SCOOP: Google Testing Gemini With Partners -- Launch Soon?

0:01Today on the AI Breakdown, we're looking at reports that Google is on the verge of releasing their Gemini model. Before that on the brief, big market excitement continues around artificial intelligence. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our Discord, our newsletter, and our YouTube channel.

0:26Hey friends, one quick thing before we get into today's show. I'm exploring a number of different ways to extend our content products into the realm of educational with a goal to really help people move from just learning about the news in AI to figuring out the tools in AI to help them reskill, start new companies, figure out how it's going to evolve their careers, and generally feel more empowered to go take on this new AI powered world. But I need some feedback before we actually launch some of those new products. I have a survey that'll take you about one minute to fill out at bit.ly slash AI breakdown survey.

1:03It's just going to ask you a couple questions about what you're most interested in learning, and I would so appreciate it if you would take the time to go take that survey. Hopefully, it'll turn into the exact type of educational and resource content that you're looking for. So thank you, thank you in advance. And again, that link is bit.ly slash AI breakdown survey. Now let's talk AI news. Welcome back to the AI breakdown brief, all the AI headline news you need in around five minutes. Today, we are starting off with a look at market enthusiasm surrounding AI expressed in both public and private markets yesterday.

1:39One of the big stories and really one of the big things that market observers were looking to as a bellwether for how much enthusiasm there remained for AI stocks was the IPO of Arm. Arm is an AI chip company that we've talked about on previous episodes who doesn't manufacture chips themselves, but sells plans for chips that others can manufacture. Now, one thing that's important to keep in note is that tech stocks in general, and IPOs specifically, have had a rough go of it for the last 18 months or so. That is, of course, driven by the fact that we've had the fastest rate-hiking cycle in history following an extended 12-14-year period of near-zero interest rates.

2:14And yet, that AI enthusiasm did indeed continue on. ARM was up 25 % from its initial offering price in the first day of trading, and the IPO was 12 times oversubscribed. Now, if you were watching or listening yesterday, you heard me talk about Goldman Sachs' recent report, Why AI Stocks Aren't in a Bubble. And I guess depending on your priors, this could either confirm or deny Goldman's thesis that we are not. For those who didn't hear that episode, basically Goldman's argument is that although, yes, there is a ton of enthusiasm around artificial intelligence, if you look at the P.E. ratios of previous tech bubbles, the top companies benefiting from AI enthusiasm aren't anywhere near the P.E.

2:52ratios of companies at the top in previous bubbles. Interestingly, Goldman also pointed out that the quote-unquote early winners were this set of infrastructure companies, including quote, makers of semiconductors needed to build AI technology and cloud service providers with the computing infrastructure to commercialize it. They noted that those near-term AI beneficiaries were up much, much more than long-term AI beneficiaries in the first eight months of 2023. Arm would definitely fall into that category, and so I think in many ways the IPO performance yesterday adds evidence to Goldman Sachs' thesis.

3:22Now, private markets were no less active, and there were at least two notable deals. The first was Databricks, which raised an additional half billion dollars at a$43 billion valuation. According to CEO Ali Goetze, the new round, which was technically a Series I, I believe, was sparked by inbound interest from NVIDIA. Ultimately, T. Rowe Price led the round and was joined by NVIDIA, Capital One, the Ontario Teachers Association, Andreessen Horowitz, and many, many more. So what does Databricks do? Basically, they help companies organize all of their data, which is, of course, a prerequisite for getting the most out of artificial intelligence.

3:58More recently, they've been helping companies use that data to train their own custom LLMs or to take advantage of and customize open source models that exist more widely. A month ago, the company made news when a tie-up with Microsoft came out, through which Microsoft's enterprise Azure platform would offer access to Databricks in a move that many saw as competitive with Microsoft's other big investment in the AI space, OpenAI. Now, to many observing Microsoft's behavior, it reflected the reality that the AI space is extremely fast-moving and dynamic, and it appears like enterprises might be a little bit more focused on getting custom solutions that work with their data without having to rely on entirely separate services like OpenAI, and this was a way for them to adapt to that reality.

