AI development is facing obstacles from both technology developments and politics. Join Mike and Paul as they explore OpenAI's Orion challenges and Trump's regulatory shakeup. Plus, Mike and Paul share their firsthand experience using Generative AI to revolutionize planning meetings - turning hours of brainstorming into minutes of focused strategy.
Listen or watch below—and see below for show notes and the transcript.
00:03:57 — Trump Administration + AI
00:17:06 — OpenAI + Model Improvements
00:29:06 — Gen AI Planning Assistant
00:39:56 — Visa Case Study + GenAI Spending
00:44:08 — OpenAI in Talks to Transform to For-Profit Company
00:46:54 — OpenAI + Chat.com
00:49:36 — Perplexity’s Near $9 Billion Valuation
00:52:20 — Perplexity CEO offers AI for striking NYT staff
00:55:16 — Claude 3.5 Haiku
00:57:35 — Amazon and Anthropic
01:00:06 — Robot AI startup Physical Intelligence
01:02:13 — OpenAI Copyright Case
01:08:33 — Google Jarvis
Trump Administration’s AI Policy
President-elect Donald Trump has signaled sweeping changes to the nation's artificial intelligence strategy.
At the center of his plans is a promise to dismantle President Biden's landmark AI executive order, which established crucial safety and privacy standards for AI development.
The incoming administration's approach appears to be shaped by several key advisors, most notably Elon Musk, who contributed over $100 million to Trump's campaign and has been outspoken about AI development.
Among the most immediate changes expected is the potential elimination of the AI Safety Institute, created under Biden's executive order to evaluate advanced AI systems.
However, a coalition of tech companies and think tanks is racing to convince Congress to make the institute permanent before Trump takes office in January.
The new administration's AI agenda seems focused on reducing regulatory barriers and promoting what they call "AI development rooted in free speech and human flourishing." This includes pushing back against what Trump's allies term "woke AI," with potential pressure on tech companies to disclose or revise algorithms deemed politically biased.
Trade policy could also significantly impact AI development, with Trump proposing a 10% blanket tariff on U.S. imports and 60% on Chinese products.
What seems to be missing is a clear replacement framework for the regulations Trump plans to dismantle.
Open AI Model Improvements
A shift is occurring behind the scenes at OpenAI, as the company grapples with an unexpected challenge: the pace of improvement in its core AI technology appears to be slowing down.
The Information is reporting that the company's upcoming flagship model, code-named Orion, is revealing the limitations of current AI development approaches. While Orion does surpass previous models, the improvement is notably smaller than the dramatic leap seen between GPT-3 and GPT-4, according to reporting from The Information.
Some OpenAI employees report that Orion isn't consistently better at certain tasks, particularly coding, despite potentially higher operational costs.
This situation challenges a fundamental assumption in artificial intelligence: that these systems would continue to improve at a consistent rate given more data and computing power.
What appears to be at the heart of this slowdown is a bottleneck: the amount of high-quality training data. OpenAI has apparently largely exhausted publicly available text and data sources, forcing the company to experiment with AI-generated training data.
However, this approach has introduced new complications, with Orion sometimes mimicking the limitations of the older models used to generate its training data.
A Practical Use Case for GenAI Planning
What traditionally would have been a blank-page brainstorming session between Mike and Paul turned into a highly productive strategy meeting, thanks to some preliminary work with ChatGPT.
Instead of starting from scratch, Paul used ChatGPT to develop comprehensive drafts of their plan, covering everything from planning and production to promotion and performance metrics.
What would have typically taken 10-20 hours of deep thinking was accomplished in just three minutes. While the AI didn't necessarily generate ideas, it provided a structured 2,000-word brief that gave a solid foundation to build upon.
The result? Rather than spending a two-hour meeting staring at a blank page, they were able to immediately dive into substantive discussions.
The key lesson? AI tools aren't just for content creation—they're invaluable planning assistants that can transform the efficiency of strategic discussions.
Today’s episode is brought to you by our AI for Agencies Summit, a virtual event taking place from 12pm - 5pm ET on Wednesday, November 20.
The AI for Agencies Summit is designed for marketing agency practitioners and leaders who are ready to reinvent what’s possible in their business and embrace smarter technologies to accelerate transformation and value creation.
You can get tickets by going to www.aiforagencies.com and clicking “Register Now.” When you do, use the code AIFORWARD200 for $200 off your ticket.
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: Neither campaign really talked about AI much at all. My opinion was. They didn't know how the public perceived AI, so there were no votes to be won by talking about AI on the campaign trails. But we knew it was going to be fundamental to whatever happened once the administration, whichever one it was going to be, came into office.
[00:00:20] Paul Roetzer: Welcome to the Artificial Intelligence Show, the podcast that helps your business grow smarter and better. 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.
[00:00:49] Paul Roetzer: Join us as we accelerate and accelerate. AI Literacy for all.
[00:00:57] Paul Roetzer: Welcome to episode 1 23 of the [00:01:00] Artificial Intelligence Show. I'm your host, Paul Reer, along with my co-host, Mike Caput. so we had an election last week, in the United States. Mike and I have a policy not to discuss politics on the show. no one cares about our political opinions and too many tech podcasts went the way of political shows, in my feeling over the last year.
[00:01:23] Paul Roetzer: So Mike and I have a commitment to you not to do that. However, when it influences AI and the future of AI, we need to talk about it. So we will be discussing, some early signs of what we think the new administration means to AI. And technology more broadly. So that is going to be a main topic today.
[00:01:43] Paul Roetzer: We're also going to get in some rumors, I guess. I don't know. It's being reported that maybe the frontier models have plateaued in their training. So we're going to kind of unpack that cause that's a very important thing. And then Mike and I are actually going to share a really [00:02:00] cool internal use case that we used.
[00:02:02] Paul Roetzer: ChatGPT for in particular for some planning that I think people could apply as they're doing their 2025 plannings. And then a whole bunch of rapid fire items. So we got a lot to cover. this week's episode is brought to us again by AI for Agency Summit. This is our virtual event. The second year it's happening.
[00:02:19] Paul Roetzer: It is occurring Wednesday, November 20th from 12 p. m. to 5 p. m. Eastern Time. You cannot make that time, zone. You can get on demand. So there will be an option to get the summit on demand. it's an incredible lineup. It's going to be packed with basically trying to figure out how to help you drive AI transformation in your own agency.
[00:02:41] Paul Roetzer: And for your clients, if you're on the brand side, let your agencies know about the summit. 'cause you want agencies that are proactively seeking to drive AI literacy and capabilities within their own firms. And this is a great headstart for them to do that. So you can go to ai four fo r [00:03:00] agencies.com.
[00:03:00] Paul Roetzer: That's ai four agencies.com. Check out the agenda, check out the speaker lineup. Then click register. Now you can use promo code AI forward 200 for $200 off. So again, that is AI for agencies.com and the promo code is AI Forward 200, and that will get you $200 off the ticket. And again, if you can't make the live event, check out the on demand ticket options as well.
[00:03:27] Paul Roetzer: All right, Mike, the topic you, neither of us really wants to talk about that has to do with politics we're going to talk about. So let's do this in our style is as objective and unbiased as we can humanly be when it comes to these things and just be, report the facts as they appear to be. So that is our kind of promise to you is we're going to do our very best anytime we talk about this topic related to politics to just give you the facts and let you all figure it out for yourselves from there.
[00:03:57] Mike Kaput: Alright, Paul, so let's dive [00:04:00] into it. According to a few reports we're starting to get, it seems like possibly President elect Donald Trump is potentially signaling some pretty significant changes to the nation's artificial intelligence strategy. So, it sounds like At the center of his administration's plans, might be a potential dismantling of President Biden's landmark AI executive order, which established safety and privacy standards for AI development.
[00:04:30] Mike Kaput: We know for a fact that the incoming administration is shaped in some ways by several key Elon Musk, who contributed over a hundred million dollars to Trump's campaign and has been outspoken about his views on how AI should be developed. We're kind of seeing among the most immediate changes expected is potentially the elimination of the AI Safety Institute that is created under Biden's executive order in order to evaluate advanced AI systems.
