Navigating the AI Landscape: A Survival Guide | 2024 MAD Landscape with Matt Turck and Aman Kabeer

All right, everyone, we have a very special episode of The MAD Podcast today. It's actually a very meta episode because we're going to talk about the MAD Landscape on The MAD Podcast. So MAD standing for Machine Learning, AI, and Data. And today I am joined by my colleague Aman, a.k.a. the intern. And we're referencing a tweet that went viral by someone named Christina Key, who screenshotted our annual 2024 MAD Landscape with that crazy busy image, which we're going to talk about.

And the tweet said, "Give the intern a raise." So, ladies and gentlemen, the intern, Aman.

That tweet was also copied on Litquidity on Instagram.

Congratulations on your newfound fame in our micro-niche little world of data geeks and AI geeks. So this is a little bit of an experiment, this podcast. We're going to go through the MAD Landscape, some trends and all the things, and you're going to play moderator. We're going to figure it out as we go.

Yes. So the MAD Landscape is our annual labor of love, something we've been doing since 2012. It's actually the 10th edition because we skipped a few years here and there, but we've been doing this since 2012. And what it is is really a couple of things. One is a big market map of the data machine learning industry, which comes in two flavors: PDF and interactive, at mad.firstmark.com. Of course, we'll put that in the show notes. That this year, again, was done in partnership with our friends at Go Fractional, formerly known as Gotta Go Fast.

And Jonathan Grana in particular. So, big thank you for the help there. But the interactive version is great to search, it's great to zoom in, and each entry comes with a data card provided by our friends at CB Insights. So hopefully a very good resource. So that's one thing. And then the second thing is this write-up of key trends in AI and data. This year we did it in 24 Ideas for 2024 kind of format. So people can find that on the blog.

Again, we'll put that in the show notes. But that's the general idea.

Yeah, so of course, that's sort of the knee-jerk reaction that anybody that gets exposed to that MAD Landscape has for very obvious reasons. For anyone that has seen the PDF, it is a little bit of an eye chart, and every year it keeps getting worse in many ways. I'd argue that, look, as much as we try to do this as a map, it's also a reflection of the territory. In other words, the reason why it's so crowded is not because of us.

It's not because we put gratuitously all sorts of logos on a map. It's just the reality of the industry that there has been an explosive amount of companies created and founded over the last few years. It's a combination of two trends, and maybe we can talk about those in greater detail. But there is one trend that largely has been played out, I would say, between 2015 and 2021 that we can call the modern data stack trend, which is all the excitement around the rise of cloud data warehouses and all the tooling that went around them, before them, like the ELT, ETL, and after them, reverse ETL, like all the things.

And the peak of the excitement was the Snowflake IPO. And there was just a tremendous number of companies created, tremendous number of companies funded. And so that's one world. And then, of course, in the last couple of years, there's been this second massive wave, which has been generative AI. So same thing: bunch of new companies, a lot of VC funding, a lot of excitement and all the things. So it's really those two trends, like one plus one. And there hasn't been much of a minus.

So the minus would be, well, companies disappearing or a lot of industry consolidation one way or the other. But the reality is that there's been a little bit, and certainly we've all been waiting for that consolidation moment for years, and so far there hasn't been that much. So there's been some startup failure, but there hasn't been a lot of M&A. And we can discuss why. One reason is that companies that are natural acquirers, buyers, in the last couple of years have had their own troubles to deal with in a context where the macro environment dramatically changed.

And another issue being this phase of antitrust law enforcement, which has been pretty severe and may have discouraged some of the bigger acquisitions. So there's been some, but not a lot. So one trend plus the other trend minus not that much consolidation: that has resulted in tons and tons of companies on that landscape.

Yes. And then no IPO, right? Or very little IPO. That seems to be starting to open, but there's some really interesting IPOs in the future. When that happens, it's going to be incredibly exciting, right? Like the Databricks of the world in particular. But there's a whole range of companies that are somewhere between, call it, $200 million in ARR and $500 or $600 million in ARR that have been sitting on the sidelines waiting to go public. And I think there's a whole wave of very high-quality companies that now got fit, as the expression goes, in the last couple of years.

I think when the IPO window reopens, and of course there's all this debate about the fact that the IPO window is always open at the right price, but when it effectively reopens, I think there's going to be a lot of very exciting stuff. But in the meantime, not a lot of exits.

