Kaput: and did not appear anywhere near any type of break even point. The investment firm Kotu

[00:37:09] Mike Kaput: Kotu that was leading the round resigned from the board at one point,

[00:37:12] Mike Kaput: and Stability recently sold off ClipDrop, an image generation platform, to Jasper in February.

[00:37:19] Mike Kaput: At one point Forbes also reported that Stability had struggled to pay wages and taxes. Now, This is a pretty dramatic reversal because just 18 months ago, Stability had raised money at a 1 billion valuation, and plenty of people at the time would have considered Stable Diffusion, the model at least, as one of the heavyweight kind of players in this space.

[00:37:44] Mike Kaput: So now it sounds like we're talking, Paul, about a company that seems to be in extremely bad shape. Like, what's going on here? What kind of lessons might we be able to take from this?

[00:37:56] Paul Roetzer: Yeah, so again, if you're a regular listener to [00:38:00] the show, you might, you know, you Have noticed, we don't talk much about Stability

[00:38:05] Paul Roetzer: AI and in some ways it's a bit, we try and be super on this show. I'm not a fan of company. Like I've really struggled to think that they were a legitimate company.

[00:38:15] Paul Roetzer: And the reason is the founding story is very bizarre. so ahmad was, um. they basically provided computing power like these, I think it was five researchers who were developing the stable diffusion model.

[00:38:31] Paul Roetzer: And in their fundraising process, Amad basically took credit for inventing this stable diffusion, which was a key breakthrough kind

[00:38:40] Paul Roetzer: of in the development of image generation and things like that. so the company has all kinds of, it just doesn't smell right. I guess there's one way

[00:38:54] Paul Roetzer: to say it. Like the instinct I have around this is there's just so many stories [00:39:00] about how it was built and it's been run. And,

[00:39:05] Paul Roetzer: I generally just kind of pay attention to what they're doing, but I don't put a bunch of stock in it.

[00:39:13] Paul Roetzer: so think part of this might be representative of it's, as we talk about, it's

[00:39:17] Paul Roetzer: very challenging world to build a Gen AI company in that's building and maintaining these big models. This one could very well also just be, it was

[00:39:30] Paul Roetzer: a company meant from the start to not make it. And again, it's just, I don't know them personally. I've, never interviewed them or anything like that, but when you start seeing all of

[00:39:42] Paul Roetzer: of stories stacking up, you know, over your career, you just. People, we all kind of like learn

[00:39:48] Paul Roetzer: to, to assess a situation that I've personally just always assessed that this was a company that likely wasn't going to maintain

[00:39:56] Paul Roetzer: uh, being a major player because things just didn't add [00:40:00] up so ool tech, but yeah, yeah, company wise, just, I haven't always been very impressed them.

[00:40:07] Mike Kaput: Right. So, we're also seeing

[00:40:10] Mike Kaput: another. leading company facing some headwinds too.

[00:40:14]  Tepid Revenue at Cohere Shows OpenAI Competitors Face Uphill Battle

[00:40:14] Mike Kaput: Cohere, which is a major AI company backed by Jeff Hinton, who's a very notable, some

[00:40:23] Mike Kaput: say maybe even the godfather of of modern aI, and co founded by Aiden Gomez, who is a major AI researcher. They're running into. Some issues according to reporting from the information.

[00:40:36] Mike Kaput: According to that reporting, Cohere was generating just 13 million dollars in annualized revenue at the end of 2023. Now, that is not a lot for a company valued

[00:40:48] Mike Kaput: like Cohere the company has raised more than 400 million dollars. They're reportedly trying to raise another 500 million. Late last year, their fundraising

[00:40:58] Mike Kaput: that were engaging in was [00:41:00] looking to value the firm at 6 billion.

[00:41:03] Mike Kaput: Which would be a crazy valuation given this annualized revenue. It's more than 450 times this annualized revenue, which is also a way higher multiple than even OpenAI, which makes a ton of money, apparently. Now, according to the information, Cohere has kind of tried to now differentiate itself from larger competitors by starting to sell to only enterprises, and in recent months, company leaders have decided not to compete with the big players by developing frontier models.

[00:41:36] Mike Kaput: Someone with direct knowledge of this decision told the information this. Instead, the company is now focused on retrieval augmented generation, RAG technology, which helps reduce AI's hallucinations and increases its accuracy. Now, this is not saying, you know, Cohere is out of money totally or falling apart necessarily, but that's extremely low revenue for a [00:42:00] company that is looking to do what Cohere is doing.