[00:04:59] Mike Kaput: [00:05:00] However, as always, tech companies, think tanks are kind of racing to have their own influence. Some of them are trying to convince Congress to make the institute permanent before Trump takes office in January. And the new administration's AI agenda seems to be aligning around reducing regulatory barriers and promoting AI development that, in their words, is kind of rooted in free speech.
[00:05:27] Mike Kaput: So this kind of pushes back against what some Trump allies kind of term, you know, quote unquote like woke AI with like biased or politically skewed results. And then of course there's also going to be What we've heard rumblings about is trade policies that could also impact AI development. Trump has proposed a 10 percent blanket tariff on U.
[00:05:47] Mike Kaput: S. imports and 60 percent on Chinese products, all of which would affect some of the hardware, and products that go into artificial intelligence. And we don't seem to [00:06:00] have any real clarity on kind of what would replace this. Would it be just an unfettered set of policies? Would there be additional regulations?
[00:06:08] Mike Kaput: We're not sure yet. So, Paul, it kind of seems like we could see some swift, sudden changes to AI policy under the Trump administration. How are you seeing this playing out right now?
[00:06:20] Paul Roetzer: Yeah, I think there are breadcrumbs, you know, there's a couple articles that came out over the weekend we'll put into the show notes about this, but, the way I, the way I kind of think about this is to follow the money is a good way to do it.
[00:06:33] Paul Roetzer: So while the campaigns, neither campaign really talked about AI much at all, and we talked about this on the show, that my opinion was they didn't know how the public perceived AI. So there were no votes to be won, by talking about AI on the campaign trails. But we knew it was going to be fundamental to whatever happened once the administration, whichever one it was going to be, came into office.
[00:06:59] Paul Roetzer: So, [00:07:00] when we look at the Trump administration, you can look back over the last year plus of who are the major technology organizations and individuals that backed the campaign. you mentioned Elon, obviously, but, Andreessen Horowitz came out a few months back and, kind of counter to their traditional views, they went all in on Trump.
[00:07:23] Paul Roetzer: Now, they published in fall 2023, we talked about this on the show, the Techno Optimist Manifesto, so if you want to understand why A16Z and Andreessen Horowitz would back Trump, you could probably go read the Techno Optimist Manifesto, and that will give you a good sense of In essence, they are very, very pro startups, so they want to, enable the startup ecosystem to innovate as much as possible with as little regulation as possible.
[00:07:54] Paul Roetzer: And crypto. Those are two, you know, two or three big areas. Startups, regulation, crypto. [00:08:00] so they, they listed several reasons they believe the Biden administration is stifling startups through over regulation and potentially needless taxation. The, I guess sort of like, the tipping point for them was the Biden administration had proposed a tax on unrealized capital gains, meaning as startups become higher and higher in value that they wanted to actually tax that increase in valuation before the money was actually realized by anybody.
[00:08:29] Paul Roetzer: And that was the, what they claim was the big, big deal. Break for them. Another major player to look into is Peter Thiel. Peter Thiel is the kingmaker for J. D. Vance. So, when Vance was elected to Senate in 2022, it was Thiel who funded that campaign, who basically pushed for him to get into that position.
[00:08:50] Paul Roetzer: And Thiel along with David Sachs and some others are the ones who basically pushed for J. D. Vance to get the ticket as the vice president. [00:09:00] So, Vance was put into power through his VC connections. you can go into Peter Thiel's history, he, co founder of PayPal with Elon Musk, Palantir Technologies Founders Fund, so he's a VC as well.
[00:09:15] Paul Roetzer: He was the first in Silicon Valley to publicly support Trump back in 2016, gave a bunch of money, was very vocal, appeared at the Republican National Convention in 2016, then pushed for, you know, Vance to be involved. He's a libertarian, which is someone who advocates for minimal government intervention in individuals personal and economic lives.
[00:09:35] Paul Roetzer: so again, get rid of the regulation, let the technology do its thing, maximize personal freedoms while minimizing the role of the state. And then there's David Sachs, the All In podcast. I'm sure there's some of our listeners that are also All In podcast listeners. As you would be aware, David Sachs, who was a founding member and chief operating officer of PayPal, back with Peter Thiel all those guys.
[00:09:58] Paul Roetzer: he was very, very [00:10:00] vocal on the ALL IN podcast and got Trump, I think Trump appeared on the ALL IN podcast if I'm not mistaken. He was also the founder, Saks was of Yammer, which he sold to Microsoft in 2012 and he also has a VC firm. And then obviously Elon Musk. So again, if you want to know the general technology agenda and where AI may go, you can look at some of those key players and what they have publicly stated about their beliefs.
[00:10:26] Paul Roetzer: they're going to have an influence. Now, not all of Silicon Valley was supportive of the Trump campaign and the next Trump administration, but they certainly all lined up over the weekend to congratulate. So, you can go and look at Bezos, and Bill Gates, and Aiden Gomez of Cohere, and Jack Clark of Pichai of Google, and Mark Benioff of Salesforce, and Greg Brockman of OpenAI.
[00:10:48] Paul Roetzer: Every single one of them had the congratulatory tweets, and we're looking forward to working with your administration. And again, whether they supported that administration or not, what you're going to have to work with them. so at the end of the [00:11:00] day, I kind of made my like scorecard, like right away, I was, I had this scorecard before I started looking into the rest of the stuff and, the losers in my opinion, climate, climate regulation done, like they're, they're going to do anything in their power to not worry about climate change.
[00:11:16] Paul Roetzer: And they're certainly not going to care about whether or not advancing AI impacts the climate. So that is. That is a direct loser. the executive order as you called for is done. They will can that as soon as possible. So that is a loser. A dark horse loser here is OpenAI and Sam Altman. And the reason for that is Elon Musk's influence.
[00:11:38] Paul Roetzer: So as we have talked about many times on the show, Elon was a co founder of OpenAI. He put the first 40 million in, he created the name. He lost a power struggle in 2019 when he tried to roll OpenAI into Tesla. Sam came to power, and Elon has a beef with Sam, and it's very public. And Elon [00:12:00] like tends to hold grudges.
[00:12:01] Paul Roetzer: And so there's a, you know, maybe Sam and Elon kind of come to peace, but if they don't, like Elon could probably, with his influence, make Sam and OpenAI's lives miserable if he chooses to. the other side to that is XAI, Elon Musk's startup AI company. You could imagine him getting way more support and power for his own AI startup.
[00:12:24] Paul Roetzer: because he's bought himself some pretty significant influence. And then open source is probably the other and maybe the biggest winner is, when you remove regulation, you limit the, you know, the government's say in the downsides of open source and acceleration of technology. So the e/accmovement, that, that acceleration all costs.
[00:12:47] Paul Roetzer: They're a winner too, and so open source probably is a byproduct of that, and that's one of the A16Z plays, is they want, like, kind of this freedom to innovate there. the one big variable in all this, and I don't know, like, [00:13:00] how this plays out, but the reality is Trump and Elon Musk both have very large egos, very large personalities, and they are both alphas.
[00:13:09] Paul Roetzer: And how those two get along for four years is going to be really interesting. Because Elon, while he may not hold some official position in the administration, he is certainly going to have influence. And if he starts getting a lot of public credit for things that happen, you almost wonder if that doesn't create some friction.
[00:13:32] Paul Roetzer: So I have no idea, but it'll be really interesting to watch how that power dynamic plays out, and what kind of ends up happening. And again, keep in mind with Elon. He, he is historically a supporter of Democrats, like he's very public about the fact that he voted for Hillary Clinton. It's rumored he voted for Biden.
[00:13:53] Paul Roetzer: When he took over Twitter in April 2022, he tweeted, for Twitter to deserve public trust, it must [00:14:00] be politically neutral, which effectively means upsetting the far right and the far left equally. So, Elon's move to support Trump wasn't This isn't, wasn't assumed all this time. I mean, Elon has tried to remain somewhat independent.