Yes. So it's fascinating, right, how we published this one, like, two weeks ago, three weeks ago. I guess March. Not that long ago. And it's a little bit like the Interstellar joke, right? Like, one minute in this world is like seven years on Earth, or whatever the exact numbers are. But it sort of feels like that in AI. And certainly last week there was this major announcement by Meta around Llama 3, which is basically two, and soon three, highly performant, apparently, models.

It's an 8 billion parameter one, then there's 70 billion, and then there's a 400 billion parameter model that is still being trained. That is like the latest entry into this whole wave of open-source AI, which has been fascinating to watch. And yeah, that raises a bunch of questions. So this one question, and just to stir the pot a little bit because we very much love open source in everything that we do as investors, for all the obvious reasons, or the reasons that have now become obvious.

But having said that, is that too much of a good thing? There's been such an explosion of open source in AI that we're hearing from customers, end users that we talk to, that it's just a little dizzying. I think we were just—

Yes, we were looking this up right before this, and we saw there's over 1 million models on Hugging Face and literally thousands added every day. And if you look at the LLM leaderboard, it seems to be changing like every other day, which, in many ways, is awesome because that's just like a testament to how vibrant the space is. But equally, again, from what we're hearing from users, it feels a bit dizzying. So eventually there'll be a power law and all the things, but it's an interesting moment in open source.

There's a question around what's going to happen to all the companies and projects. It's a question around monetization. When the product you build is the model and the model is open source, how do you build value around it? So there's emerging business models around that, whether that's selling through Azure or services for self-hosted, on-prem kind of versions. But if that's your business model and then you're competing with Llama 3, which is not really trying to make money with those models, what does that mean for the future of your business?

The Zuck interview last weekend was super interesting from that perspective because he did say that, yes, it would be fair to have some kind of arrangement with the cloud providers, and maybe they already have them, where, yes, you can deploy Llama 3 on Azure or other services. But he very much dismissed that as, like, a small part of the overall idea. And he made it very clear that there will be benefits around community input into the models that will make them better for the ultimate benefit of Meta AI products.

But it sounded like the key reason was really a power play in the industry where he was saying that if you look at the mobile industry, that's very much been dominated by the duopoly of Apple and Google. And he said that sucks and he doesn't want that to happen. So if you're an open-source AI provider, how do you make your business model work when you're competing with a player that also provides very highly performant open-source models but doesn't really care about the money and does that for their overall kind of power position in the industry?

This is going to be fascinating to watch.

Yeah, very much so. And the really interesting question is, for now, the open source models are not as highly performant as the commercial models, but maybe there's a world where they do become as performant and maybe even more performant. So what does that mean in terms of overall sort of distribution of value and who grabs the value where in the industry? That's going to be another part that's going to be fascinating to watch. So in the meantime, for the commercial LLM players, it's been like everybody's favorite question.

Question: given the enormity of the amounts raised in a very short period of time by all the LLM players, are we witnessing this crazy incineration of capital? People have called LLMs, I think the expression was, the fastest depreciating asset in history. Those are certainly interesting and valid points with a lot of strong rationale behind them. But equally, if you look at OpenAI and Anthropic, they're doing very well, thank you very much. They're just growing at an astounding pace in terms of revenue.

So there's a little bit of cognitive dissonance, and obviously what is true now may not be true in the future. But I think for me, it really comes down to what is a product that you sell. If your product is the LLM itself, then maybe that's tricky. If you think, if you build something and the value of what you build is there, maybe getting more open source and all the things becomes tricky, and everything does get commoditized.

But I think the OpenAIs and the Anthropics are actually going for kind of like a richer play where, yes, there is a model, but they have a layer on top. And the layer can be developer tools, can be a consumer engine, which is ChatGPT, famously, can be enterprise products. OpenAI already has released all sorts of data plays. And so it's more of a full-stack approach, and it feels like a lot of the value is actually going to be captured and retained through that application layer.

In a way that is maybe not entirely different from the cloud vendors. You could very much argue that the storage layer of an AWS versus a GCP versus an Azure is entirely not differentiated. And the game companies have played is really to capture and retain people through all the tools on top of the storage layer. And so far that seems to have worked very well in terms of growing the pie for the industry and capturing the hearts and minds of developers and customers.