[00:42:03] Mike Kaput: And Paul, I still remember you messaging me when Cohere was first announced. I mean, it's super significant in terms of the people backing it,

[00:42:09] Mike Kaput: people who founded it. These are legit AI leaders who are running this company. Um, you've been watching since literally day one. What can you tell us about Cohere's current troubles and future prospects?

[00:42:22] Mike Kaput: Yeah.

[00:42:50] Paul Roetzer: win.

[00:42:50] Paul Roetzer: like this move to betting on retrieval augmented generation or RAG makes a lot of sense in the current state of [00:43:00] things. But just in the last like five days, I've seen so

[00:43:03] Paul Roetzer: much context in, in tweets from like industry leaders about how these bigger models just outperform everything.

[00:43:15] Paul Roetzer: And especially when they're given time to think the system to thinking we talked about, there was the one example. and we'll

[00:43:22] Paul Roetzer: get to this a little bit with the Claude example, where it was it, was it mollick that shared where if you give. GPT-3. 5 think basically, that it actually outperforms outperforms GPT-4.

[00:43:37] Paul Roetzer: in essence, like, it seems like we're not even fully comprehending what these bigger models are capable of.

[00:43:46] Paul Roetzer: And those bigger models are only getting bigger and smarter. And so the question for all of these companies becomes when we have GPT

[00:43:56] Paul Roetzer: even GPT 6 class models. Gemini2, [00:44:00] Gemini3 class claude 4 class models.

[00:44:03] Paul Roetzer: Does it even matter anymore dubrek? Like, is that even necessary? or is the context window so massive and it just remembers everything and like doesn't and it doesn't hallucinate? Like, do we solve all these inefficiencies?

[00:44:17] Paul Roetzer: And so is there really a market long term these companies that are trying to find the weakness in the current models and solve for it?

[00:44:27] Paul Roetzer: If that weakness doesn't exist in 12 months. And that's where I said, like, don't know, like, I am not the expert. I've

[00:44:34] Paul Roetzer: never built a model like this. Like, I don't, I'm not the inside person working on these things every day. We're just consuming and synthesizing whole bunch of information and trying to connect the dots. And

[00:44:47] Paul Roetzer: the dots seem to be leading to like, Anything you think is a weakness today probably won't be in 12 months or it'll be largely solved.

[00:44:57] Paul Roetzer: and so I think that's the challenge of the environment [00:45:00] is

[00:45:00] Paul Roetzer: how do you build and scale an AI tech company in this environment where 12 months from now what's true today won't be true anymore.

[00:45:08] Anthropic, AWS, and Accenture team up

[00:45:08] Mike Kaput: Well, the good news is it's not all bad news for major AI companies. in companies. a more positive development, Anthropic just announced a partnership with AWS and the consulting firm Accenture to build AI solutions for the enterprise.

[00:45:26] Mike Kaput: According to an Anthropic announcement, over 1, 400 accenture engineers will be trained as specialists in using Anthropic models on AWS, allowing them to provide customers with end to end support that accelerates their AI strategies from concept to production.

[00:45:45] Mike Kaput: So basically, they are teaming up to be able to provide it. Anthropic models running on AWS to Accenture's customers, which seems to plug in a couple of those gaps that we talked about in the A16z research, which is interesting. [00:46:00] Now, also

[00:46:02] Mike Kaput: according to the reporting from the information we just talked about, in reference to Cohere, they also mentioned that Anthropic appears to be doing okay when it comes to revenue.

[00:46:12] Mike Kaput: They, the company has projected it's going to generate more than 850 million in annualized revenue by the end of 2024. Now you can take that estimate with a

[00:46:23] Mike Kaput: of salt if you want, but it sounds a lot better than 13 million per year. Now, Paul, we're going to talk in a bit, a little bit more about Anthropic's flagship model, Claude3 Opus, but how do you evaluate their place in the overall AI race right now?

[00:46:41] Paul Roetzer: They're a major player. And again, if you, if you're not familiar with them, new to the podcast,

[00:46:46] Paul Roetzer: to the space, 2021, 2021, 10 percent of OpenAI staff left to create Anthropic. And the premise was to put safety

[00:46:57] Paul Roetzer: first, they felt OpenAI was no longer [00:47:00] safe AI and the level they would like see it done. So Dario Amadei and a few other people left OpenAI to do it.

[00:47:09] Paul Roetzer: the interesting. Complexity to all of this is like, how is this achieving that mission? So if, if they're racing ahead and Claude3 is truly like GPT

[00:47:19] Paul Roetzer: 4 class model, and they're just building the most models in the world, how are they becoming any different from, or their original mission of like safety first?