4:38Around this new funding, Databricks CEO Ali Goetze again talked about that exact weird relationship. Goetze said, it's the nature of the business. We're both a customer and a competitor, but all of the clouds would rather have our business than not. It's a land grab right now in AI. Now, certainly given this funding and Databricks hotspot in the AI markets, many are looking at them as a promising IPO prospect as well. However, currently reports are that the company is burning a huge amount right now. Indeed, they appear to be in that old world zero interest rate mindset of growth at all costs rather than profitability.

5:09Although Goetze did say that his company could flip to profitability within two quarters if it felt the need to do so. Another startup turning heads with its fundraise is a European startup called Helsing. The company raised a$223 million Series B led by General Catalyst, and the round was also joined by Swedish heavy industry and defense group Saab. Now, this company is notable for a couple reasons. First, it's backed by Spotify founder Daniel Ek and represented part of his pledge to put his money to work funding deep tech moonshot projects. He had previously funded Helsing with a 100 million euro investment.

5:41Now, in terms of what this company does, as TechCrunch puts it, its AI platform claims to be able to boost defense and national security specifically for liberal democracies by making them more efficient using live data. By way of example, quote, in June this year, the German government selected Helsing and Saab to provide AI-enabled electronic warfare capabilities for the Eurofighter jet fighter. And this month, Helsing and its consortium partners won the contract to provide the AI backbone for the Future Combat Air System program. There is so much that we could dig into here about what trends this reflects.

6:09Of course, on the technology side, it's an AI company, but it's also an AI company in the defense of the military space, which is something that we are seeing just a huge amount of. What's more, I do think it also shows geopolitical changes. In the wake of Russia's invasion of Ukraine, obviously the European military establishment has gotten a very different perspective on its own existence and has really started to ramp up investment, and this appears to be an example of that. One more fundraise for the sake of completeness, Pixies, a company that promises an AI-powered platform for brands to monitor and orchestrate their marketing campaigns, also announced an$85 million Series C round.

6:43Now, I've talked before about how marketing is just one of the most obvious areas for AI. It's an area that benefits from understanding and taking advantage of huge amounts of data. It's an area that involves creating a ton of content. And candidly, it's an area that already has made lots and lots of trade-offs between creativity and artistic integrity, and raw efficiency in salesmanship. That's not a critique, that's just saying that marketing is an area that is going to have less scruples, let's say, about adopting AI than perhaps some other fields will. Speaking of fields that perhaps have a little bit more of a challenge with the adoption of AI, the strikes in Hollywood continue, but an interesting new announcement this week shows that in addition to union negotiations, part of the path forward may be new technology companies and different business approaches to AI as well.

7:27On Thursday, a company called Metaphysic announced a partnership with CAA, through which it claimed that CAA's talent, such as Tom Hanks, Rita Wilson, Paris Hilton, and many others, will effectively be able to manage how third parties use AI versions of their likeness, including their face, voice, and performance data. Now, the announcement is fairly scant on details, but they say that the platform includes consent, compensation, issues of copyright. It basically represents a way for celebrities to try to take hold of AI representations of themselves and benefit from them rather than be cut out and relegated to the sidelines.

8:00Now, I don't know the specifics of metaphysics yet, and obviously there remain big questions about how the union negotiations resolve. But at the same time, it does feel inevitable in any entertainment sector that part of the answer at least is going to be sanctioned AI platforms and sanctioned AI uses that cut the people whose likeness or talent or voice or face is being used into the economics of that usage as a new opportunity and revenue stream. I think it is by far the most likely outcome that this is what we see in music as well, with some amount of bifurcation between actual artist-created music and AI-generated music using artist-voiced likeness that still goes through some whitelist or blacklisting process and gives those artists the ability to make money just from existing.

8:41Lastly today, one more area where the AI future is a little bit controversial. One might think that the gaming industry is completely ripe for not disruption by AI, but to leverage AI for really innovative and new types of experiences. Obviously, we've seen efforts around non-player characters using ChatGPT for more realistic dialogue. There are really interesting companies out there that are making it much easier to create 3D worlds. But on the flip side, you have companies like Valve, who through their Steam platform seem to be taking a hardline stance against AI-generated content in games. Well, Bain Consulting has come out with a report in which they suggest that while less than 5 % of video game content today is developed with generative AI, within 5-10 years, an entire 50 % of video game content will be developed with support from generative AI.