So that part of the discussion very much falls in the camp of like, okay, we're industry observers. We're not investors in OpenAI, but just like everybody else, we've been fascinated with the company and its everything, right? The scale of ambition, the pace of innovation, and of course, all the drama from the CEO ouster that lasted a minute and a half to safety developers or engineers that were let go recently. I mean, it just seems like something new happens every minute, and it's fascinating.

It truly is amazing to see a company that sort of escapes the sphere where we all operate that's dominated by the sort of laws of gravity, where raising money is hard and all the things. This is a company that seems to have unlimited access to everything, including capital. And it's probably a good thing because they presumably are burning an immense amount of money and an immense amount of compute for their very ambitious goals, so that it sort of feels like they kind of have to continue raising, but they're very successful at it.

So it's just fascinating to watch. The question is whether they can continue to escape the laws of gravity forever. I don't know, but it's a fascinating effort at sort of doing all the things at the same time. Like, as we were saying a minute ago, they seem to be doing the consumer thing and then the enterprise thing and then the developer thing, and then to do it all around the world. And that seems to be a lot, and that seems to be a formidable execution challenge.

And then it seems that Sam Altman, who's the Steve Jobs of our generation—we're actually joking, the Steve Jobs of the TikTok generation—unlike Steve Jobs, who was fired and came back years later, he left and came back within what felt like a minute. But Sam Altman, being the Steve Jobs of our generation, seems to be doing a lot of different things as well. He's doing OpenAI, he's doing the Worldcoin project, he's doing apparently that multi-trillion chips company, he's doing the AI hardware consumer thing with Jony Ive, formerly of Apple.

So this company doing lots of different things, the CEO doing lots of different things—how long can you just completely escape the laws of gravity that every single startup seems to be subject to?

Like, where the mantra in the startup world is like, focus, focus, focus. You could argue that Coinbase is actually a good precedent for it. It's also a company that was early-ish to—well, actually early to—a very large industry and has been doing the consumer thing, the enterprise thing, the institutional thing. So maybe it's doable. But anyway, it's an incredible way of building a company. And it feels like a lot of the industry is now waiting with bated breath for GPT-5.

And how much of a jump GPT-5 is going to be in terms of overall performance over GPT-4, I think, is going to be super interesting. Very much like a bellwether of where the industry might be going. It sort of feels like if it's an incremental improvement, that may signify that this part of the hype cycle is ending. If it's crazy, as everybody's hoping, then yes, then it feels like we're still very much in that exponential curve somewhere. We don't know where exactly, but then that'll be a very exciting moment.

It feels like OpenAI does have a lot of responsibility, so much impact, and therefore responsibility for the whole industry.

Yes.

So, look, I mean, it feels like there are some cracks in the hype cycle. Look, there's always a little bit of, like, what goes up must come down at some point. And for a society that, as a whole, seems to have a little bit of an ADD problem, it feels like we've all been excited about generative AI for a number of months now, years and years. So it's going to be interesting to see how long that lasts. It seems like it's constantly fueled by the new thing and the new thing and the new thing.

And we were talking about Llama 3 a second ago, but there's also the Microsoft speaking images thing. And there's always something new to be excited about, and that has been fueling the hype cycle. At the same time, here and there, again, what feel like cracks. So in particular, there seems to be perhaps the recognition that if you're not OpenAI, if you're not Anthropic, but you are just a little behind as a startup, things are a bit tougher than they may appear. So I'm thinking about a couple of recent examples: Inflection AI and Stability.

And so Inflection AI, as was abundantly reported in the press, had this super interesting sort of acquisition, non-acquisition, which I tweeted that the Microsoft legal department is in GOAT mode these days. But that's certainly an example of it. Raised $1.3 billion and was very early in its development, decided to sell itself. So again, like, a perfect example of something I'm not privy to. I don't have the details. What was the CEO's personal journey? What was the level of success of Pi as a chatbot?

I don't know. But it sort of feels like if they had something unbelievable going on, they would probably not have made that decision. And same thing for Stability. So the CEO had been controversial for a while, but, yeah, same idea. If the company was doing incredibly well, whether the CEO was controversial or not would not have been as much of an issue. So I think those are two telltale examples of what might be happening. And then there's, like, all the stories we've read in the press or heard talking in startup circles, like the Jasper example.

The most recent thing is Tome that laid off some folks. Bottom line is, like, yes, there's a lot of hype. Yes, it's easy if you have the right backgrounds and the right story to raise money. It doesn't mean that all of this transforms overnight into a very successful business. And there's still a number of questions around, okay, well, what are the actual use cases for all those things? Right? So we all loved playing with ChatGPT. We all loved playing with Midjourney.