[00:47:29] Paul Roetzer: And I've listened to plenty of interviews with Dario that they need to build these bigger models so they can build the safety in. But maybe we'll make a billion a year

[00:47:37] Paul Roetzer: in the process like, I don't know, like it's just, a, I've always found Anthropic to be, a bit of a confusing company. That they have this stated mission,

[00:47:49] Paul Roetzer: and I don't know if the, I don't know how they justify becoming one of the major players in building this massive enterprise revenue model.

[00:47:59] Paul Roetzer: [00:48:00] and still pursue the original mission. Maybe there's some way to do it, I don't know,

[00:48:03] Paul Roetzer: But all that being said, this does definitely fit with the a16Z thing, and I think it also, you know, along the lines of does Cohere, you know, make it, how, you know, Inflection was trying to make it. In this case, this is a

[00:48:18] Paul Roetzer: distribution play. So what Anthropic did is they aligned themselves with Accenture that has the distribution of. Thousands of enterprise clients,

[00:48:26] Paul Roetzer: And so now you immediately have your model baked into there, and then it

[00:48:30] Paul Roetzer: aligns with the cloud service provider choice of if you're on AWS and you already trust them with all your data. You can build right on top of Claude3

[00:48:39] Paul Roetzer: with your data already living in aWS. So it definitely sense. And again, I think if you zoom out from this episode, you can really start to see the trend lines for how adoption is going to work with these models and some of the roadblocks that

[00:48:54] Paul Roetzer: The model companies are going to have.

[00:48:57] Claude 3 Opus + Prompt Library

[00:48:57] Mike Kaput: Alright, we're gonna talk a little bit more [00:49:00] about Anthropic and Claude3 Opus, their flagship model. Um, we're gonna kick off this discussion, Paul, by talking about a LinkedIn post

[00:49:09] Mike Kaput: that you recently put up about a prompt library that Anthropic released where they have collected a ton of really useful business and personal prompts for users to employ using Claude or anything else, frankly.

[00:49:24] Mike Kaput: It seems like it got a lot of attention. Do you want to maybe talk about how that came about and what found valuable about it?

[00:49:30] Paul Roetzer: Yeah, I think it's, I actually got their last

[00:49:34] Paul Roetzer: Anthropic email like different things at Claude3 and they had this prompt library in there. So I was just kind of checking it out. It did get a lot of on LinkedIn, but I think it's cause it's so practical. I

[00:49:45] Paul Roetzer: do think a lot of people still just struggle with prompting. Like it's, it's weird. that blank page syndrome.

[00:49:50] Paul Roetzer: you're staring at the, put your prompt in thing. It's like, what do I say?

[00:49:53] Paul Roetzer: How do I tell it? And so. I think it's very helpful people just to see these prompts. So yeah, again, we'll put link in [00:50:00] the show notes.

[00:50:01] Paul Roetzer: OpenAI has a similar one where they give prompt examples. But in this case, gives like the system prompt, the user prompt, and a sample output. And there's some fun ones in there. There's like play prompts, they

[00:50:11] Paul Roetzer: call them, there's work prompts.

[00:50:12] Paul Roetzer: And so I highlighted a few, there was like a meeting scribe, a grammar genie, a lesson planner, an interview question

[00:50:18] Paul Roetzer: crafter, and I think the key is like, just to look at how these are structured. So Anthropic is telling you, this is the best way to prompt our

[00:50:27] Paul Roetzer: This is what the prompt should look like. But what we know is these models, oddly enough. Function all in a very similar way. And so if a prompt

[00:50:35] Paul Roetzer: well on Claude, it's likely going to work well on Gemini and chatGPT.

[00:50:40] Paul Roetzer: So that was the thing for me is I just thought it was kind of cool. I've seen a ton of positive feedback on Claude3. I've only been able to test it a few times through perplexity. So if you're a

[00:50:50] Paul Roetzer: Perplexity page user, you can go in your preferences and choose which model to use.

[00:50:56] Paul Roetzer: So they have their own model. They have claude.

[00:50:59] Paul Roetzer: They [00:51:00] have, GPT-4 and forget others. There's like five different models you can choose from. So I've experimented with Claude recently, Claude Opus through perplexity. but I

[00:51:10] Paul Roetzer: know, Mike, you've spent more time playing around with it. So like, what has your experience been with Claude3 claude3

[00:51:14] Paul Roetzer: Opus?

