The AI Daily Brief: Artificial Intelligence News and Analysis - The AI Competitive Landscape One Year After ChatGPT

Episode Date: October 22, 2023

NLW looks at where companies fit in the competitive matrix of artificial intelligence - from OpenAI to Anthropic to the big techs in Meta, Google, Amazon and Microsoft. Today's Sponsors: Listen to the chart-topping podcast 'web3 with a16z crypto' wherever you get your podcasts or here: https://link.chtbl.com/xz5kFVEK?sid=AIBreakdown  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 the competitive landscape of artificial intelligence around one year after chat GPT's launch. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown. Not Network for more information about our Discord, our newsletter, and our YouTube channel. Welcome back to the AI breakdown. I had a chance last week to have a really interesting conversation with a group of people

around how I saw the competitive landscape for artificial intelligence. I think it's an interesting moment to ask that question around what the competitive landscape is and how things have evolved because we're coming just up on the year anniversary of the launch of ChatGPT. I've said it before here that I think ChatGPT is the zero to one moment and one that has set the tone for the entire year. So what I want to do for this weekend edition of the AI breakdown is talk through where I see the big players first and how the competitive landscape has evolved. And then we'll talk about some of the other important trends heading into the one-year anniversary

of the generative AI space as we know it. So let's start first with OpenAI and Microsoft. OpenAI has, of course, been the company that has dominated the space, has set the tone for everything that has come, because A, their chat GPT was the first LLM interface that actually hit with people, seeing something like 100 million people try it in its first six weeks, which at the time made it the fastest growing tech product in history, by a significant margin, by the way. And that combined with GPT4, which has throughout this entire year remained the state of the art and most advanced foundation model, has put open AI in the

pole position in more than just technology. It's put it in the pole position in terms of influence, in terms of people paying attention, in terms of people wondering what comes next. Now, I won't go into depth about exactly where OpenAI is now, because I kind of did that a little bit on Friday. The TLDR is that on the one hand, there have been some small stumbles recently. Most notably, the information reported that the Iraqis model that OpenAI was hoping would provide a GBT4 level experience, but in a much more efficient and thus less costly package, ultimately had to be abandoned in the middle of this year, leading to some tension with their partners at Microsoft, as well as just leading to time wasted on that model when they could have

been focused on other areas. Still, the next big event in the generative AI space is in ways the developer event happening on November 6th, which is OpenAI's first developer event, and in which many people anticipate that we will see AI agents, as well as a lower cost GPT4 offering. I'm sure we will come back to OpenAI before this episode is over, but let's move over to Microsoft. Now, Microsoft's partnership with OpenAI, I think, set a very different tone for the way that this space could have evolved had it not been there. Specifically, it announced to the rest of the market that the big guys were not going to wait on the sidelines and that they were going to find ways, whether it was through building their own tools or through partnering

with others, to be at the forefront of this technology revolution. Now, Microsoft has not only made a bet on OpenAI. For example, over the summer, we got an announcement of a partnership with Databricks that will allow Microsoft's Azure customers to have much more choice in which models they actually work with, which some saw as perhaps Microsoft falling out of love with OpenAI, but I saw as adapting to what market signals were telling them, which we'll get into in just a moment. Still, I think what is inescapable, and what One of the things that I don't think people would have anticipated a year ago is the extent to which incumbents have an advantage in this AI space relative to other technology changes

in the past. Historically, it's always been startups who are the main drivers of some new technology, but in AI there are some different dynamics. One part of it is just the incredible capital cost to train foundation and frontier models, which automatically means that it's only a small handful of companies that can actually do that. But there is something else as well. People have realized that AI is going to be better, the more that it has access to

the proprietary data of either the person using it or the company using it, and that creates a real trust issue. Companies are understandably worried about leakages and about sensitive data that they put into something like chat GPT, ending up training the models, and perhaps even finding its way into other people's interaction with those tools. Now multiply that across all the different categories of generative AI, and you have a situation where in particular enterprise buyers are, I think, a little bit more predisposed than they might have otherwise been, to work with their existing partners who they already have trusted relationships with around their data. That creates an advantage in a moat that hasn't exactly been as present in other technology

startup spaces that I think these big guys are really leveraging to their advantage. Now, Microsoft is also, I think, representative of a key trend right now heading into the end of this year and the beginning of the next year, which is that even as competition around state-of-the-art models increases, and that particular arms race, in fact, accelerates, there is another thing going on which is much closer to the consumer, which is the integration of AI into the actual software and tools they are. already using. Now, Microsoft's example of this is, of course, bringing their co-pilot AI assistant across not just a single software suite, but across the entire Windows operating system. Between that

