The AI Daily Brief: Artificial Intelligence News and Analysis - OpenAI Is Now On A Billion Dollar Revenue Pace

Episode Date: August 30, 2023

On the Brief, NLW looks at new reports that OpenAI is clearing $80,000,000 a month, even before the launch of ChatGPT Enterprise. Also on the Brief Snapchat Dreams is live; an AI-powered defense system around D.C. and more. On the main episode, NLW looks at all the announcements from Google Cloud Next including Vertex AI, Duet AI and an updated partnership with Nvidia. ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at all of the various AI announcements from Google Cloud Next. Before that on the brief, OpenAI hits a milestone billion dollar revenue run rate. 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 YouTube, our Discord, and our newsletter. Welcome back to the AI breakdown brief. All the AI headline news you need in around five minutes. We begin today, which you can tell if you're looking at the video and this excellent image, behind me, with the story that OpenAI has hit a milestone billion-dollar run rate.

Now, to really understand why this is significant other than just being a big number, I think it's important to go back a couple weeks to a story that even at the time I said was stupid, but still got a lot of traction. The trade publication payments wrote on August 16th is going for broke with ChatchipT bankrupting Sam Altman's OpenAI, and basically what generated this and other articles like it, was a report that OpenAI was spending $700,000 a day maintaining their underlying infrastructure. Now, that number wasn't confirmed, but we did know for a fact that the company had lost $540 million in 2022. Combined that with what many were reporting as the drop in chat GPT's usage in June, and you have this potent little narrative concoction

where costs kept rising, usage seemed to be going down, and maybe this whole AI thing wasn't as big and disruptive as it seemed. That's a much better media story than it is reflective of reality. And two, in addition to not taking into account whatever money they were making, it also didn't take into account A, the $10 billion they had in the bank from Microsoft, which could sustain that burn rate for a very long time. And B, the fact that it was highly likely that if Open AI was ever in trouble from a capital perspective, there would be a nearly unlimited spigot of investors who would like to get in. Amadma stock from Stability AI said as much as well, quote tweeting something that was eventually deleted about the $700,000 a day story and saying,

what is this rubbish? OpenAI lost twice that a day last year and this year raised $10 billion from Microsoft, enough to maintain that burn for 37 years. This is cheap R&D relative to impact, way more bang for the buck than Web 3 Metaverse or whatever. One of the commenters on the post, Steve Stojik wrote, the AI thread boys are desperate for something to talk about, and candidly, I don't think he was all that wrong. Still, that is the context in which yesterday's report from the information came in. On August 29th, the information writes, OpenAI passes $1 billion revenue pace as big companies boost AI spending. Now, a couple important things about this reporting. First, according to the information sources, this is, quote,

far ahead of revenue projections the company previously shared with its shareholders. Second, it shows significant growth from $28 million in revenue for the entirety of last year to more than $80 million in revenue per month right now. But third, there wasn't necessarily as much information as we might like about where that revenue was coming from. As the information writes, the percentage of revenue OpenAI generates from chat GPT subscriptions versus selling access to GPT4 through an API also couldn't be learned. But in March of this year, OpenAI had between 1 million and 2 million chat GPT subscribers paying $20 per month, set a person with knowledge of the internet.

the figure. One other interesting aspect of the story is that they talk about how big Wall Street trading firms such as Jane Street have become some of the biggest customers for OpenAI and Microsoft. Again, from the information, for traders at top trading firms such as Jane Street who are aiming to make profitable bets and markets, AI language models can help by quickly parsing large amounts of data or paraphrasing lengthy memos, giving traders' answers that can inform their rapid-fire decisions. Another high-frequency trading firm Citadel has gotten close with OpenAI and other language model developers as it looks to incorporate the technology. Another financial use case comes from Morgan Stanley, who apparently commissioned OpenAI to build a bespoke customized model, train on over

100,000 documents that come from Morgan Stanley itself. Jeff McMillan, who runs data analytics and innovation for the wealth management division at Morgan Stanley, said in an interview earlier this year that the tool was making it about eight times faster for wealth managers to compare a bull and bear case for a specific investment, from 40 minutes down to five minutes. Now, I think the most interesting thing about these numbers, big as they are, is that this isn't even counting ChatGPT Enterprise, which just launched yesterday, as we know, and which feels very likely to me to be a much more significant driver of revenue than even these retail subscribers. Amad again from Stability AI shared an image of the information article and writes before enterprise adoption even kicks in properly.