[00:51:15] Mike Kaput: Yeah, so to kind of tee this up, Claude, the Claude3 family of models came

[00:51:21] Mike Kaput: out like the beginning march, like these are maybe three weeks old, I think. And they pretty quickly changed. Change the game is probably the wrong phrase, but they have really opened

[00:51:32] Mike Kaput: people's because there's three models. They're all just varying levels of power and speed. So there's Claude3 Haiku, which is the fastest and weakest, middle of the road Claude3 Sonnet, and the most intelligent and slowest, and that really kind of where it comes into play

[00:51:48] Mike Kaput: when building on top of it, right? Not when you're using it necessarily as a user.

[00:51:53] Mike Kaput: It's not really that noticeable. Claude3 opus. And that's the one I've been around with. You can get access to [00:52:00] that in paid, like Claude version. You can buy,

[00:52:04] Paul Roetzer: 20 bucks

[00:52:04] Mike Kaput: month? Is that their model too? It's 20 bucks a month. Yeah.

[00:52:06] Mike Kaput: Yeah. So it's very similar to ChatGPT so what I'm doing, we give this advice to people all the time is my approach to these tools is have dozens of core use cases and prompts that I'm continually refining that I'm then testing across.

[00:52:24] Mike Kaput: Major tools because this is exactly what we've kind of advised people is

[00:52:29] Mike Kaput: Sometimes a model comes along that makes use cases possible, or changes the game a little bit. And this is one of them.

[00:52:38] Paul Roetzer: Which fits into that multi model approach we talked with A16Z.

[00:52:41] Paul Roetzer: It's like, don't into a single, like you always got to be

[00:52:44] Mike Kaput: testing. 100%. 100%. And yeah.

[00:52:47] Mike Kaput: I have started testing especially my content generation and writing related prompts with Claude3.

[00:52:56] Mike Kaput: and it has blown me away. It is crazy good at [00:53:00] writing. It is really, really good at mimicking style. And some of that has also been on

[00:53:06] Mike Kaput: my end, testing more extensive prompts along the style that Anthropic has recommended.

[00:53:13] Mike Kaput: A system prompt to prime the tool to get ready to do what I want it

[00:53:18] Mike Kaput: do, and then a user prompt. And in my case, they're starting to get to pages long because I want to provide examples, context, things like that, so it actually does

[00:53:28] Mike Kaput: I want it to do at a really, really high quality level. And like the writing results I've gotten from Claude3 have been really, really impressive, to the point where I would probably default to it for like a basic writing task.

[00:53:43] Mike Kaput: And this is again, not like we're putting out blog posts that Claude3 is writing, but stuff like summarizing, interpreting transcripts, turning those into like first drafts I can really use to polish up and to get ideas from and riff on headlines, stuff like that. It's [00:54:00] incredible. I've just been really, really impressed.

[00:54:03] Mike Kaput: this

[00:54:03] Mike Kaput: is actually a thing I did a pretty big deep dive in into in. My first ever generative AI mastery session. So this is part of our AI mastery membership that we actually started offering, which is a yearly membership with a number of different benefits each month for people who really want to continue down that journey of AI mastery, on an ongoing basis.

[00:54:29] Mike Kaput: And one of the deep dives I did where I demonstrate. Actual technology with actual prompts was Claude3. We spent a huge amount of time on it in that session, which I think actually dropped a couple hours ago.

[00:54:41] Paul Roetzer: came out, Friday

[00:54:42] Mike Kaput: yeah.

[00:54:42] Paul Roetzer: yeah.

[00:54:43] Mike Kaput: So if that is of interest, that's well worth checking out that membership, but regardless, absolutely.

[00:54:52] Mike Kaput: You need to be checking out Claude3. Claude3. and I think, especially if you're not using any of the paid versions of these models, you're going to [00:55:00] be blown away.

[00:55:01] Paul Roetzer: Yeah, it's one

[00:55:02] Paul Roetzer: of those, I want to get in and spend more time with it, that and gemini, which we're going to talk about in a minute.

[00:55:08] Paul Roetzer: yeah, so the AI Mastery Membership, Mike mentioned, you go to marketingacademy. aIt'll take you right to that information.

[00:55:13] Paul Roetzer: This is a new thing that we did launch at the end of January or the end of February, I think. Yeah. And so we do, in addition to all the other things

[00:55:22] Paul Roetzer: that we offer free, is a paid program. but Mike and I do a quarterly AI Trends Briefing. The Gen AI Mastery Series is

[00:55:29] Paul Roetzer: quarterly thing that Mike does, and then we do a quarterly ask me anything session with me in addition to a host of other

[00:55:37] Paul Roetzer: So yeah, if you're interested the Mastery membership, it is again, marketingacademy. ai.