and their integration of these tools into office, you're going to have huge swaths of the population that maybe haven't even tried out something like ChatGPT yet, who are interacting with and actually starting to integrate AI into their workflows in the months to come. Now, let's talk about Amazon. Amazon was a company whose place in this competition was definitely impacted and shaped by the launch of ChatGPT. Reporting suggests that Amazon even had their own models that they had been working on, but scrapped them when they saw just how far ahead OpenAI was when ChatGPT came out. The company quickly pivoted and shifted gears and instead started focusing on what they call Bedrock. Bedrock is the environment through which their existing customers who work with them on cloud

and enterprise services can customize and train models with a variety of choice in ways that are specifically relevant for those particular enterprises. In other words, Amazon has decided to make a big narrative and product bet on the idea that there will not be a winner take-all, and that instead companies will want lots of choice and customization, but still value the relationships that make that choice and customization feel safe. That said, a couple of weeks ago, a news announcement suggested that Amazon might not be so unconcerned with foundation models as they might have previously seen. Specifically, they hooked up and partnered with Anthropic with a $1.25 billion investment

that might turn into as much as $4 billion. Now, in addition to that investment, the companies will also be collaborating on Amazon's AI chips. This is yet another big dimension of this competitive landscape in which the dominance of invidia and the difficulty of access and compute has pushed basically all of the big players to also think about and try to design their own chip solutions. We've had a number of rumors about that at Microsoft, with an expectation that we might actually get some more information about their project Athena soon, whereas Amazon has been more out and focused on it, and indeed part of this Anthropic deal is focused around helping them improve AWS's Traneum and Infersia chips.

Bringing this story together, the information's Jessica Lesson sat for about three hours with Amazon's Andy Jassy, and he articulated what he saw as three layers of the AI stack. The first are the models, obviously Anthropic being a big part of their approach there. The second is applications built on top of the models such as ChatGBT, and then the third layer of the stack being those chips. Now, we didn't get a lot more new information from that, but we did get Jassie's perspective that, quote, we are like three steps into this marathon. What about Google? Google has more than anyone suffered, at least in market perception, from how far ahead open AI is perceived to be.

It seems like the strong sense among many would have been that Google, of all these big companies, should have been the farthest out ahead, but that just hasn't been borne out by reality. Now, Google is on trend in one way, which is that they are working very quickly to integrate their barred AI into all of the workspace apps within Google Drive, like Google Docs. And of course, this is similar to what we were just discussing before around this battle to not just have state-of-the-art models, but to have them actually work in the workflows that people are already working in. There is also a broad sense that perhaps Google's big bet, and the thing which it wants to use to jump back out into the pole position, is coming up in the form of Gemini. At the end of August, the semi-analyst blog, got a lot of attention, basically saying that Google was using so much more computing power when it came to training this model that it made everyone else look GPU poor. A big, big question for me when it comes to how the competitive landscape shifts or doesn't shift

in the months to come is just what the performance of Google Gemini actually is and when it finally comes out. And now a word from today's sponsor. Are you interested in how two top-of-mind trends AI and crypto can work together? If so, I have the perfect podcast recommendation for you. Web3 with A16Crypto, the chart-topping show brought to you by venture firm Andresen Horowitz. Web 3 with A16Z Crypto is your definitive resource for the future of the internet. Whether you're already building in these spaces or simply curious about what's next.

If you need a place to start, they recently released an excellent episode with Stanford Cryptography Professor Dan Bonay and former Google Xer Aliya in conversation with host Sonal Choxi about the intersection of AI and crypto. From fighting deepfakes and proving humanity to large language models like ChatCBT, they cover it all. I highly recommend checking it out, especially if you'd like to learn more about how AI and crypto will impact our everyday lives. Beyond crypto and AI, this show is for creators seeking more ways to truly own their work, for business leaders trying to prepare for the future today,

and for innovators exploring trending tech topics. So go ahead, listen to Web3 with A16Z crypto wherever you get your podcasts. Now, there was another blog post on semi-analysts that got a lot of attention earlier this year, and that was a reprint of an internal memo from an employee at Google that was called We Have No Mote and Neither Does Open AI. It was all about how very unexpectedly, open source had raced out ahead and was arguably more competitive than closed models like either Google or Open AI were deploying. The main beneficiary of that was, of course, meta. Now, how much meta's open approach to AI was driven by a long-term plan and philosophy

versus just opportunism of seeing the open lane, it's undeniable that a huge amount of developer attention has been built around the Lama ecosystem because of how much more open and available it is than other closed sources. Meta's Lama 2, which was recently released, was a huge step because it came with officially a commercial license option. And in many ways, meta has made itself not only the big progenitor of open models, but also the great defender of open source in the public debate. I'm not even really touching on the policy discussion, but I do think it will have a big impact on how all of this plays out over the course of the next year to come.