[00:14:15] Paul Roetzer: but there was a shift and some believe the shift actually occurred when he got snubbed at the electric vehicle event at the White House a couple years back when he wasn't invited there. and it ended up being because it was like a, in essence, a union event and Tesla is not a union shop. But it seems like that may have played a role in him feeling burned by the current administration and, shifting his support.
[00:14:43] Paul Roetzer: But who knows? But, yeah, it's, it's a dynamic situation and, like I said, Mike and I'll keep tabs on it and report the facts to you, again, as an unbiased and objective way as possible, just observing the space. And so those are some of the things that [00:15:00] I see early on, but I think we're going to learn a lot more in the, in the months ahead leading up to.
[00:15:05] Paul Roetzer: The Inauguration in January.
[00:15:07] Mike Kaput: Yeah, we'll definitely see how this all plays out in practice, but you mentioned this briefly, it just seems like, at the moment, I would not be betting on anything other than EAC. movement, the effect of accelerationism seems to be the order of the day here.
[00:15:22] Paul Roetzer: And I think this probably also, you know, I hadn't thought about this one.
[00:15:26] Paul Roetzer: It probably accelerates the States like California being way more aggressive. So if you remember when SB 1047, that's right. Yep. Yeah. When that, when, Newsom did not sign that bill. What we said on that episode was I thought the Biden administration and Nancy Pelosi told them to pump the brakes, that, that federal wanted to have a say in this.
[00:15:55] Paul Roetzer: And I, and Newsom called like all hands in [00:16:00] California, legislation last week after the election. And so I would not be surprised at all if you don't see states like California race to put some state level legislation in place. I, again, I hadn't really thought about that until now, but I could definitely see that being a major play, to try and bypass some of the stuff that they're, you know, that they know that this administration is going to do too.
[00:16:27] Mike Kaput: Yeah. And while I don't certainly don't expect Silicon Valley to go anytime soon, you could see some really big moves depending on how regulation at the state level happens where some of these companies are built.
[00:16:38] Paul Roetzer: Yeah, for sure. And I mean, Elon's already made that. pretty public with his efforts to build more in Texas, but again, like the SB 1047, the key to that legislation is it doesn't matter if you're in California or not, it's if you do business in California.
[00:16:53] Paul Roetzer: And so, you know, California more than any other state can have an impact on the economy with their [00:17:00] choices. so yeah, I don't know, it's going to be fascinating.
[00:17:04] Mike Kaput: Alright, so next up, we are seeing a bit of an unexpected challenge confronting OpenAI. The pace of improvement in their core AI technology Might be slowing down.
[00:17:20] Mike Kaput: We got a report this past week from the information that the company's upcoming flagship model, which is codenamed Orion, is kind of revealing the limitations of current AI development approaches. So the information is reporting that while Orion does appear to surpass previous models, The improvement is notably smaller than the big leap we saw between GPT 3 and GPT 4.
[00:17:45] Mike Kaput: And some OpenAI employees report that Orion just isn't consistently better at certain tasks, even particularly coding, despite higher operational costs. So, this basically is starting to [00:18:00] challenge one of the fundamental assumptions in AI that we've talked about a bunch, which is Scaling laws. The systems will continue to improve at a consistent rate given more data in compute.
[00:18:11] Mike Kaput: And it seems like that data piece could be the bottleneck here. They are reporting that the amount of high quality training data is creating a real struggle for OpenAI. Apparently they have largely exhausted their publicly available text and data sources. They've started to experiment with AI generated training data.
[00:18:31] Mike Kaput: But, this is creating complexity. The information is reporting that Orion sometimes mimics the limitations of the older models used to generate its training data. So in response, OpenAI is trying a couple things it sounds like. They have established a dedicated foundations team led by Nick Ryder to address the data shortage and explore the future of AI scaling.
[00:18:54] Mike Kaput: They're also looking more at post training improvements, like developing new reasoning [00:19:00] models like O1, that take more time to think before providing answers. So, Paul, I guess my first and biggest question is, are we actually slowing down here? Yeah,
[00:19:13] Paul Roetzer: this is fascinating. So I don't feel like anything in this article was new, but this thing was cooking on Twitter over the weekend.
[00:19:21] Paul Roetzer: Like everyone was reacting to this. There was a bunch of OpenAI employees responding to it. So the co founder and CTO at Writer, which is a, you know, a GenAI writing platform. That we're very familiar with, he tweeted, we've been discussing this for some time. There's minimal improvement or return beyond a trillion parameters with only very small gains from around 150 billion to 1 trillion.
[00:19:49] Paul Roetzer: We have publicly stated that our Palmera LLM achieves improvements by deepening the model architecture, not by increasing the number of parameters. Um. But [00:20:00] then you had other people, like Clive Chan from OpenAI, he's like, what comes next is relatively little new science, but instead years of grinding engineering, to try all the new obvious ideas and the new paradigm, scale it up, speed it up.
[00:20:13] Paul Roetzer: Maybe there's another wall after this one, but for now there's 10Xs as far as the eye can see. So he's like, You know, just the headlines misleading, basically. yeah, Dan Schipper, who's a co founder and CEO at Every. the message that this headline conveys is that odds with what people inside the big labs are actually feeling saying, it's technically correct, but the takeaway for the casual reader.
[00:20:34] Paul Roetzer: AI progress is slowing is exactly the opposite of what I'm hearing. We have Adam. GPT, who's an OpenAI employee, said traditional scaling laws which focus on pre training larger models for longer is absolutely still a thing. That aspect of scale is still foundational. There now happens to be another scaling thing, and together those two things are poised to unlock amazing capabilities.
[00:20:58] Paul Roetzer: Noam Brown, who's [00:21:00] fundamental to the O1 model, we've talked about Noam multiple times, He said there won't be a slowdown in AI progress any time soon. yeah, so, and then like, ironically, Sam Altman did an interview with Gary Tan from Y Combinator. And this is last week. We'll put the interview in the show notes.
[00:21:19] Paul Roetzer: I was watching it on YouTube. I don't know if it's a podcast too or not, but it's definitely on YouTube. And Sam said these things are going to compound. We could hit somewhat unexpected wall or we could be missing something, but it looks like there's a lot of compounding in front of us. We are not near the saturation point.
[00:21:35] Paul Roetzer: The models are going to get so much better, so quickly. And when Gary asked him, what are you excited about in 2025, he said, A GI . So I don't know. Um. So, I think like, there was a couple of things I went back to on this, because again, the article, I guess maybe it was more explicitly saying that it thinks the labs have run into a plateau on the scaling laws, [00:22:00] but I don't.
[00:22:01] Paul Roetzer: I don't know that that's new because, like, we talked recently about, Ilya Sutskever and he was talking about the need to, like, how these labs all know to do, like, more data, more, more trading time. But that there needs to be some new paths where you push hard and that the challenge a lot of these labs were facing was which new path to bet on.
[00:22:28] Paul Roetzer: That there was different ways now to go about trying to drive. like these leaps forward, but the thing we talked about, I can't remember what episode it was on. We'll have to go back and look. We'll put in the show notes. But what I said was, it might've been when we first talked about Orion, that the thing that to me was the trillion dollar question was, could OpenAI create a new frontier model that stayed at the top of the leaderboard for another two years?
[00:22:54] Paul Roetzer: Because when they introduced GPT 4 in March of 23, [00:23:00] Everybody chased that model since then, and it seems like everybody just sort of caught up to that. Like, no one is clearly ahead of a GPT O model, but they're all sort of comparable. And so the question became, well, why hasn't someone taken a leap yet?
[00:23:22] Paul Roetzer: Is it because the training in this way is just kind of like, this is it, this is the smartest models we get? The investments coming from big tech certainly didn't imply that. The demand for NVIDIA chips certainly didn't imply that. So the unknown was, well, has OpenAI or is anybody else? Cracked the next breakthrough that allows them to create a model that is so far ahead of everybody else like GPT 4 was.
[00:23:50] Paul Roetzer: And that's the part that seems up in the air at the moment. Now, as we talked about last week, OpenAI seems to think [00:24:00] reasoning is the key, that their series of O1 models. And again, I'm, I'm a believer that we're going to get the full O1. Sometime between November 21st, when I think that there's a developer day, an OpenAI developer day, and November 30th, when we would be at the two year anniversary of ChatGPT coming out.