[00:55:43] Mike Kaput: So, yeah, I say, in terms of a model you absolutely have to check out and continue trying to cycle through, couldn't recommend right now Claude3 Opus enough, so definitely give that a whirl.

[00:55:56] Mike Kaput: you know, on the theme of things [00:56:00] that are just blowing my mind here, we, Google has now granted everyone access to Gemini 1.

[00:56:07] Mike Kaput: 5, which is its latest, most powerful model. So they have a sandbox environment through their AI studio. Will you, we will put the link in the show notes.

[00:56:17] Paul Roetzer: It's just aistudio. google. com, right?

[00:56:19] Mike Kaput: I

[00:56:19] Paul Roetzer: believe so,

[00:56:20] Mike Kaput: yeah. I'll check it while you're talking. Yeah, I will.

[00:56:23] 

[00:56:23] Gemini 1.5 API + Google AI Studio

[00:56:23] Mike Kaput: so this is not even just Gemini Advanced, we have a paid subscription to, which is Gemini 1. 0, which is a super powerful new model that Google released.

[00:56:34] Mike Kaput: This is the next version. this is also making a a ton of waves because it has a 1 million token context window. So in addition to just being a far more advanced model, In terms of reasoning, Gemini 1.

[00:56:50] Mike Kaput: 1. 5 pro can, according to Google, process vast amounts of information in one go. That includes things like one hour of video, [00:57:00] 11 hours of audio,

[00:57:01] Mike Kaput: code bases with over 30, 000 lines code, Or over 700, 000 words.

[00:57:08] Mike Kaput: So, since anyone can access Gemini 1. 5 Pro right now, in this test environment, we actually went and tried it out. So, for a test, we used a topic that we're probably going to be talking about. Next week, a little more on the podcast. Right before recording, we found the White House had released its annual economic report of the President, the ERP, which is an annual report produced by the Council of Economic Advisors.

[00:57:35] Mike Kaput: It's on every topic imaginable, and one chapter this year was about AI. Now, the problem is, this report is literally almost 500 pages

[00:57:46] Mike Kaput: long.

[00:57:46] Mike Kaput: That chapter alone 50 pages. And it is not easy to read writing, let tell you. Um, you. shout out to all the very smart

[00:57:56] Mike Kaput: people in the U. S. government, but it is very academic, let's say [00:58:00] that. But, we actually just dropped the entire report into Gemini 1.

[00:58:06] Mike Kaput: 5 Pro. And just ask it to summarize the AI chapter. Now, it'll actually tell you, when you upload a file, how many tokens it is. And this was, the entire report was 675,

[00:58:17] Mike Kaput: 675, 000 tokens. So we didn't even hit full 1 million token limit.

[00:58:24] Mike Kaput: Gemini appears to have done a pretty good job. I went through and kind of spot checked and asked a couple questions, opened up the actual report, started reading it a bit. The summary was, in my experience, perfect. Like, it was on point. I checked the themes it talked about. It referenced specific language and frameworks that the government was advising, so I went and tried to hunt those down, make sure it wasn't making those up. Those were accurate. I asked where it got, where did the government get the data from, because it referenced at one time the government used a bunch of data sets, so it listed those [00:59:00] out.

[00:59:00] Mike Kaput: I didn't check every single one, but a spot check looked like it was accurate. In the government report, like an academic paper, they cite work by researchers and put their name, their last names in parentheses and say like, hey, we're citing so and so's research from like 2018 or whatever. So I said, hey, could you give me the list of researchers cited in the AI chapter?

[00:59:23] Mike Kaput: Again, I got, you know, probably 50 names back. I didn't check every single one, but I checked 10, 15 of them all accurate. It was incredible. This was like looking the future of what large context windows are. So

[00:59:37] Mike Kaput: I'll stop with that there at that experiment, Paul, but like, you really have to take seriously that I don't know if we're going to get a million tokens in like a 20 buck a month, Gemini

[00:59:49] Mike Kaput: advanced account. Bye. It's going to be crazy. Because one of the much content.

[00:59:54] Paul Roetzer: researchers tweeted, yesterday that the 10 million context is working too. Yes, [01:00:00] Like they're, you yes. Yeah, and

[01:00:04] Paul Roetzer: back to what we said about how hard this is, like, a year, year and a half ago, Mike, I know you talked

[01:00:09] Paul Roetzer: these companies, like, there was, there was people raising funding to build PDF summarizers.

[01:00:14] Paul Roetzer: Yeah. Like, that was a company idea. a year and a half ago. And now just play around in the sandbox you

[01:00:22] Paul Roetzer: drop a 300 page PDF and it analyzes it. it is aistudio. google. com. It is more technical. It's like a a sandbox, for users, but you can, there's even a, a new tuned model.