Folks don't understand how stupidly large this market is going to be. A thousand companies will spend $10 million. A hundred will spend $100 million. Ten will spend $100 million. Ten will spend a billion in the next year or so as it becomes enterprise ready. 30 billion plus, cheap, given the impact it will have refined. So when all is said and done, I don't think we currently have to worry about Sam Altman and Open AI going out of business any time soon. Next, a little bit of a follow-up from something that we had heard last week. A mobile leak suggested that Snapchat was working on a new AI feature that they were calling

dreams. It was a feature that would allow the user to put their selfie in any environment that they could imagine, and that feature is now not only confirmed, but has been released in Australia and New Zealand. According to The Verge, the plan is to roll that feature out globally in the coming few weeks. Interestingly, in addition to just staying up on the generative AI craze, it sounds like Snapchat is also attempting to make this a moneymaker. The Verge writes that the first pack of dreams will be free, but that each additional dream will cost $1 as an in-app purchase. The Verge author, for his part, didn't give the feature super high marks. The company apparently shared an early

version with him, and he said, several of the dreams I received didn't really look like me, although there were a couple funny ones I would consider sharing as a joke. Most of the images didn't seem to capture my likeness relative to the selfies I've made in Lenza and other AI apps. Will the feature catch on? Only time will tell. But it's another indication of just how much social companies are looking to new generative AI experiences as the next frontier for their apps. Speaking of apps and generative AI, one new app that has quite a bit of buzz is ideogram. Yesterday the company tweeted, Today we're opening ideogram to everyone on the planet.

Sign up at ideogram.a.iogram enables you to turn your creative ideas into delightful images in a matter of seconds. It's free and has no limits and it can render text. Now basically this is just like if Mid Journey had its own social network. Instead of sharing photos that you've taken on your phone, you instead share images that you've generated through a prompt. It creates a feed like any other social app, and you can see images from people that you're following, recent images from across the community, images that are trending, etc. I did a little experiment yesterday after securing the NLW handle, as I always do, trying to create a nostalgia-filled image of a neighborhood at Halloween in the 1990s. It did okay, except for some of the scariest wonky faces that I've ever seen from an image creation app, suggesting that from a technology standpoint this may be still a little bit behind some of the other solutions out there. Still, people are excited about it.

Swicks, who you've heard on this show as a guest, says the generative Instagram killer is here. X image and team actually shipping. I bet it is killing both Google and Facebook that the next Instagram slash TikTok will probably not be created by them. Lastly, today, one from a very different area. One of the things that I think is interesting right now watching AI as closely as I do is the contrast on the one hand between the discussions of AI safety and AI policy and AI ethics that happen on Twitter and in the halls of Washington and the speed with which the Defense Department and other military institutions are just rapidly adopting AI. The Department of Defense announced this week that they will deploy an AI-enabled detection system over DC as a way to improve its efforts to protect the DC airspace.

It's notable to me not only that they are doing this, but that they think it's worth announcing in a press release. Make of that what you will. For now, that is going to do it for today's AI breakdown brief. I'll be back soon with the main AI breakdown. Welcome back to the AI breakdown. At this point, it seems like every single week, one of the big tech companies is having some major conference announcing a slate of new AI features, and this week it was Google Cloud's turn. The division of Google held their annual Next event, and unless you think it was about anything other than AI, Google CEO Sundar Pichai tweeted about the event saying, at Google Cloud Next, we've shared ways we're helping customers and partners benefit from generative AI, including

increasing the number of models in vertex AI, expanding access to duet AI, are always on collaborator and Google Workspace and Google Cloud, and more. So today what we're going to do on the episode is go through the announcements, situate them in the larger context not only of Google, but of the AI competitive landscape, and see if there's anything that you really need to be paying attention to. First up, let's get a little bit more of a summary of this event and what they announced. On their blog post, the team writes, this week, Google Cloud will welcome thousands of people to San Francisco for our first in-person Google Cloud next event since 2019. The post then herald's milestones they've achieved, including a $32 billion annual revenue run rate,