[00:24:22] Paul Roetzer: So I think we're going to get the O1 full model. Now, what it's capable of doing beyond what we already see, I'm not sure. but I think that there, there is a real question here about whether or not these frontier models are sort of commoditized at this point, that they have somewhat plateaued in they're just going to keep leapfrogging each other every three to six months, and there isn't a breakthrough really left at that model of text in and text out, but where I think this becomes interesting is like, what are the things that are [00:25:00] going to differentiate as we move forward?
[00:25:03] Paul Roetzer: And the things that seem, because Demis Assabis has said the same thing, that there's like two to three breakthroughs left before we just get to AGIOT. And it, reasoning seems to be a big one, like that, that everybody's working on that. and so O1 is sort of the first one to market that, has that clearly, but we know everybody else is working on it.
[00:25:24] Paul Roetzer: Multimodal seems to be really critical. So again, keep in mind, a lot of these previous generation models were text in, text out, so they were trained on. The text of the internet. They were not trained on videos, audio, images as a single model. But we know that's what Google is trying to do at Gemini. That you train it on multimodal, and then that opens up a whole new universe of data to train these models on.
[00:25:51] Paul Roetzer: And maybe that's a path to go down, is pushing hard on the training there. The other one that comes to mind to me is The path, and maybe these are some of [00:26:00] the things Ilya is thinking about when he's saying like, there needs to be, you got to pick which path to bet on. So you could bet on reasoning, you could bet on multimodal training, you could bet on symphony of models where the frontier model functions as the conductor, and then all the smaller models do their thing within specialized areas that don't require as much compute.
[00:26:17] Paul Roetzer: And so you can imagine pushing hard on This sort of like central hub, which is the frontier model, and then all of this symphony that allows them to kind of work in collaboration together. And then the other one where, where Google has the massive advantage is self play and recursive self improvement.
[00:26:36] Paul Roetzer: And that comes from reinforcement learning, like alphas go and alpha zero and things like that. And so those are four. Again, I have no idea what the next breakthroughs are. But from everything I've heard in these interviews, they seem to think they know collection of things it could be. And what they need to now do is push compute, push different [00:27:00] testing, and see which of these things plays out to unlock these frontier models to maybe then take the leap.
[00:27:07] Paul Roetzer: And maybe it's like a couple of these things in combination. But all that being said, for our listeners, this is all fascinating, but for our listeners, here's the reality. It's irrelevant to you if they make a leap forward next year, like they probably will. But from what Mike and I see every day talking to big enterprises and small is the absorption of the current capabilities is so low that the value you can create in your company using today's models is so significant and so untapped.
[00:27:41] Paul Roetzer: That it doesn't really matter, like, do we get GPT 5 or Gemini 2 or Quad 4, or did their training runs not work? Like, it's all fun to talk about, but for you, focus on using what we have today to do your plans for next year. To [00:28:00] build a more efficient team, to drive productivity and creativity and innovation.
[00:28:04] Paul Roetzer: Like, the next main topic, Mike and I are going to share a way we did this, but like That's my main message to you is don't get caught up worrying about all this stuff, just go do things, like take action, because it's, there's so much value sitting there to be created with the models we already have.
[00:28:23] Mike Kaput: And realistically, we don't need that many more breakthroughs for there to be even more disruption.
[00:28:30] Mike Kaput: Right? I mean, even if it's not as big a leap next time forward, these are still improving. Yeah. And the hospitals are still improving. We're just debating, like, how much.
[00:28:39] Paul Roetzer: Yeah, I think diffusion of the current capabilities would be enough disruption to last us, like, five years. Yeah. So. They're going to get smarter, they're going to get more generally capable, they're going to be able to take actions, they're going to do all these things we talk about, have worldviews, things like that, that, that stuff [00:29:00] may not create value for your company for another year or two, but what we have today can transform your company right now.
[00:29:06] Mike Kaput: So let's maybe then kind of ease into the third big topic and talk about a way to do that. Like we wanted to, you know, you and I had talked before this episode, Paul, and we just kind of wanted to share like a practical use case of how. Even just using the latest capabilities of chat GPT alone last week, we dramatically accelerated planning and innovation for both marketing AI Institute and Smarter X.
[00:29:30] Mike Kaput: So do you want to maybe walk us through what you and I had worked on?
[00:29:35] Paul Roetzer: Yeah, so, so basically the way this works is, you know, as A-C-E-O-A, a lot of my work is on kind of the vision and the high level strategy for the organization. You know, thinking through our revenue channels, thinking through. Our current growth opportunities, future growth opportunities, things like that.
[00:29:52] Paul Roetzer: And so as I was going through building our 2025, like, growth matrix, I sort of landed on this idea [00:30:00] that we'd had a couple years back, and it's related to like the media content side of our business. So Mike, as you're aware, is the chief content officer. So I go to Mike with this idea. I was like, listen, I think we have this opportunity to really scale what we're doing, but to create like tremendous value, like a high velocity of value creation across different industries and for different personas.
[00:30:23] Paul Roetzer: And so I kind of explained the concept to Mike and he's like, yeah, I love it. Like we should do that. And it's like, okay, now what do we do? And so Mike and I spent, you know, what, nine years at a marketing agency together, and we've built plenty of strategies and campaigns. And there's always that, like to go from idea to action is really hard.
[00:30:44] Paul Roetzer: Because someone has to commit the time to build the brief, to go do the research, to do the initial planning, so that you can then react to that together. So, for me, I was like, okay, I gotta put a forcing function in place. I got a bunch of travel coming up, like, I'm just going to [00:31:00] put a meeting for Mike and I.
[00:31:01] Paul Roetzer: It was last Friday, I think we met. And so by putting that meeting on the calendar, it was like, okay, this will force us to now talk about it. But the Thursday, the day before, we had nothing. We had like blank page. It's like, okay, we have this idea, but like, what do we do? And so what I, what I, what I then did is rather than Mike and I showing up to that meeting and spending two hours just kind of bouncing this idea around, I use ChatGPT to develop drafts of the plan, thinking through various planning, production, promotion, performance, like kind of like the key here is we look at different ways.
[00:31:38] Paul Roetzer: And I'm basically just giving some prompts to this thing, but saying like, okay, I've got this business idea. It's for our SmarterX brand. And this, I'm using my co CEO GPT that I built. And so it understands that brand and knows what we do and it knows our revenue channels. And I just said, help me think through this business model.
[00:31:53] Paul Roetzer: And it's like, okay, great. Like, what do you, like, what do you want to do? And so I kind of like had a conversation around that. And then It was [00:32:00] really good. And this is all happening over like three minutes. And then I said, okay, create a task list for the planning and production of the content that we're going to create without getting into all the details.
[00:32:10] Paul Roetzer: And then I was like, okay, you know, we're going to do this for different industries and personas. How do we identify and prioritize verticals and personas, you know, to help drive our decision making? And it created, again, an amazing brief on this thing. Now, as I said, Mike and I did this stuff for a living.
[00:32:26] Paul Roetzer: Like we have the ability to do this, but the reality is for us to do this would probably have taken 10, 20 hours to do the kind of planning that went into this. Now, there was nothing that ChatGPT output that we wouldn't have probably thought about if we had enough time. Like if Mike or I could go away for three days and just think deeply about this idea, we may have come up with 80, 90 percent of what ChatGPT did.
[00:32:54] Paul Roetzer: But the reality is ChatGPT did this in three minutes. And so my thinking here [00:33:00] was rather than Mike and I sitting staring at a blank page and ideating from zero. We now had, I don't know, it's probably like 2, 000 words or so structured really nicely in an outline and brief for us to react to. And so we get into the meeting on Friday and I said to Michael, I was like, let's just walk through what ChatGPT did.
[00:33:20] Paul Roetzer: And let's start talking through things we like, things we don't, if we have any other ideas. And there came a time, what, like, like, 15 minutes into it, we're like, hold on a second. This is actually a really interesting idea. Let's, like, lean into this for a couple minutes. And honestly, it ended up Leading to a conversation that may change, like, our whole go to market strategy for Smart RECs next year.