Now, one area in which meta is alike the other big tech competitors is in this integration of AI across the existing places that people operate. Now, for meta, that's less work tools and more consumer experiences. And of course, we're seeing huge experimentation with AI chatbots that have different personalities, although whether that actually sticks remains to be seen. Now, there are a million other examples of how much energy right now is being focused on bringing AI into existing workflows. Adobe has been relentless this year in bringing generative AI experiences into their creative suite in a way that for some long-term Adobe Watchers is almost head spinning.

But it's also working. Popular AI content creator Matt Wolfe recently tweeted, I have a lot of monthly subscriptions for AI and creative tools. The only one I can think of that I'll likely never cancel is Adobe CC. Insane value. Now, Adobe for their part have a bunch of different competition. On the one hand, they are competing with startups who are trying to offer an alternative to the creative suite, most notably Canva,

who also just made a big AI announcement recently with their Magic Studio that has a slew of new AI design features, and even integrates partnerships with text-to-image generators and text-to-video generators, but Adobe is also competing on a model basis. While Mid-Journey has spent most of the year in the lead when it comes to text-to-image generation, they now face increasing pressure from both Dali 3 from OpenAI, as well as from the new just-released Adobe Firefly 2. One of the things that makes Firefly 2 so interesting

is that it was trained on images that Adobe owns the rights to. For enterprises and business customers that might be worry about the legal implications of copyright and AI training, this seems like it could be a huge strategic advantage, especially if the quality is similar, which it seems to be. Now, there are a million other areas that we could focus on and different categories of generative AI tools that are still coming into their own. Video generation is a big area that could see a lot of development over the next year, with companies like Runway and PICA Labs at the top of the heap,

voice and video cloning are also getting huge. Haygen has captured a huge amount of attention with their ultra-realistic avatar videos that can be used for business and education, and then of course over in voice cloning land, not only are there an incredible number of competitors who are trying to provide content creators with the ability to clone their voices and use it for various purposes,

but they're also part of a big push which genuinely could see language barriers as we know them absolutely destroyed forever. 11 Labs, Wondercraft AI, YouTube, Spotify have all announced dubbing features recently that will allow content on their platforms, be it videos or podcasts, to be automatically translated into other languages

in the voice of the original speaker, even if that speaker doesn't speak the language that it's being translated into. Now, of course, when it comes to developers and where the bleeding edge of people building in the AI space is, a huge amount of the energy and attention has been around AI agents. In other words, moving from tools that are good at helping us with specific tasks, i.e. can you create an image of X, to agents that can figure out how to solve a problem, which might involve a complex set of different tasks, and even using a bunch of different tools. Ever since April in the launch of AutoGPT and Baby AGI,

there has been a never-ending streak of AI agent startups and open-source projects that have announced themselves. And while it's not clear that there's any sort of breakout tool or even really breakout use case yet, the sheer wadage of intellectual firepower and engineering energy that's spent in this domain is likely to bear fruit in some way or another. Now, of course, I could go on. Each of these areas has so many different dimensions. But the TLDR is that there is still so much up for grabs and unshaped, about how generative AI is going to play out. Big company incumbents, major startup labs,

and even indie devs all have different things to contribute to the space and are jostling for relevance in a constant never-ending dance. There are major constraints which we haven't even really touched on yet, including access to compute and I think in the future policy. And there's also a real push to put these tools into real users' hands and into people's lives. I think as much as we saw this year, 2023, as the year of dabbling and first tests with AI, 2024 is likely to be the year where we see the utter normalization of AI at the center of so many tools that people use day in and day out. Now, if this sounds a little bit like an end-of-the-year reflection post, don't worry. We still have two months before that type of content comes out, and I'm sure that things will have changed yet again by then.

Hopefully you found this interesting or useful, however, and until next time, peace.