but really the emphasis was almost entirely on AI. They write, today at Google Cloud Next, we're proud to announce new ways we're helping every business government and user benefit from generative AI and leading cloud technologies, including AI optimized infrastructure, vertex AI, duet AI, and more. Let's talk about Duet first, as it is the most consumer-friendly and potentially most generally relevant for the average person who's going to be listening to this podcast or watching this video. The verge sums up, Google's Duet AI is now available in Gmail docs and more for $30 a month. You can now use Google's AI to make spreadsheets, whip up slide decks, and summarize all those documents you were never actually going to read.

So from that description alone, you get the idea that what this is is Google's version of an AI assistant for its existing suite of business apps, including Gmail, drive, slides, docs, and more. Now, Duet was first announced at the I.O. Developer Conference earlier this, year, and at that event, they showed off a number of ways that they imagined duet being used. For example, converting information from one format to another, such as turning a Google Docs outline into a deck in slides, having it automatically generate charts of data and spreadsheets, having it help you write email responses or generate images, and even helping you query

things that are in your drive or summarize your documents, etc., etc. Now, the pricing of $30 per user is the price based on large organizational usage. The company hasn't yet announced pricing for smaller teams, but what's notable about that $30 price point is of course that it's the same price that Microsoft is charging for its AI system copilot. In many ways, you can make a one-to-one comparison between Duet's AI assistant for Google's online office tools and co-pilot with Microsoft's primarily offline office apps. I know, of course, that Office is now fully integrated online, but you get the point. The Verge writes, if you're a workspace user, duet is going to start showing up in practically

all the apps you use. In some places, it's a separate menu, which you have. access by clicking the duet icon in the top right corner. In other places, you can ask Duet for help right from within the body of your email or document. Giving Google's penchant for putting its newest features front and center even when it annoys users, you probably won't be able to ignore duet even if you want to. So a couple things that I think are interesting about this. One, it's clearly reflective of the next generation of competition between these big tech players. Microsoft has, of course, been trying to leverage its seeming lead in AI as a way to get more people into its suite of tools, but it was inevitable that everything that other people were already using,

such as the suite of Google Workspace Tools, were going to have those features as well. And that I think brings up the second interesting point, which is the commoditization of AI features that is happening so incredibly rapidly. Very quickly, these types of AI integrations are simply becoming table stakes in basically every enterprise tool that you might be using, whether you're a big business or you're just simply using the business version of tools for your individual pursuits. I think it's going to be increasingly difficult for startups who are trying to offer some sort of business use case wrapper on top of an open AI API or something like that to compete because people are just going to have so many options

that are already integrated into the tools they already use, be it Google, Microsoft, Salesforce, you name it. That doesn't mean that startups won't be able to differentiate and ultimately win. And obviously, the use of chat GPT so far has suggested that people will add new tools to their portfolio. But there is, I believe, an open question of how much people will be willing to leave the tools and platforms they already use to go seek out things that are only nominally better. In other words, the question is how much better does an external or new tool have to be relative to the AI that's simply sitting in what you're already using? Then again, we don't necessarily know that all of these AI integrations actually have product

market fit. So far, the business model for these things has more been put AI everywhere and see what sticks. We're not really even at the see what sticks phase yet, so the landscape could look very different a year from now as business users actually figure out which. workflows are going to be fundamentally changed by this technology versus where it just becomes another new feature that was once exciting, but is relegated to the dustbin of history. Now, one area where it does seem AI has product market fit is in the world of developers. For that, Google offers Vertex AI, which they say enables customers to build, deploy, and scale

machine learning models. The company notes that they've seen the number of generative AI customer projects growing 150x between April and July this year. And at this event, they announced that within the Vertex Model Garden, customers have access to more than 100. hundred foundational models, including not only Google models such as Medpom 2 and Sec Palm 2, but also external third party and open source models. Meta, for example, yesterday tweeted that Lama 2 and CodeLama were now available via the Vertex AI model garden. Now, of course, this puts Vertex as a competitor for some of the other AI work spaces

that have come out, such as Amazon's bedrock, and potentially the forthcoming partnership between Microsoft and Databricks, which is also designed to help companies either spin up their own models or take advantage of and customize existing open source models from within a trusted sandbox. In addition to just supporting new models like Lama and Claude 2, the vertex announcement also included updates in their own first party models as well as the tooling surrounding them. For example, Google claims a 25% quality improvement in their coding assistant, Cody. Now, maybe the most interesting individual announcement within this vertex section of the event was a new AI watermarking system that was powered by something called Synth ID from Google DeepMind.