[00:33:43] Paul Roetzer: And if we hadn't had ChatGPT develop the brief, I don't know that we would have got there. And then I go away, like, this week, and then we run into AI Agency Summit, and then it's Thanksgiving and all of a sudden it's middle of December and Mike and I haven't made any progress. Instead, we're going to spend the next 30 days [00:34:00] pushing hard on the few things that came out of it that were actionable.
[00:34:04] Paul Roetzer: So, that to me is like a fundamental way to use these tools today. Use them as planning assistants. And then, the other thing we did is we traditionally use Zoom, but for this one, you know, I really wanted to try Google Meet. Now, I had had a meeting last week where we did the video, the transcript in the summer with Google Meet, and I was pretty impressed.
[00:34:26] Paul Roetzer: Like, it was, it was really solid. And so we have Google Workspace. So I said to Mike, like, let's try Google Meet for this one. And so we did Google Meet and we used it, did the video, the transcript, the summary, which was great. So the whole thing, Mike and I met for like two hours. Yeah. And I feel like we made a month's worth of progress by just infusing ChatGPT and like that brainstorm process.
[00:34:49] Paul Roetzer: So I don't know. How did you feel about it, Mike? I mean,
[00:34:51] Mike Kaput: you lived through the experience too. Yeah, no, I felt very similarly. What really struck me too is like, Everything you just [00:35:00] described is like if we had a really smart employee like brief us on initial ideas, which then gave us the bandwidth and the time to actually I would argue do what we should be doing and do best, which is actually exploring and operationalizing more ideas or creating our own based on that.
[00:35:18] Mike Kaput: We would never have gotten there. We would have spent all the time getting the initial brainstorming done. And like you said, nothing would have happened as fast as it happened.
[00:35:28] Paul Roetzer: Yeah. One other note I'll make here, and I mentioned this to Mike was, so I did all this in ChatGPT. Then, because we're a Google Workspace shop, I copied and pasted everything because I wanted to share it with Mike.
[00:35:42] Paul Roetzer: Now, I can share the chat, but we wanted to be able to comment on it. We wanted to be able to, like, interact with it. So I needed that ChatGPT output into Google Docs. So you copy and paste, and unfortunately, all the formatting goes away. I mean, I don't know, maybe there's something I'm doing wrong there, but I've tested multiple ways and it just jacks up the [00:36:00] formatting and puts all the pound signs and everything in there.
[00:36:02] Paul Roetzer: And so then as Mike and I are going through, I'm like changing the format, it's kind of annoying. Now that's an, that's an opportunity for Google if you think about it, because direct integration into Google Docs would be brilliant, but I'm not saying ChatGPT per se, like if you use Google Gems, you can export to Docs right away.
[00:36:22] Paul Roetzer: There's actually a button to send it to Docs, like sharing to the Docs. And it does nice formatting. The problem is GEMS seems way behind custom GPTs at the moment. So like, I went in and tried to do the same thing with a GEM. One, I can't find any information about the context window for the instructions.
[00:36:40] Paul Roetzer: I don't know what the character limit is. Like I know in ChatGPT it's 8, 000. I can't find that anywhere for GEMS. There's no guide when you're in there. There, there's no ability to do conversation starters, which is like a really awesome feature of ChatGPT. I don't know how to do those. I think the gems are 100 percent private, they don't train on them.[00:37:00]
[00:37:00] Paul Roetzer: I think if I share them with you, only you can see them. But I have no idea, like, and so just a note to the Google team, like, gems is a huge opportunity, but I, I can't find anything other than like the blog post announcing that they exist in August.
[00:37:15] Mike Kaput: Yeah.
[00:37:15] Paul Roetzer: and then like some user generated guides to it, but even those didn't answer the questions I had about character limits and conversation starters and things like that.
[00:37:24] Paul Roetzer: So I feel like If Google Docs had Gemini baked right in, because if you go into Google Docs and just use the Help Me Write thing, it does not do anything like what ChatGPT or Gemini are capable of doing. So like, I just want the functionality to do my planning right in Google Docs with Gemini without having to go through all this other stuff.
[00:37:47] Paul Roetzer: And that does not exist, but because Google has Google Workspace. There's a big opportunity if they can crack actually making Gems highly functional and valuable and that integration with Google Docs right away. [00:38:00] Now Microsoft obviously has the same capability with ChatGPT and Word, but that would be a great unlock in terms of value creation.
[00:38:10] Mike Kaput: Yeah, and you and I had even tested, obviously, there is Gemini within Docs, but even then, I don't know if it's due to what information it pulls, what model is being used, but like, we tried more sophisticated prompts right in Google Docs, and it's just not there. The same prompt works really well if you go open up your own instance of Gemini.
[00:38:30] Mike Kaput: So it's very limited in
[00:38:32] Paul Roetzer: capabilities in Docs. Right. It is not the same chat, for sure. Maybe that's the, maybe that's the need is like, stop trying to be one thing in one doc and one in the other. It's just integrate the things. So yeah, so the moral story here is push on using these tools as a planning tool for next year.
[00:38:51] Paul Roetzer: Like there, there's a ton of value sitting there and it can really accelerate things. Like again, I always use these examples of what would I pay [00:39:00] to have that function? So forget 20 bucks a month for ChatGPT, for MAICONI it's 40 bucks a month or whatever that number is. If I would have done that myself, we're talking 10 to 20 hours, what is 10 to 20 hours of my time worth?
[00:39:14] Paul Roetzer: Right. A lot to me. So the fact that I didn't have to do that, if that was just like, hey, you can use it to help with this plan, what would you pay for it? I'm probably like, I don't know, 1, 000, 2, 000? Like if some, if a ChatGPT is going to create this for me, that Mike and I could just spend two hours reacting to it instead?
[00:39:30] Paul Roetzer: I would have happily paid thousands of dollars as a business user for that one
[00:39:34] Mike Kaput: use case. Yeah, that's incredible and like we've talked about a couple times, it's like, whether you're trying to figure out more use cases or trying to figure out how to get started, go create a GPT that does, helps you do your job.
[00:39:48] Mike Kaput: Like Co CEO. Just think of it that way. Put your job description in and be working with it regularly.
[00:39:56] Mike Kaput: Alright, let's dive into our rapid fire topics this week. [00:40:00] So first up, we have another interesting case study, not ours this time around, but Visa, the credit card company, has apparently deployed over 500 generative AI applications across its operations, according to a new report in the Wall Street Journal.
[00:40:16] Mike Kaput: These applications span a ton of different functions, from security tools that detect bugs in code, to specialized chatbots serving as subject matter experts. This initiative, led by the company's technology president, Rajat Taneja, reflects a deliberate kind of go fast approach designed to address two critical challenges.
[00:40:36] Mike Kaput: Staying ahead of increasingly sophisticated fraud attempts. And maximizing AI's potential benefits before Visa's competitors do. So they've actually invested over 3 billion in AI and data infrastructure over the past decade to support this vision. one particularly notable implementation targets something called enumeration attacks, [00:41:00] which currently cost the company over a billion dollars annually in fraud losses.
[00:41:05] Mike Kaput: Other tools they've developed to help customers customize billing cycles and streamline various operational processes. And what's really interesting here is just kind of how all in the company has been on this. They established really strong governance infrastructure and data protections first, and then encouraged their teams across all different business functions to participate in AI implementations.
[00:41:28] Mike Kaput: Looking ahead, Tenasia, who is the technology president in charge of this, envisions a future where human employees each oversee Eight to ten different AI powered digital workers creating kind of a hybrid workforce model. So, Paul, this certainly seems like an enterprise succeeding with AI and getting real value out of it, which is a little contrary to some of the reports we've seen saying that nobody is getting value out of this stuff.
[00:41:53] Paul Roetzer: Yeah, we're always on the lookout for brands that are telling their stories of success. They're hard to come by, [00:42:00] honestly. so yeah, this is interesting and it makes sense like a lot on the fraud side that they would have a ton of use cases there, but simultaneously we, we came across a information exclusive for like their pro, subscription with which Mike and I have that are the generative AI spending of 50 companies from Coke to Walmart.