And Midjourney has built what seems to be an unbelievable business with people effectively playing with stuff, whatever their revenues are, like $200 million or $300 million with, like, 40 people. So presumably insanely profitable. But now we're getting into that stage where, like, okay, all right, this is awesome. Clearly this is the future. The world is changing. We want to be part of it. But what does that mean in terms of my personal use cases? How often do you or I use Claude or ChatGPT on a daily basis or not?

And in the enterprise, there's the coding use case, there is the kind of, like, internal search/knowledge management use case. But what are the other use cases where generative AI is going, as promised, to just completely change everything? That's still being determined.

Yes. And then, nicely going into that, there's certainly the competition from Microsoft and all the things, which is the other big question that everybody has been asking since the beginning. And I don't know that we know the answer better today than we did then, but it seems that for stuff that's very horizontal, startups are going to have a harder time building long-lasting businesses because in this cycle they are facing a very different type of incumbents compared to prior platform shifts, where this time you're fighting the Microsofts and the Googles and the Adobes, and everybody has read the memo and nobody wants to be the sitting duck that didn't see the new trend coming.

And so they are incredibly reactive to it and have formidable advantages around the amount of data and the access to talent and all the things.

Yeah, so look, in terms of a head fake, we don't know. What we do know, working with a number of players in the enterprise arena in particular, is that a lot of the money that generative AI purchases came out of was innovation budgets rather than actual operational budgets. And that's a feature, not a bug, because that's what innovation budgets are for. But the big question is whether now we are going to go from this kind of innovation-budget moment, bring in Accenture, let's do a POC kind of thing, to now, okay, we're serious, we've identified the use cases, we're grabbing money from operational budgets where we may actually be removing money from one bucket to put it in this bucket kind of thing.

And that's going to be the big, big opportunity and challenge of 2024. So it seems to be moving in that direction, but we don't know yet.

We love the space. Hence all the MAD Landscape, The MAD Podcast, all the work we do, and all the investing we do. It's been really interesting to see how the VC world in general has reacted. There seems to be a whole spectrum. There seem to be, on the one hand, new funds that are entirely specialized, doing basically generative AI. There seem to be, within existing firms, new fund vehicles being raised to do generative AI. There's the sovereign stuff that may or may not be happening, like the Saudi Arabia $40 billion AI fund.

So there's all of that. And then at the other end of the spectrum, you actually have a handful of very respected venture firms that say, "Well, we kind of actually are sort of sitting out of the whole thing for now." And some of them are saying, "Well, our AI strategy is pretty simple. We're just going to dump a bunch of money into OpenAI, and we're going to make one investment," which is a fascinating concept. Where we are at FirstMark as a firm, we, I think, are both enthusiastic but also careful, and ultimately don't feel like we are in a massive rush.

So we're sort of keeping our investing pace consistent with prior years. I think the comparison with prior platform shifts is very apt, and maybe it's moving a bit faster this time, but not necessarily much. But if you look at the '90s and early aughts, a lot of the companies that were built—the classic example is Google—there were a bunch of AltaVistas and Ask Jeeves and Lycos that were created before Google, but then Google, a few years later, showed up and became the dominant company.

I think there's some variation of this that may be happening. The reason why we have so many questions here is we don't know. And I think everybody's sort of trying to figure it out in real time. So, very excited about the space, actively investing in it, but equally not feeling under any particular rush to just do anything and everything around AI and things that we're kind of looking at, excited about.

Yes, we use the term modern AI stack in the write-up, which I'd been trying to resist for a while, but it's sort of convenient. And the reason why I'd been trying to avoid it was because obviously it's a parallel to the modern data stack, which maybe we can talk about in a minute. But the modern data stack was both exciting, but also reflected a moment of excess. And the modern AI stack may be taking the same general direction because there's a lot of companies, there's entire new categories that seemingly are getting crowded overnight.

And a lot of those companies, to put it bluntly, don't have a lot of traction yet because it turns out that if you're trying to do something—and I'm not picking on them—but like evaluation or monitoring or all the things, well, you need to have LLMs to monitor, right? So your customers need first to define what it is that they want to do, select an LLM, bring it in and all the things, and then you can spend more time on the evaluation front.