Google describes Synth ID as, quote, a state-of-the-art technology that embeds the digital watermark directly into the image of pixels, making it invisible to the human eye and very difficult to tamper with without damaging the image. Demis Heshabis, the CEO of Google DeepMind, actually tweeted about this as well, saying really excited to share the beta release of SynthiD, a watermarking tool created by Google DeepMind and made available by Google Cloud to help tag and identify AI generated images. Now, this is obviously a big concern for that set of regulators and policymakers that are thinking about AI right now. specifically how to make sure that people know what is quote unquote real as we think about it now

and what was generated through one of these new generative AI tools. The challenges of this technology are numerous. One, there's just the technology challenge of if it actually works. Two, there are questions of standards. If one platform can identify images that were created by a Google tool, but another platform can identify images created by an open AI tool, obviously that reduces the utility of any one tool because it's so confined to a specific source. More practically when it comes to concerns like influencing elections, the fact that these technologies are all embedded deeply within the image so that you can't see it with the naked human eye, while understandable, we obviously wouldn't want watermarks all over the images that we created that would make them much less useful. It still

doesn't really solve the fact that people might believe something that they see on some random news website or social media. Think about the example that many people have feared, where the day of a presidential election you might see some terrifying AI-generated image that, would sway just enough people at the last minute before there was time to really confirm that it wasn't real. Is this type of watermarking system going to solve that? Hard to say. Still, obviously, it's an important area of development, and good to see that there are lots of efforts ongoing in this area. Finally, Google announced a bunch of infrastructure updates. They write, the advanced capabilities and broad applications that make Gen. A.I. so revolutionary demand the most sophisticated and

capable infrastructure. Our AI optimized infrastructure is a leading choice for training and serving gen AI models. In fact, more than 70% of Gen AI unicorns are Google Cloud customers, including Anthropic, Cobra, Mosaic, Replit, Runway, and Typeface. Now, they had a slew of announcements in this infrastructure area, but two that I wanted to call your attention to. The first was that they announced the fifth generation of their custom AI chips, what they call their tensor processing units. TechCrunch writes, the company notes that it built this edition of the chip with a focus on efficiency. Compared to the last generation, this version promises to deliver a 2x improvement in training performance per dollar and a 2.5x improvement in inferencing performance per dollar.

Now, the other big announcement was a collaboration with NVIDIA, that as of next month, they would be making NVIDIA's H-100 GPUs available to developers as part of their A3 series of virtual machines. In fact, NVIDIA's CEO Jensen Huang was actually at the conference talking about the arrangement. Basically, it's just another delivery mechanism for Nvidia compute, which has become so, so in demand, and frankly at a shortage around the world. One thing that was clear is that the markets liked the deal. Nvidia shares were up 4.2% after the announcement, ultimately closing at a new record high. Overall, the stock is up 234% in 2023 and is trading at over $500 a share.

In many ways, what is notable about this event is how many announcements there were and how much it demonstrated just a new business as usual. Google obviously was very excited to share all of these developments, but it didn't have the feel of I.O. even a couple months ago, when Google went to the first. out of its way to try to plant its flag in this AI space. What it more felt like is that this sort of generative AI integration across the suite of business and infrastructure tools is just the reality going forward. Yes, there was lots in there for developers and for enterprises and for individuals, but it was all sort of the obvious next thing because this is just where we are now. We'll have some

additional events from big companies coming up over the next few weeks, including Salesforce's Dreamforce, and I'll be interested to see to what extent they have the same feel, that generative AI has become de rigueur and just part of the expected suite of tools for modern professionals. If that is what we find, it will cement what is one of the fastest adoption cycles I think I've ever seen for any technology. And anyways, hopefully you feel caught up on what happened with Google yesterday. Thanks as always for listening or watching. Until next time, peace.