Götterdämmerung: Has Alphabet Lost Its Edge?
This is the author's opinion only, this is not financial advice.
By the mid-90s, the nascent Internet had grown into a massive stream of data by the standards of the time. A few visionaries had the foresight to see that whoever could tame this mighty stream and extract its gold could forge it into an instrument of power and might. What sounds like a cyberpunk version of Richard Wagner's Ring des Nibelungen, is the story of Google, as Alphabet was then known. In those early days of the Internet age, the struggle for control of the world's digital data streams began. We saw the rise and fall of big names, some quickly forgotten, others prevailed: Alphabet, Meta, Amazon, Microsoft, to name the most important. Over the past two decades, the amount of data on the Internet has grown by orders of magnitude, and Alphabet's Google Search has been the compass most people use to navigate this ocean. This put Alphabet in a clearly superior, if not unbeatable, position for years.
Since then, Google and later Alphabet has been at the forefront of technological innovation. As early as 2001, Google Search implemented machine learning. And in 2005, Google Translate, an AI for text translation, was launched. In 2014, Google made it clear that it saw huge potential in AI to shape the future when it acquired DeepMind for about $500 million. In 2015, Google released TensorFlow, which quickly became the state-of-the-art framework for AI research and development. In 2016, AlphaGo defeated the world champion in Go, a game that for decades was considered too complex for artificial intelligence. In April 2022, scientists from Alphabet's DeepMind submitted a paper to the prestigious journal Science, showing for the first time that an "artificial intelligence system has performed competitively in programming competitions." The scientists concluded that "the development of such coding platforms could have a huge impact on programmers' productivity. It may even change the culture of programming[...]". The paper was published in early December 2022, when OpenAI's ChatGPT had already taken the world by storm.
OpenAI's scientific publications up to the release of ChatGPT seem to me to avoid making such ambitious statements. In June 2022, for example, OpenAI scientists published a paper in which they proudly announced that they had trained an AI to play the video game Minecraft using online videos, and that they were "first to report computer agents that can craft diamond tools, which can take proficient humans upwards of 20 minutes (24,000 environment actions) of gameplay to accomplish". This is exciting, no doubt, and also kind of cute. I wouldn't be surprised if they weren't too worried about the competition in AI at Alphabet back then.
No wonder: years earlier, in 2017, it was actually Google's own scientists who, in the quiet of their labs, forged a groundbreaking new tool in this battle: transformer-based neural networks. But in November 2022, it was OpenAI, a Microsoft-funded startup, that brought this technology to market maturity, shaking the balance of power in the digital space to its core - Alphabet called it a "code red". Since then, Alphabet has made many statements to make it clear that they are not far behind when it comes to AI, and in February 2023, Alphabet released its own chatbot, Google Bard, which in my opinion is inferior to OpenAI's GPT 3.5 and can't keep up with OpenAI's GPT 4 at all. Some might call this the innovator's dilemma, but perhaps Alphabet simply missed these developments out of hubris.
AI will take over many tasks in the future, but I see a lot of potential in its ability to generate program code. Programming skills are also a good way to objectify an AI's capabilities: Can the AI solve a given problem at all? Is its solution efficient? Is the code elegant and does it follow common paradigms? I mean, if an AI can't even reason in artificial languages whose only purpose is to be understood by computers, how can an AI understand the tiniest nuances in human languages? For the past year, I've been using Open AI's GPT 3.5, GPT 4, and Microsoft's GitHub Copilot to help me with my programming. I have become increasingly impressed with the code these AIs can write. I think Microsoft saw the potential of automated programming very early on, which is why it acquired GitHub for $7.5 billion in 2018. GitHub is by far the most popular platform for programmers to share and store their code in the cloud. It is where almost all major open source projects are created and managed. This gives Microsoft access to millions of lines of source code on which to train OpenAI's AI models. This puts Microsoft in a position that Alphabet will not easily be able to match in the future.
I have also had Google Bard write programs for me, and I would kindly describe the experience as underwhelming. I haven't smashed a keyboard in about 15 years, but if I had to use Google Bard all the time, I'd probably have to up the dose of my antihypertensives. When you give ChatGPT (both GPT 3.5 and 4) complex tasks, the result often does not work at first. However, if you show the AI the resulting error message and simply communicate with it, the results improve, and so far I have mostly been at least satisfied, sometimes really impressed. Meanwhile, Google Bard just gets stuck like a record and keeps spitting out the same nonsensical code.
Last week Alphabet announced its new AI Gemini, which will be released in 3 different versions: Nano, Pro and Ultra. According to Google Deepmind's recent report, Gemini Ultra outperforms OpenAI's GPT 4 in almost all benchmarks. This claim was supported by a video that turned out to be staged (I wouldn't call it a fake, as the real prompts were also published). But if Gemini Ultra is finally superior to GPT 4 and this can be proven by many benchmarks, why does Alphabet continue to hide its prodigy? Why does Alphabet take the risk that OpenAI will release GPT 4.5 in the near future and thus sweep away all the doubts about its superiority?
Hasn’t Alphabet’s innovation lead always been a big part of the investment thesis for Alphabet shareholders? Now Alphabet is trying hard to show that its secret AI, Gemini Ultra, can keep up with the already publicly available GPT 4 — marketing aside, the published report on Gemini Ultra doesn’t say much more than that. An AI is not a gasoline engine. You cannot simply put a number on its performance and then claim that a single-digit percentage difference is evidence of one’s superiority over the other. I am not an expert in quantitative assessment of AI, but AI is ultimately the result of statistical learning, and in my experience the output of the new AI chatbots is often not reproducible (especially when generating images). Accordingly, shouldn’t the significance level of the measured performance differences be specified? Couldn’t these rather small differences just be statistical noise in favor of Gemini? Also, the results given are not reproducible because we cannot test Gemini Ultra. And again, if Gemini Ultra is so superior, why can’t we just try it? So has Alphabet lost its edge? I don’t know, but with every statement without a tangible product, my doubts about the company’s superiority grow.
And let's not forget the giant behind OpenAI with a 49% stake: Microsoft, which has a product ecosystem that literally screams for AI implementations. And I'm not just talking about Word, which will soon be writing for us, PowerPoint, which will magically put together our next presentation, or Excel, which will soon be giving no one a headache: Microsoft Visual Studio and Visual Studio Code are widely used integrated development environments (IDEs) and industry standards. Microsoft's code-generating bot, GitHub Copilot, has been available as a plug-in since 2021. So if AI is going to have a real impact on business productivity and therefore our economies, it could be through the infrastructure that Microsoft provides. So even if Gemini Ultra beats GPT4 and OpenAI cannot deliver a better AI in the near future, Microsoft is providing the services where AI could become a real productivity driver.
And what will Alphabet have left if it loses this AI battle? People use its cash cow, Google Search, today largely because they've been used to it for two decades - that's worth a lot, but maybe Bing can slay even that metaphorical dragon with the magic of OpenAI, unaware of its often-called unbeatable dominance. So maybe the Google saga is made of the same stuff the old stories were made of: Ambition, power, and the inevitable fall that comes with hubris. In the meantime, I'm really looking forward to trying Gemini Ultra myself.