9:26This is getting a lot of chatter on Twitter and X, but frankly, I'm taking the over on this one. I think that the ability for game designers to create these incredibly immersive worlds, to have more realistic interactions with people, all of the opportunities basically of AI are going to so radically outweigh the pressure to go the other direction. I just think it's going to become a huge tool in the games industry, in spite of these furtive efforts to take some sort of ethical stance against it. But I'm by no means a games expert. I could be wrong. In any case, that is going to do it for today's AI Breakdown Brief.

9:58I hope you enjoyed this, and I'll be back soon with the main AI breakdown. Welcome back to the AI Breakdown. We've got a fun one for your Friday. One of the things that everyone is paying attention to is the competition between the biggest players, Meta, OpenAI, Google, around the future of LLMs, and in particular, whether anyone is going to exceed the capabilities of GPT-4. This week, we got yet another scoop from the information that suggests that the release of Google Gemini is right around the corner. And this is something that has been getting buzz for weeks now. On August 27th, Semi-Analysis, which is the same blog that published that Google has no moat memo from earlier in the year, wrote a piece called Google Gemini Eats the World.

10:45Gemini smashes GPT-4 by 5x the GPU pores. Basically, this piece made a very simple argument, which is that Google's Gemini has access to more compute than anyone else. They called Google the most compute-rich firm in the world and said basically that they had 5x the computing resources that GPT-4 did. Now, if you need evidence that this post touched a nerve, OpenAI's Sam Altman, usually a pretty reserved guy, took to Twitter a couple days after its release saying, incredible Google got that semi-analysis guy to publish their internal marketing and recruiting chart, lol. However, this came off exactly as it seems like it came off, as defensive and nervous.

11:26Indeed, Elon Musk responded, are the numbers wrong? To which the author of the piece Dylan Patel wrote, they are correct. More than anything, this conversation reflected just how much attention there is on this particular battle. And this was the environment in which we got a report from John Victor at The Information yesterday, called Google Nears Release of Gemini AI to Challenge OpenAI. The post reads, Google has given a small group of companies access to an early version of its highly anticipated conversational artificial intelligence software. Giving outside developers access to the software known as Gemini means Google is getting close to incorporating it in its consumer services and selling it to businesses through the company's cloud unit.

12:04The piece reinforces that Gemini is meant to be multimodal. It writes, Gemini comprises a set of LLMs which can power everything from chatbots to features that summarize text or generate original text, such as email drafts, song lyrics, and news articles based on descriptions of what users want to read. Gemini is also expected to help software engineers write code and to generate original images based on what users ask to see. Mainly, though, the article points out just how high the stakes are for Gemini's launch. Quote, Google is banking on the software to power everything from its barred chatbot to new features in its workspace software in addition to boosting its cloud server rental business.

12:38But is there anything in here that gives us more information about how Gemini might actually compete? Well, the piece writes, quote, Gemini has an advantage over GPT-4 in at least one respect, said a person who has tested it. The model leverages reams of Google's proprietary data from its consumer products in addition to public information from the web. As a result, the model should be especially accurate when it comes to understanding users' intentions with particular queries, and it appears to generate fewer incorrect answers known as hallucinations. Now, I don't know if the person referenced in this was Brian Romley, but Brian Romley did quote tweet this piece and said earlier today, I have been testing a version of Google's Gemini and find it very interesting.

13:19It is equivalent to ChatGPT4, but with newly up to the second knowledge base. This saves it from some hallucinations. The information also got some info about how Google plans to release it. Apparently, they are going to give companies access to it through their Google Cloud Vertex AI service and will in fact release different size versions so developers can customize it to what they need it for. For example, paying for a less sophisticated version to handle simple tasks and even one that's small enough to run on a personal device. According to the sources for this article, Google is currently giving developers access to a relatively large version of Gemini, but not the largest version it is developing, which would be more on par with GPT-4.

13:56So all in all, only a little bit new in here, but confirmation of some things that we expected, and an interesting reminder of this advantage that Google has, taking advantage of the proprietary data that it already has access to from its consumer products. This is one of the most interesting areas of potential differentiation. Earlier this week, we discussed how Apple's release of the A17 chip, and the way that they're using it with the new double-tap watch features, gives an indication that they are trying to also take advantage of their comparative advantages, which is their focus on running software on device rather than in the cloud, to increase privacy and improve performance.