[01:00:37] Paul Roetzer: you can come in here and it says, like. You can tune a model from an existing structured prompt or create one by importing from Google sheets

[01:00:44] Paul Roetzer: or a CSV. Tuning only works text, but this time we recommend 100 to 500 examples. So if you want to like. Train a model

[01:00:51] Paul Roetzer: to do something, you, you can come in here and do this stuff again. Like I can't stress enough the two things we [01:01:00] always like say

[01:01:01] Paul Roetzer: are so critical as AI literacy and competency the competency comes from experimentation if if don't go and like see what these things are capable yourself, or if it's not

[01:01:13] Paul Roetzer: you like make sure someone in your company is constantly testing. These are like, Mike and I push other all the time on stuff. Like we were having coffee

[01:01:21] Paul Roetzer: this morning, I met for breakfast and we're like going back and forth on this and like, Oh, we're recording like five hours. Hey, let's drop that 300

[01:01:27] Paul Roetzer: PF and see what happens.

[01:01:29] Paul Roetzer: This is the kind of culture you have to have. Like have to have people, whether it's your AI council or more informal, where you're just watching what's going on and saying, man, this might, this might be

[01:01:38] Paul Roetzer: significant. This could change the way we do X.

[01:01:41] Paul Roetzer: so yeah, hopefully that's part of what this show does. Does is like you the inspiration to get out there and try stuff because listening to us talk about, it's only going to get

[01:01:50] Paul Roetzer: so far but this stuff is crazy like it's the kind of stuff I just oh and I just noticed there's prompt gallery in here just like we were talking about with Anthropic.

[01:01:56] Paul Roetzer: You can go in here and see a whole bunch of prompts. [01:02:00] it's the kind of like I wish I had

[01:02:02] Paul Roetzer: three hours this afternoon. To actually go play around myself these things, but it's not always the case. And so again, depending on your role in the company, you might not be the one that gets to go

[01:02:11] Paul Roetzer: do these things, but create the culture that enables it, that everyone kind of learn and share internally.

[01:02:16] Mike Kaput: Yeah, and one last note here, and I've harped on this before and I'm sure I will in the future, is you have to have a use case first mentality here. You need to be documenting, even if you have not tried out AI for certain use cases, I need you

[01:02:30] Mike Kaput: to like,

[01:02:31] Mike Kaput: start thinking about what

[01:02:32] Mike Kaput: would I love to if it were possible AI could

[01:02:35] Mike Kaput: for me, because I can tell you for a fact, several of the things that I put through Claude3 Opus, and Gemini 1. 5 were things that were aspirational, that I was not able to

[01:02:48] Mike Kaput: do before that are now unlocked and I'm ready to go. able to demonstrate the technology and start doing more things and essentially gaining more superpowers

[01:02:58] Mike Kaput: I prepped in advance. [01:03:00] So it's helpful to continually do that.

[01:03:04] Mike Kaput: All right. A couple more stories here until we, before we wrap up our rapid fire,

[01:03:09] Chrome AI Rollout

[01:03:09] Mike Kaput: Google Chrome has also quietly rolled out

[01:03:11] Mike Kaput: a couple of AI

[01:03:13] Mike Kaput: that may be worth experimenting with.

[01:03:15] Mike Kaput: And the biggest one is an AI powered writing suggestion. feature right within Chrome. So this feature is called Help Me Write, and it will give you AI writing suggestions when you write on the web. So once it's enabled in Chrome, you right click on an open text field while using the Chrome browser, and then click Help Me Write.

[01:03:35] Mike Kaput: A prompt window opens up and you can prompt Chrome's AI

[01:03:38] Mike Kaput: AI to write

[01:03:39] Mike Kaput: or rewrite text as you need. Now, Chrome also, at the same time, released a way to generate your own custom Chrome visual theme using AI, which is kind of fun, but not super important. But, Paul, you had flagged especially the Help Me Write feature, and I think you've maybe experimented with it a little bit.

[01:03:57] Mike Kaput: What were your thoughts here? [01:04:00] Gab.

[01:04:01] Paul Roetzer: it takes you to the What's New tab.

[01:04:03] Mike Kaput: Yeah.

[01:04:03] Paul Roetzer: And it is just like, This infusion of help me write button throughout. And then there's also a, this, what you create an an AI based theme and stuff.

[01:04:11] Paul Roetzer: And then there's like a smart organizer for tabs. we first talked about this back on episode 81, which was end of November?