So there's a lot of companies that are getting started with this wave and that sometimes are a little bit hard to differentiate from one another.

And yes, it's starting to feel a little bit like the modern data stack. And maybe all of this is a feature, not a bug. Like all the clichés, nothing great has been achieved without some level of irrational exuberance. So maybe we are in that moment. But yeah, lots of new companies. So having said that, it does feel like there's an opportunity ultimately for some of those companies to be very important. Without getting too much in the weeds, the reason why you need a separate stack is that, one, you're doing something a little bit different from the modern data stack in terms of use cases, but you're also dealing with a different kind of data.

So the modern data stack was very much the world of structured data, which means the data that fits neatly in rows and columns. And I'm caricaturing this—it's a lot more nuanced than that—but for purposes of this discussion, it was all about how you'd extract that data from your databases or your SaaS application, then move it into Snowflake, which ultimately is a big analytical database. Again, I'm caricaturing a bit. And then you would make sense of the data using transformation and then business intelligence dashboards, kind of analysis.

So that was a world of structured data. The world of generative AI, which is much more of a predictive analytics, predictive pattern recognition—like all those things, prediction of words, prediction of pixels—that operates on a different kind of data, which is unstructured data. So that's your text, your images, your videos, your sound. And that requires different instruments, different tools to process, manipulate, turn into the kind of format that large language models need to be able to consume the data. So that requires different kinds of tools.

So we've seen a bunch of things. So there's the vector databases, there's the sort of unstructured ETL, there is the concept of routers—companies that, if you want to use a small language model and large language model, where do you send which query to which? There's been prompt engineering platforms. There's that world that we were talking about of monitoring and evaluation. There's governance tools, there's all sorts of different things. There's orchestration, like the LlamaIndexes and LangChains of the world, and so on and so forth.

But all of this is very much early and playing out.

Yes. Yeah, you and I have talked quite a bit about the vector databases world. So vector databases are a key part of that emerging modern AI stack because they are the heart of RAG, retrieval-augmented generation, which is basically the way you bring in your data as an enterprise into a generative AI context. In particular—not just for, but in particular—to avoid hallucinations. So you just sort of compare what comes out of the model with your enterprise data. And a vector database is what holds information or data in a vector format, which is what can be consumed.

So it's sort of the step before the LLM in that kind of modern AI stack architecture. But anyway, the point being that there's been a bunch that either have appeared or have accelerated dramatically in the last couple of years. So the Pinecones and the Qdrants and the Chromas and the Weaviates of the world. And I'm sure I'm forgetting LanceDB. A lot of very smart entrepreneurs, a lot of very great products from everything we're hearing. And a lot of them have had banner years.

But equally, there's been this emerging question around, okay, well, do you really need a specialized database to do that? Or in a context where some of the more general-purpose players, like the MongoDBs of the world, have started announcing vector capabilities, is that good enough? The answer to that question will be enormously impactful to how successful those companies can be and whether they can be longstanding independent public companies. So, early days and really interesting to see how that's going to play out.

Yes.

So everybody's favorite area right now is agents.

Agentic AI agents is really kind of like that last mile where the model goes from recommending something or creating content for you to doing an action. So in a consumer context, that could be anything from like, okay, well, you should go to Italy and go to that hotel, which is like sort of the recommendation part, to here's a ticket I bought for you, or I'm about to buy for you.

So that part is the agent part. And in the enterprise, variations of kind of like RPA, where processes can be understood and then automated. So all of this is obviously the holy grail because you go from something that's cool and informative and intriguing to something that's like, oh my God, this AI is sort of taking over the world and doing all the things. So everybody's working on this. OpenAI, very clearly, a bunch of others. We know of emerging companies, stealth companies, that are doing this exclusively.

And there's other things. It's a really interesting question because fundamentally, AI in general and generative AI in particular is very much stochastic and probabilistic in nature. So that means it doesn't give you the right answer 100% of the time. So what does that mean in a context where you actually automate actions? If the thing hallucinates and makes stuff up and that turns into booking a ticket for your vacation or invoicing a customer, what does that mean? What are the guardrails?

What are the use cases where agents can be used or cannot be used? So again, super early, but definitely a really interesting area. And then another interesting area that we mentioned in the write-up is this whole edge AI kind of world, where a very exciting variation of all of this is AI being on all our devices and the Internet of Things. I don't know if that's still a thing that people talk about, but all sorts of, whether that's your Vision Pro, whatever, local devices, or your Meta glasses.