[00:42:20] Paul Roetzer: And so we were kind of scouring through that, looking at seeing like which models they use, what are their use cases? And, you know, I think similar to Visa, some of the things that surface is customer support, like everybody is building chatbots for customer support and success, that, that seemed almost universal, marketing and content generation, which is no surprise, operational efficiency, etc.
[00:42:42] Paul Roetzer: you know, companies like Goldman Sachs and Toyota deploying AI for internal tools. coding is obviously a big one and then sales enablement. Those are categorically like the big things that jumped out from these 50 companies and again, they're, they're big brands. so they were looking at IPG and DoorDash [00:43:00] and AT& T and, Coca Cola I mentioned.
[00:43:02] Paul Roetzer: So yeah, I think going into next year, we're going to start seeing a lot more companies talking publicly. I did think it was interesting how he was sort of wording. around the job part of this. Yeah. You mentioned, you know, the employees managing AI generated digital employees. Why eight to 10 is a number.
[00:43:21] Paul Roetzer: I don't know. I would be interested to see where that's coming from. but he also says we don't invest in AI to displace our talent. We invest in AI to help our employees be more productive, continue to protect consumers from fraud, and to drive constant innovation and payments. That was a spokesperson actually from Visa that said that.
[00:43:39] Paul Roetzer: So, you know, again, I think that the media, every time they're hearing these stories about all this efficiency and gains, they're going to ask the question about jobs. And I think these are kind of like the boilerplate answers we're going to get for a while, that it's not meant to replace them, like it's to give them tools to unlock things and whether that actually, because in that article, they talked about layoffs at Visa.
[00:43:59] Paul Roetzer: [00:44:00] And so it's just, they're trying to kind of head that that's not why the layoffs are happening, which may or may not be true.
[00:44:08] Mike Kaput: Alright, so in some other news, we're seeing OpenAI taking steps to transform their corporate structure. They have entered into preliminary discussions with regular regulators to convert from a non profit to a for profit entity, which we knew was happening, and we're getting reports that they're currently engaged in early talks with both the California and Delaware Attorney's General Offices, marking the beginning of what is likely to be a complex regulatory review process.
[00:44:36] Mike Kaput: Now, This transition is not as straightforward as just changing your status. They have to figure out how to properly value and transfer the company's assets, including its AI technology portfolios. According to OpenAI's non profit board chairman, Brett Taylor, any restructuring would ensure the non profit's continued existence and fair compensation for its current stake in the [00:45:00] four profit entities.
[00:45:01] Mike Kaput: The company apparently plans to become a public benefit corporation, which allows it to, at least on paper, say, maintain its social mission while operating as a for profit business. And like we've talked about on past episodes, this timing here is a bit crucial, because under their recent fundraising terms, the investments that they've raised could convert to debt if the restructuring doesn't occur within a couple of years, within two years of the money being raised.
[00:45:29] Mike Kaput: So, Paul, we've talked about this being kind of one of the core near term challenges for OpenAI. Like, how hard is this going to be for them to pull off?
[00:45:39] Paul Roetzer: Well, I'm kind of like laughing to myself at the moment because I'm thinking if Musk wants his vengeance, this is how you use your newfound influence and power is you find a way for Federal to throw a wrench in all of this.
[00:45:55] Paul Roetzer: that would be the ultimate. Because this is why he was suing them. So this isn't [00:46:00] me just like making up some drama. Musk sued them over this thing that they, that they didn't function within this. And they, you know, the money that was put in, they weren't functioning as this nonprofit. And so like, there's, there's history here and, if they can't do this, they're, they're in a whole heap of trouble if this process doesn't work.
[00:46:20] Paul Roetzer: so yeah, it'll be, it'll be fascinating to kind of follow this along. I mean, obviously they're not going to want information about this process to be leaked out because it's going to be challenging. But, yeah, I feel like this got a lot more interesting now, given Musk's influence on the incoming administration.
[00:46:42] Paul Roetzer: My guess
[00:46:43] Mike Kaput: is this week, the timeline for this accelerated pretty quick. Yeah. They're going to try to,
[00:46:47] Paul Roetzer: I don't think you can do this in a two month period, but they're going to push every button possible, most likely.
[00:46:54] Mike Kaput: Alright, in some other OpenAI news, they have acquired one of the Internet's oldest and most [00:47:00] valuable domain names, chat.
[00:47:01] Mike Kaput: com. So this domain, which now redirects to ChatGPT, is, has kind of an interesting history, because the domain was originally purchased last year by HubSpot co founder, Dharmesh Shah, who we've talked about before. For over 15 million, that makes it one of the highest priced domain sales ever publicly reported.
[00:47:23] Mike Kaput: Now, Shah also revealed this past week that OpenAI was the unnamed buyer he had talked about selling the domain to, and he suggested that he may have received payment in the form of OpenAI shares. So chat.com is now totally in the hands of OpenAI. I mean, Paul, you know, Dharmesh, given your background, you own HubSpot's first ever partner agency, like, seems like a bet for him that kind of paid off here.
[00:47:49] Paul Roetzer: Yeah, it was. I thought it was hilarious. 'cause Sam literally just tweeted chat.com and on November 6th and then like the tech world just like went crazy and Dharmesh posted the [00:48:00] story of how it kind of happened on LinkedIn and on x. And, yeah, he implied, like, You know, he actually used the O1 reasoning model, he gave a prompt, like how much did Dharmesh sell it for?
[00:48:13] Paul Roetzer: And then it was like, he gave a prompt you could use in O1 to try and figure it out. As far as I know, they didn't disclose anything. My guess is he didn't sell it for much of a profit, if any, he, he likely just exchanged it for OpenAI shares. Dharmesh is, is a notorious, domain name collector. I have an issue with, shoes.
[00:48:36] Paul Roetzer: Like I like to buy Nike shoes. I think Dharmesh buys domain names. Like, I don't know how many he owns, but my guess is it's in thousands. so yeah, this is, I think for Dharmesh, like he said in his LinkedIn post, like he doesn't need the money, like he's doing just fine. So I think this is just like fun for him.
[00:48:55] Paul Roetzer: and I think the fact that. OpenAI is using it, [00:49:00] you know, gives Dharmesh some enjoyment and hopefully he, you know, made out with some OpenAI shares along the way. I mean, Dharmesh is one of my favorite people in the world. He's, he is, one of those few people who lives up to your expectations when you actually get to meet someone in person.
[00:49:14] Paul Roetzer: he has always been, he's a wonderful, wonderful person for me, for, for my business. I've known him going back to 2007. so yeah, if you don't know Dharmesh, it couldn't happen to a better guy.
[00:49:27] Mike Kaput: And if you have a, an idea for a hot domain name you want to buy, maybe check with Dharmesh first. Yeah, he probably already got it.
[00:49:36] Mike Kaput: Alright, so next up, Perplexity appears to be on the verge of securing a 500 million investment round led by institutional venture partners, IVP. that potentially values the company at 9 billion. So this represents literally a tripling of the company's value from its previous funding round earlier this year.
[00:49:57] Mike Kaput: It would mark Perplexity's fourth funding round [00:50:00] in 2024 alone. IVP was a significant backer already. They led Perplexity Series B. honestly, Perplexity's growth trajectory, Paul, seems kind of nothing short of explosive. Like, we've talked about, you know, in comparison to Google, It's still very much a small fish in the search market, but this certainly seems like they're on the right trajectory.
[00:50:21] Paul Roetzer: Yeah, they got to get out of their own way. They're still making a whole bunch of like PR missteps. And, you know, I think Arvind has said he's, you know, learning lessons as a CEO. Like they've done some very questionable things from a business practice perspective, in my opinion. I love the technology. I'm not a huge user of Perplexity, but I feel like they're one of those like rocket ship startups that needs.
[00:50:46] Paul Roetzer: probably needs some, I don't know if this is the right way to say it, but like adults in the room. Yeah. You know, in the early days of Google and things like that when these Huge explosions of growth happen, like you got to go get some people who know what the hell they're doing [00:51:00] and don't, have so many just self inflicted wounds.