[01:04:20] Paul Roetzer: Or end of January, sorry, 2024. And at the time, what we were saying, and this is why I think this becomes so significant, is they have this massive distribution of this thing. So there's.

[01:04:32] Paul Roetzer: What, over a billion users, I think, or they have 63% of all internet users use Chrome. And what we said at the time was this complexity.

[01:04:42] Paul Roetzer: of this landscape. Like I have Cohere, I have GeminI have Anthropic, Claude. I have all these

[01:04:47] Paul Roetzer: options. but we're more. advanced users of like, we're seeking the latest thing, we're trying to go find these tools.

[01:04:56] Paul Roetzer: The vast majority of people aren't doing what we're doing. [01:05:00] You know, if you listen to this you're likely in our camp. You're an early adopter, you're to like be a first mover. You're trying

[01:05:04] Paul Roetzer: to figure this stuff out. The average internet user though, the help me write button from Chrome might be the first time they actually use an AI tool, like knowingly use a AI

[01:05:14] Paul Roetzer: AI writing tool. And so I think that's the key here is, I don't know how good it's going be, but I know it's going make a lot of the other things you have

[01:05:23] Paul Roetzer: Redundant. So if I have the Chrome extension turned on and I go into LinkedIn, well, I already have a help write button in LinkedIn. Am I just going to

[01:05:30] Paul Roetzer: to ignore that one or rewrite button in LinkedIn? Or. if I'm in HubSpot and I have my iWriter, or if I'm using Jasper or Writer or Claude or like whatever I'm doing,

[01:05:39] Paul Roetzer: this thing's just going to follow me everywhere. if there's a form where I can write something, it's gonna be there with me. Yep. And that starts to get into like, you could start to see the changing of behavior. And if that is the gemini model, like if, if six months

[01:05:53] Paul Roetzer: now we have Gemini 2 and this 1. 5 model we're talking about that has a million tokens of context, [01:06:00] if that's the thing, that's a free extension baked

[01:06:03] Paul Roetzer: right into Chrome, that changes things. And so that's I find so fascinating about this is like, how

[01:06:10] Paul Roetzer: how is this going to affect and it, I don't have the answers at all. Obviously neither of us do Mike, but this is like, it's just going to be such a wild

[01:06:20] Paul Roetzer: time this year. I don't know. So I'll report back if I actually like really start testing it. But right now I think it's going to be more of the non advanced user who's going to start experimenting with this stuff all the time.

[01:06:33] Paul Roetzer: Like I could see my saying like, Oh yeah, there's like, help me write button popped on my Chrome. Like, should I use it? Like, I don't know.

[01:06:41] Mike Kaput: So our last news item really,

[01:06:45] Mike Kaput: I think, brings all of these themes today together. a new deep dive was just published from the famous tech writer

[01:06:53] How Eight Google Employees Invented Modern AI 

[01:06:53] Mike Kaput: Stephen Levy in Wired Magazine, detailing the behind the scenes story of how In their words, quote, eight Google [01:07:00] employees invented modern AI. The story looks at how eight AI researchers within Google in 2017 came to publish the research on Transformers, which is the technology that kicked off the generative AI revolution.

[01:07:16] Mike Kaput: The paper that introduced Transformers is called Attention Is All You Need, and it's

[01:07:20] Mike Kaput: Possibly the most famous paper in modern AI. Now, according to this article, approaching its 7th anniversary, the attention paper has attained legendary status. The authors started with a thriving and improving technology, a variety of AI called neural networks, and made it into something else.

[01:07:38] Mike Kaput: A digital so powerful that its output can feel like the product of an alien intelligence. Called Transformers, this architecture

[01:07:47] Mike Kaput: the not so secret sauce behind all those mind blowing AI products, including ChatGPT and graphic generators such as DALL E and MidJourney. All eight researchers are no longer at Google, [01:08:00] and all of them are kind of their own type of celebrity in AI and

[01:08:04] Mike Kaput: include people who have gone on to found notable AI companies like Aiden Gomez, a co hero like we just talked about. Now this story is well worth a

[01:08:11] Mike Kaput: read. It gives an inside look at all the work, the conviction, kind of the sheer accidents that had to happen

[01:08:17] Mike Kaput: for this paper to come together and this technology. to transform the world. It also prompted a response

[01:08:24] Mike Kaput: Yann LeCun who noted that the story leaves out other notable innovations on the way to transformers that also clearly inspired the researchers.

[01:08:33] Mike Kaput: Now Paul, this is another AI topic you have followed since almost day one. Can you unpack for us why a 2017 paper and the people behind it are worth paying attention to in 2024?