14:31That could obviously translate to a very unique approach for AI models, and it seems like Google accessing its proprietary data from its consumer products is a version of that as well. Now, while he might not have talked extensively about Gemini itself, Google CEO Sundar Pichai has talked a lot about AI more generally lately. Google just had its 25th anniversary, and in part to commemorate that, Pichai wrote a memo to Googlers around the world. Unsurprisingly, AI is at the very center of how they see the future. In a section of the memo titled A Healthy Disregard for the Impossible, he writes, Google has been investing in AI since almost the beginning.

15:07We were one of the first to use machine learning in our products starting in the early 2000s for spelling corrections, improving the quality of ads, and showing suggestions and recommendations. Pichai talks about the first time he went and saw a demo of a neural network in practice in 2012, and said, it was the first moment I thought to myself, this is really going to change everything. He goes on, I had a similar feeling when I saw the groundbreaking interdisciplinary research happening at DeepMind focusing on understanding the nature of intelligence. This progress deeply influenced my thinking when I became CEO in 2015 that Google should pivot to be an AI-first company.

15:40Now, in addition to going through the products that they've released, Vichai talks about the societal implications as well. He writes, as excited as we are about the potential of AI to benefit people in society, We understand that AI, like any early technology, poses complexities and risks. Our development and use of AI must address these risks and help to develop the technology responsibly. The AI principles we launched in 2018 are an important part of how we do this. These principles prompt questions like, will it be helpful to people and benefit society, or could it lead to harm in any way?

16:09Now, this is interesting because one of the big things that sped up the conversation around AI safety this year was Turing Award winner Jeffrey Hinton leaving Google and arguing that it, among other companies, was getting more reckless in how it was thinking about releasing AI models because of the pressure of market competition. To the extent that he feels that way, Pichai has certainly tried to play this down. Insider reports Google's CEO says he isn't worried about catching up to open AI. Quote, I feel very comfortable about where we are. Moving back to this 25th anniversary memo, the looking ahead section is basically entirely about AI.

16:43He writes, as we look ahead, I've been reflecting on the commitment from our original founder's letter in 2004 to develop services that improve the lives of as many people as possible to do things that matter. With AI, we have the opportunity to do things that matter on an even larger scale. He points out that a million people are already using generative AI in Google Workspace, that their AI-powered flood forecasting now covers places where 460 million people live, that a million researchers have used the AlphaFold database, which covers 200 million predictions for protein structures, and he says, quote, we've demonstrated how AI can help the airline industry to decrease contrails from planes, an important tool for fighting climate change.

17:19Still, he writes, and this is really the money shot, there is so much more ahead. Over time, AI will be the biggest technological shift we see in our lifetimes. It's bigger than the shift from desktop computing to mobile, and it may be bigger than the internet itself. It's a fundamental rewiring of technology and an incredible accelerant of human ingenuity. Making AI more helpful for everyone and deploying it responsibly is the most important way we'll deliver on our mission for the next 10 years and beyond. AI will allow us and others to ask questions like, how could every student have access to a personal tutor in any language on any topic?

17:53How could we enable entrepreneurs to develop new forms of clean energy? What tools could we invent to help people design and create new products and grow new businesses? Et cetera, et cetera, et cetera. He concludes, as these frontiers come into view, we have a renewed invitation to act boldly and responsibly to improve as many lives as possible and to keep asking those big questions. Our search for answers will drive extraordinary technology progress over the next 25 years. And in 2048, if somewhere in the world, a teenager looks at all we've built with AI and shrugs, we'll know we succeeded.

18:22And then we'll get back to work. Now, I think that the takeaway from this is that as much as it's easy and exciting, frankly, to get caught up in the AI race, the model competition and what Gemini might mean, these companies are making long-term longitudinal bets that AI is going to remake the architecture of everything. That means when we're discussing AI, when we're thinking about its implications, when we're trying to understand how companies will and won't behave around it, we have to be thinking not on two and a half month terms, but on 25 year terms. In other words, Gemini might be the next thing to be released, but it won't even be close to the last thing to be released.

19:00Until next time, guys, peace.

19:10You