[00:51:06] Paul Roetzer: So I hope that they figure it out and they keep growing and they, modernize the look and feel as I mentioned last week, I think, you know, ChatGPT search sort of made perplexity feel obsolete to me as a, from user experience. So, you know, I hope it keeps going and competition is good, in the search market.
[00:51:26] Paul Roetzer: I think it's, you know, pushing Google to, to think in a more innovative way and Microsoft and others. yeah. So I would, a couple of things that would not surprise me in 2025 is an adult, like a seasoned leader is brought in to help, you know, keep, keep things moving in the right direction. if the regulatory stuff, this, this is the other thing is like, I didn't think about this until this moment.
[00:51:51] Paul Roetzer: acquisitions might heat back up with the Trump administration because there, there's so many issues right now from a [00:52:00] regulatory perspective that someone like, oh, Google or Microsoft, I don't think they could ever get an acquisition of perplexity through in today's environment. Yeah. But, come next summer You know, maybe perplexity starts becoming a really interesting acquisition target if, if things cool off from a regulatory perspective on acquisitions in the tech space.
[00:52:19] Paul Roetzer: I don't know.
[00:52:20] Mike Kaput: Well, perplexity is also in the news, not for as good a story as we just covered. Exactly to your point, because perplexity CEO Aravind Srinivas has stepped into also the middle of a labor dispute at the New York Times. Time's tech workers went on strike over wage increases and workplace policies, and at the same time, Srinivas publicly offered his company, AI company's services to the newspaper, which sparked some backlash.
[00:52:53] Mike Kaput: This was right around just days before the U. S. presidential election. Time's publisher A. G. Sulzberg expressed concern [00:53:00] about the strike's impact on election coverage. And Perplexity's CEO responded directly on social media, offering Perplexity services to ensure coverage remained available through the election period.
[00:53:12] Mike Kaput: Now, you know, this is, he kind of tried to later attempt to clarify this was an offer for merely technical infrastructure support, but some people pointed out that that's exactly what the striking workers were providing. So, Kind of put his foot in it a little bit. Like Paul, how seriously should we be taking this?
[00:53:31] Mike Kaput: Is it just like bad timing, bad communication? Is there more going on here?
[00:53:36] Paul Roetzer: yeah. I mean, I think it's just like I was saying, it's a tech CEO isn't seasoned in this stuff. I don't think it was a strategic PR move to get the publicity, like just to do it. I think he thought it was a clever idea and a way to get some attention.
[00:53:53] Paul Roetzer: And probably doesn't think through the ramifications to the brand and things like that. I, [00:54:00] again, I don't know him. I don't know the company deeply, but like from observing them for the last 12 to 18 months, pretty closely, there's a lot of this stuff where it's just like, if you had someone else in the room who's been through these things before, it could be a head of communications.
[00:54:18] Paul Roetzer: It could be a president. It could be someone on the board, like whatever it is. It's. There needs to just be somebody there that, that, that's helping along to avoid stuff like this. Right. it's just sometimes these self inflicted wounds are so obvious what the outcome's going to be. And again, I don't think he wanted that backlash.
[00:54:39] Paul Roetzer: Some people do this stuff for the backlash, like Elon notoriously will do things just to cause the backlash. I don't think Aravind is, is that type of CEO yet. I don't know. I think he just thought it was a good idea and it ended up not
[00:54:56] Mike Kaput: being a great idea. Yeah, and clearly, like, [00:55:00] we're also seeing that, from a narrative perspective, people are extremely sensitive to anyone related to AI trying to And that perhaps is why it's potentially harming human workers.
[00:55:11] Mike Kaput: That's a very real concern that people have.
[00:55:16] Mike Kaput: Alright, so next up, Anthropic has unveiled Claude 3. 5 Haiku, which is an upgrade to their fastest AI model, and it now matches the abilities of larger models while maintaining its speed. It has very rapid response times. And it turns out Claude 3. 5 Haiku surpasses Claude 3 Opus on many intelligence benchmarks, even though it is optimized directly for speed.
[00:55:40] Mike Kaput: The model is being rolled out across multiple platforms, including Anthropx API, Amazon Bedrock, Google Cloud's Vertex AI as well. So, the pricing structure appears to reflect Anthropic's push for widespread adoption. It is 1 per million input tokens and 5 per million [00:56:00] output tokens. And there are potential cost savings of up to 90 percent through techniques like prompt caching and additional savings via Anthropic's Message Batches API.
[00:56:12] Mike Kaput: Now companies like Repl. it have noted substantial improvements in code related tasks, including tasks including reductions in errors and improved reasoning capabilities. So, Paul, this is kind of related to one of our main topics, like while there may be some issues with just how much progress we're seeing between each model generation, it sure seems like we're getting way better models for way cheaper still.
[00:56:39] Paul Roetzer: Yeah. And the, the implications to the average listener, the non developer, non technical listener here is the models are getting more powerful. They're getting way cheaper to build things on and to, to run. and so people are going to innovate and build more and more applications that you're going to be able to use in your [00:57:00] industry.
[00:57:01] Paul Roetzer: or if you're at a bigger enterprise and you can go build things, it's getting cheaper and cheaper and it's going to continue to get cheaper for your developers to build things for you. it's kind of, again, like, and this is just going to keep happening. So, as these big frontier models we're talking about.
[00:57:17] Paul Roetzer: The reality is like your company may more likely to use like a 6 to 12 month old version of something or the smaller version of the current one as confusing as all this is. to build things to achieve what you want to achieve without having to pay for like the big model.
[00:57:35] Mike Kaput: So some other anthropic news, apparently Amazon is in discussions to make another massive investment in Anthropic.
[00:57:43] Mike Kaput: They, that follows their initial four billion dollar commitment last year. However, the new deal comes with some strings attached. Amazon wants Anthropic to use servers powered by its own Tranium chips, while Anthropic prefers Amazon servers that use NVIDIA's [00:58:00] AI chips. Now, for Amazon, getting Anthropic to adopt its chips could reduce dependency on NVIDIA hardware, so it makes sense for them.
[00:58:08] Mike Kaput: for Anthropic, the decision could affect its flexibility to use multiple cloud providers, Or it operate its own data centers in the future. Now, Paul, can you kind of maybe unpack for us the dynamics here? On one hand, it kind of, you know, sounds a little bit like some inside baseball, but it does matter if certain companies are tying people into their ecosystems as part of their funding agreements.
[00:58:35] Paul Roetzer: Yeah, I don't know. On the surface, I feel like there's just got to be a lot more to this story and like how this partnership would work. I can't see Anthropic locking themselves in like 100 percent to something like this. And I'm sure there's, again, it's, there's probably a whole bunch of other details.
[00:58:51] Paul Roetzer: But I, a more interesting thing is like Amazon's already put like 4 billion into Anthropic as is, or four and a half billion. So, I mean, we're talking about almost like [00:59:00] 9 billion, now we're starting to get into the Microsoft OpenAI range where they've put like 13 or 14 billion into OpenAI, and again, like, go back to what I just said about regulation and acquisitions and stuff like that because, you know, when, when I mentioned all the executives that were tweeting, You could throw Jeff Bezos in that group who tweeted big congratulations to our 45th and now 47th president on an extraordinary political comeback and decisive victory.
[00:59:30] Paul Roetzer: No nation has bigger opportunities. Wishing at Real Donald Trump all success in leading and uniting the America we all love. So, you know. If regulations start to come down and Amazon sees an opportunity to make Alexa actually work again. And I don't know, like, that's one of the things I'm just really anxious to watch is like how this all plays out.
[00:59:49] Paul Roetzer: But they would be at, again, eight and a half billion dollars invested in a company that's Supposedly valued at what, 40 billion is the rumor. So you're talking about a 20 percent equity stake in [01:00:00] one of the frontier model companies. That's, that's not chump change.
[01:00:06] Mike Kaput: In other news, a robotic software company is raising some eyebrows after securing a massive 400 million in early stage funding.