[01:08:45] Paul Roetzer: Yeah, we, I'll refer people back to episode 81. So, from January 30th. 2024. We, we, I did a piece on where are they now, where I actually was some research and I polled the eight people

[01:08:56] Paul Roetzer: and kind of what companies they were with and, um, they'd [01:09:00] collectively raised what, 1. 3 billion or 1. 5 billion or something like that.

[01:09:05] Paul Roetzer: so that episode goes into like the the, the full history behind this. quick,

[01:09:11] Paul Roetzer: Um, highlights, I guess. So again, I think most people listen to the show are probably aware of this, but GPT stands for generative transformer.

[01:09:19] Paul Roetzer: That is where this name came from. So chat, GPT.

[01:09:24] Paul Roetzer: these eight people also did an interview with Jensen Wong from the CEO of Nvidia this week at

[01:09:29] Paul Roetzer: GTC conference. So you can actually go watch the interview of the eight of them together. And then, the, thing I caught my attention, Mike, I know you highlighted

[01:09:39] Paul Roetzer: in like your notes when you're reading through it. is, one of them was involved Q*. what was the, what was the Q that up in the article?

[01:09:48] Mike Kaput: So the gentleman's name, his last name is Kaiser, and an excerpt, and I highlight this because with our other topics, I was like,

[01:09:56] Mike Kaput: this seems really strange they've mentioned this. Kaiser. So. [01:10:00] According to Stephen Levy, the journalist, Kaiser is the only one who hasn't founded a company. He joined OpenAI and is one of the inventors of a new technology called Q*, which Altman said last year will, quote, push the veil of ignorance back

[01:10:12] Mike Kaput: and the frontier of discovery forward.

[01:10:15] Mike Kaput: Here's the part. When I attempted to quiz Kaiser on this in our interview, the OpenAI PR person almost leaped across the table to silence him.

[01:10:27] Paul Roetzer: Oh man. Soap opera. Like I said, it's a soap opera. So yeah, I

[01:10:33] Paul Roetzer: I just read the word article, go listen to episode 81 where we talked about it, just check the timestamps. can zoom right to that part. but

[01:10:42] Paul Roetzer: I think just, said, this is a great way to end it. This is kind of like the people that sort of started this

[01:10:51] Paul Roetzer: evolution back in building on, as Yann said earlier, breakthroughs. and so it's kind of where we started and [01:11:00] Q* might be where we're going

[01:11:02] Paul Roetzer: If we have PR people leaping across the table to shut people up and Sam Altman saying, we're not ready to talk about that in an angrily tone. So, oh man. All right. Good stuff. All right. Yeah. Thanks for pulling it all together, Mike.

[01:11:16] Mike Kaput: Yeah, no problem, Paul. Thanks for unpacking all this and all the connections between these stories. We really appreciate it. Until next week.

[01:11:24] Paul Roetzer: Alright, everyone. Thank you. Oh, and as Mike always says, check the newsletter. There's like things that didn't make the cut this

[01:11:29] Paul Roetzer: including Project Groot, which I really wanted to talk about. Maybe we'll do that next week. So that was a cool one. NVIDIA is doing this crazy stuff robotics.

[01:11:36] Paul Roetzer: We've got to talk about that one next week.

[01:11:38] Paul Roetzer: All right, everyone. Have a great week. We will talk to you again soon.

[01:11:41] Thanks for listening to The AI Show. Visit MarketingAIInstitute. com to continue your AI learning journey. And join more than 60, 000 professionals and business leaders who have subscribed to the weekly newsletter, downloaded the AI blueprints, attended virtual and in person events, [01:12:00] taken our online AI courses, and engaged in the Slack community.

[01:12:04] Until next time, stay curious and explore AI.

Related Posts

[The Marketing AI Show Episode 28]: The AI Inflection Point is Here

Cathy McPhillips | January 4, 2023

This podcast episode is dedicated to theorizing about AI’s inflection point and how it will reshape business in 2023.

[The Marketing AI Show Episode 57]: Recap of 2023’s Marketing AI Conference (MAICON), Does Sam Altman Know What He’s Creating? and Generative AI’s Impact on Jobs

Cathy McPhillips | August 1, 2023

This week's episode of The Marketing AI Show talks about MAICON 2023, a mind-blowing article on Sam Altman, and generative AI's impact on jobs.

Inside the Massive Shakeup at Inflection AI

Mike Kaput | March 26, 2024

We just got some huge news about a leading AI company. And it's a wake-up call for any AI startup trying to take on the tech giants.