What happens when we're able to fit incredibly powerful LLMs on those? And that's a really exciting part of the world, which, by the way, is leading to that whole small language models, large language models. Yes, that's one fork. And the other fork is like the revival of consumer hardware, like the Humane Pin and the Rabbit R1 and all the things. It's super cool. Very exciting.

But yes, early adopter.

Yeah, so I think we certainly are excited about vertical tools, or tools that have a pretty kind of narrow scope. There's this whole discussion around, is AI going to kill SaaS? And I think the conclusion we came to is that the answer is that AI is probably not going to kill SaaS as such, as much as give rise to a new generation of AI-native SaaS companies, where I think those companies will have a full-stack approach where they focus on the application and the workflow and the collaboration as much as the model.

And so we certainly gravitate towards companies where founders can do both, but certainly have the technical chops for going directly into the model and doing fine-tuning, RAG, customizing with the data, all sorts of different things one can do at the model level. So I think those full-stack AI-native vertical applications are going to be a very exciting area to watch.

Yeah, and maybe that's something we could have mentioned at the beginning. But look, a big reason why we have on the landscape both all the data infrastructure stuff and all the AI stuff, which is basically the left side of the landscape and then the right side of the landscape, is because it's very much a symbiotic relationship between those two parts. And to be, as we were saying, to be able to feed those models, you need data, and data needs to be grabbed from somewhere. It needs to be prepped, it needs to be organized, it needs to be processed in a certain way.

So that's the reason why we, on one landscape, can have both the generative AI trend and the modern data stack trend. So now, to your question about the modern data stack, I don't think that the underlying reality of it is dead. I think the idea that you should have a bunch of your data in one of those amazingly elastic repositories, whether that's a data warehouse kind of thing like Snowflake or Redshift, or one of those data lakes, lakehouses like Databricks, all of this has been converging.

I think that idea is very much as valid as ever and very vibrant. I do think that you need tools that are going to enable you to manipulate the data before and after and above and below those repositories. The question is, who provides those tools? Is that Databricks? Is that Snowflake? Is that Microsoft Fabric, which is the new entrant in the space? Or are those tools provided by startups? And the modern data stack, which was a functional assembly chain, but also very much a de facto marketing alliance between startups, I think that is getting impacted by that dynamic.

The question is whether the market can sustain a bunch of those startups that were created and funded, again, in the context of rabid VC interest and possibly ZIRP years around the time of the Snowflake IPO, before and after. So fast forward several years. Now you're in a context where you still have a lot of those companies. You have a bunch of those companies that are still struggling to have traction. A lot of those companies are very much overvalued by today's standards.

And yeah, the question is, what happens to those in a context where, one, buyers have had much lower budgets? There's been a lot of budget cuts. So the idea that you could be grabbing a lot of different solutions, best-of-breed solutions, and stitch them together—nobody really has the time and money anymore to do that, or resources or engineers, in a context where there's been a lot of RIFs and all the things. And two, the puck has moved in terms of action in the market and where the heat in the market is.

That's gone from the modern data stack to generative AI. So VCs and corporate investors have moved away from that modern data stack area to the new cool thing, which is generative AI. So it's tougher times in the modern data stack for sure. So again, to summarize, the concept of it is as vibrant as ever. The reality of the situation for a number of companies in that space is tougher. And what that means is potentially some failure, which is never a fun thing to talk about.

Hopefully some consolidation, and then the big getting bigger. So whether that's Databricks or Snowflake or dbt or Fivetran, or pick your scale-up, getting bigger by adding functionality and buying some of those companies.

Well, thank you.

Yes, exactly. Well, so what are we doing? We are updating the landscape in the next couple of weeks, which is the way we tend to do it every year. So we got a bunch of comments, and thank you very much to everyone. But it's going to be pencils down at some point soon, like in the next two to three weeks.

So we're going to do one last rev, both for the PDF and the interactive version. And then we're probably going to leave it at that for this year.

That's exactly right. Things change so fast, but we're probably going to leave it at that for that version. And look, yes, things do change very quickly, but I think the general direction of all of this remains very valid for a number of months, plus or minus a few companies. And then we'll try to do the 2025 version earlier than we did this year. So maybe January 2025.

Only nine months.

Okay, well, that's—

Yeah, the pace things are going, that could be next week. So we'll see. All right, well, thanks so much.