[01:00:15] Mike Kaput: So this company is called Physical Intelligence, and they just raised this money from some pretty significant tech industry heavyweights, including Jeff Bezos. OpenAI and prominent venture capital firms. So this investment actually values this startup at two billion dollars and basically what they're trying to do is create foundational software that can work across any robot platform.
[01:00:41] Mike Kaput: Their flagship software is called PiZero, and it has already demonstrated some interesting capabilities, like it has successfully enabled robots to perform everyday tasks like folding laundry, bagging groceries, and handling kitchen operations. So, this is kind of coming as we get [01:01:00] people like Elon Musk forecasting we're going to have literally billions of humanoid robots in the next couple decades.
[01:01:07] Mike Kaput: So, Paul, like, this is a pretty big early stage funding round. Like, how closely should we be paying attention to this company?
[01:01:15] Paul Roetzer: Well, I mean, obviously with the investors they have, they're worth paying attention to. I think the bigger story here is just, like, robotics is, is going to be a major, major, area of investment and progress.
[01:01:25] Paul Roetzer: You're going to see probably a whole bunch of really cool demonstrations going into 2025. They're not going to be. wide scale, it's not going to be mass market, but you're going to start to see really, really impressive stuff, especially as these multimodal language models are embedded in like basically the brains of these robots.
[01:01:44] Paul Roetzer: And the hardware keeps making progress from, you know, NVIDIA to Boston Dynamics to, you know, Figure, which, right, or, oh, Amazon has a big investment in Figure, if I'm not mistaken. So I just, this is going to be a huge area to pay attention to, and [01:02:00] sometime later this decade, you'll, you'll, you'll see it.
[01:02:02] Paul Roetzer: Start to see these things really making an impression from a commercial perspective, like actually having commercial value and being productized, but we're a ways away from that still.
[01:02:13] Mike Kaput: So next up, OpenAI has secured kind of an interesting legal victory. A New York federal judge has dismissed a lawsuit filed by news outlets Raw Story and Alternet.
[01:02:25] Mike Kaput: that challenged OpenAI's use of their articles to train its AI models. Now, this case focused not on direct copyright infringement, but on the removal of copyright management information from articles used in AI training. So, Judge Colleen McMahon, while dismissing this case, left the door open for the outlets to file an amended complaint, though she actually expressed skepticism about their ability to demonstrate sufficient legal injury.
[01:02:54] Mike Kaput: So this comes obviously as OpenAI has a number of open lawsuits, including [01:03:00] one from the New York Times, which sued them in December 2023. And the judge's decision here actually highlighted a crucial distinction in the case that could have implications elsewhere. She noted that the real issue at stake isn't about copyright management information, But rather about compensation for the use of articles in AI development.
[01:03:21] Mike Kaput: So, Paul, when we're looking at this, obviously, as always, we are not lawyers, but how big a deal is this for OpenAI?
[01:03:29] Paul Roetzer: Yeah, it seems like this is potentially a really important step. Again, like, we got to talk to our IP attorney friends, but, the one thread I was looking at said, you know, called out a few key points.
[01:03:43] Paul Roetzer: Facts on which LLMs train are not copyrightable. That seems like a really important distinction. Gen AI models synthesize, they don't copy. Datasets they're trained on are vast, so no one, one work is ever likely to be plagiarized. And regurgitation is, in quotes, By an [01:04:00] early LLM version is irrelevant if current versions won't do it.
[01:04:03] Paul Roetzer: Those are four very interesting notes from the finding. So, yeah, I don't know. I mean, this is We weren't expecting to see case law kind of, like, emerging this soon. And I wonder how much of an impact this one might have. And that's actually another area to consider. with the next administration is what impact that has on the U.
[01:04:27] Paul Roetzer: S. Copyright Office's review of, you know, these models and how they're used and how they're trained. And if, if we don't maybe see some acceleration of changes to copyright law as a result of this because that would allow more innovation. And we know that's what they're going to want to do.
[01:04:43] Mike Kaput: And these four points you mentioned appear to me to be literally the foundational logic people like the New York Times are using saying their stuff was stolen.
[01:04:53] Paul Roetzer: Yeah, again, I don't know how this stuff works, but it sure seems to imply that you could see some other cases thrown out if [01:05:00] this holds up.
[01:05:02] Mike Kaput: Alright, our last piece of news this week, Paul, is that Google accidentally, it appears, leaked details of its new AI assistant, which is called Jarvis. Of course, it's Jarvis.
[01:05:14] Mike Kaput: Of course, it's Jarvis. Everybody names everything Jarvis. So this happened in the Chrome Web Store and there was like this premature reveal that kind of showed off a few things about what Google's thinking for Jarvis. Jarvis is able to actually take direct control of web browsers to complete everyday tasks, at least according to the leaked description in the Chrome Web Store.
[01:05:37] Mike Kaput: So according to that description, which I believe is now taken down, The AI assistant can independently handle activities like purchasing groceries, booking flights, and conducting research. Google appears to have removed this store page. This was originally planned to be unveiled in December. Paul, regardless, it sounds like we're about to get a possibly competent AI agent [01:06:00] from Google.
[01:06:01] Paul Roetzer: My guess is we're about to get an impressive demo of an AI agent from Google. Like, I have said it many times on this, like, I just, I To use a tool like this, the computer use that Anthropic showed, OpenAI's got the same thing that everybody's working on this. They are not precise, they are not reliable, they have massive security concerns, because you have to give them access to your credit card or your bank account or all these apps.
[01:06:27] Paul Roetzer: I just don't see This being transformational technology in 2025. I think we're going to get a ton of demos, there's going to be a bunch of hype, there's going to be a bunch of overreaction from media and influencers online who are like, game changing and blah, blah, blah, and this is the end of the world. Like, it, it's just, it's just progress being made on an inevitable technology.
[01:06:53] Paul Roetzer: That may still take a year or two before it becomes mainstream. That's kind of like high level how I think about all this stuff. [01:07:00]
[01:07:00] Mike Kaput: Gotcha, so maybe temper our expectations a little.
[01:07:03] Paul Roetzer: Yeah, I, they may show something and it may look amazing and so did Sora eight months ago. And what, as I say that, they're probably going to release Sora in like the next three weeks.
[01:07:12] Paul Roetzer: But like, you get my point, like it's, and even, even though when they release it we're going to have, be able to get 10 seconds, not 60 seconds and things like that. So, this is how this stuff works, like you show these cool demos, you talk about it, And then like, you know, maybe a year or two, like it's actually reality.
[01:07:27] Paul Roetzer: because ChatGPT changed how this stuff works. They dropped ChatGPT on the world and it changed everything. Now we're in this preview everything and then don't release it for eight months, 12 months, 16 months, whatever it is, or release some early version of it, and that's kind of where we find ourselves in this AI timeline is everyone races to release demos of things that don't actually work.
[01:07:51] Paul Roetzer: work or aren't really ready for us to use. So, yeah, I think you just got to like have realistic expectations, but it's probably a really impressive [01:08:00] tech.
[01:08:01] Mike Kaput: All right, Paul, that's a wrap for this week. We'll have a lot of material to cover. Appreciate you as always, demystifying everything for us, giving us all the context we need.
[01:08:11] Mike Kaput: As a couple quick housekeeping reminders. If you have not already, please go subscribe to our newsletter, marketingaiinstitute. com forward slash newsletter. It's called This Week in AI. It gives you all the news you need to know each and every week in a very digestible format. Also, if you have not already and have the ability to, please leave us a review.
[01:08:32] Mike Kaput: We really appreciate all the feedback and it helps us make the show better. Paul, thanks so much.
[01:08:38] Paul Roetzer: Good stuff, Mike. We will be back next week. We'll figure out if we're doing a Thanksgiving week episode after that, but next week we will be back. and, final call, AI4Agencies. com. If you are an agency or you use agencies, it'd be a great event for you to attend.
[01:08:55] Paul Roetzer: All right. Thanks, Mike. Thanks, Paul. Thanks for listening [01:09:00] to The AI Show. Visit marketingainstitute. 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, taken our online AI courses, and engaged in the Slack community.
[01:09:21] Paul Roetzer: Until next time, stay curious and explore AI.