Gemini did not move the needle. Not yet, at least.
But the AI Arms Race sure got a lot more intense today
Today Google announced the release of Gemini, the long-awaited response to OpenAI’s GPT-4, the brains behind Microsoft’s Copilot.
According to Google DeepMind’s Demis Hassabis, Gemini was “built from the ground up to be multimodal, which means it can generalize and seamlessly understand, operate across and combine different types of information including text, code, audio, image and video.”
The press coverage was largely positive. It should have been a day of celebration.
Instead, Alphabet stock fell on the news. (Update: on December 7, Alphabet shares are trading up again, about 5 dollars higher. Apparently it takes the distributed intelligence known as the stock market 24 hours to process this kind of information).
What happened? According to The Information, “investors known when something isn’t a big deal.”
The version of Gemini released today is comparable to GPT-4, on performance metrics that the average Web surfer won’t notice. But this news did not impress investors because, after all, ChatGPT is more than a year old. Which is an eternity in the AI field.
Is matching the performance of a year-old software product from a non-profit organization really the best that Google can do?
No, no, no, we’re told. Just wait a little longer for something really special. Google promises us an even better version is on its way. Some time next year, in January maybe, the more capable version, called Gemini Ultra, will be deployed.
Maybe.
A month or more is a long time in the AI racket. New products are released daily, and performance improves noticeably on a monthly basis. Microsoft, OpenAI and the rest of the pack won’t be sitting still, waiting for Gemini Ultra.
No wonder investors are skeptical.
Google has been consistently behind in the race to launch ever-more-powerful AI models and apps. It's not because Google lacks the expertise. Remember, Google invented transformer models. They certainly have the know-how.
What Google seems to lack is the desire to win.
Which strikes me as kinda sad because Google is actually the good guy in this saga.
The search giant has acted with restraint, refraining from wiping out everyone on the AI landscape when they had the chance to be first. They also seem to heed their AI safety and ethics advisors. At least, that’s what they say.
Consider Microsoft for comparison.
Satya Nadella practically salivates with lust when he talks about Copilot. His glee is evident. He sees conversational AI as a way to dethrone Google, demolish their grip on search and advertising, and exact sweet revenge for all the damage that Google did to Microsoft's cozy monopoly on the desktop.
Nadella doesn’t mind coming across as a demented Bond villain because he is aiming for category domination. He knows that whichever company wins this next round of AI has a chance to dominate *all* of software and *all* of smart enterprises for a generation. That much is evident to anyone who watches video of him rhapsodizing about AI. He does a pretty good job of playing the Bond villain, actually.
If you recall the Microsoft monopoly of the late 1990s, then you'll probably agree that Microsoft winning in AI might not be so great for everyone. Back then, Microsoft snapped rival software firms like matchsticks and sucked all of the profit out of the PC category.
The stock market is sending a pretty strong signal: investors do not give a toss about AI safety, AI alignment or AI regulation. Nor do investors care about Google's self-inflicted struggle to come from behind.
Investors seem to care about one thing.
They are going to reward the company that demonstrates the hunger and the aptitude to dominate this market.
Personally, I question the wisdom of that singleminded focus on market dominance. But I am frequently reminded that it is foolish to second-guess the stock market.
Now watch as Google attempts to satisfy investors by integrating some version of Gemini into everything from maps to browsers to email to docs to cloud computing and Android phones, as well as smart devices in the home. They will do this as fast as possible. They have no choice because they are attempting to catch up just as Microsoft deploys Copilot in every software product in the portfolio.
The upshot: billions of people are about to experience integrated conversational AI; this deployment will occur in the next year. It could be the fastest engineered change in human behavior in history.
At some point next year or in 2025, it will be considered normal to talk to machines. And the machine will talk back. (Personally I can’t wait til Siri gets a cognitive upgrade). The recent Mercedez Benz integration with ChatGPT is a preview of that future. In a few years, talking cars will be commonplace.
Today, Gary Marcus shared a "hot take" about Gemini Pro today in his newsletter, which I enjoy immensely. (Get his newsletter here). His point is that ChatGPT has been displaced from its singular position in the marketplace. That is bound to tarnish OpenAI’s shiny halo, particularly as a followup to the spectacular display of board-level chaos two weeks ago.
Customers demand an alternative because OpenAI is obviously not a dependable organization. No customer can tolerate the risk of being single-threaded on OpenAI’s GPT-4… or the product of any one vendor.
By offering a comparable alternative, Google has complicated matters for OpenAI.
Gary Marcus goes a little further. He also speculates that large language models may have plateaued. For instance... why hasn't OpenAI released GPT-5 yet? Even Bill Gates seems to be backpedalling away from it.
If Marcus is right, then OpenAI faces a terrible problem. If they cannot continue to escalate capabilities by rolling out ever-more-powerful language models trained on ever-bigger data sets, then Google (and others) will eventually catch up.
With the advent of Gemini, OpenAI now faces at least one competitor with a comparable product who has the expertise, the vast amount of cash, and the immense compute power necessary to bust OpenAI's lead-to-market advantage.
If Google manages to get their act together and win the next round when they launch Gemini Ultra, then the pressure on OpenAI will become intense.
Google has no choice; they must act now, because the competition with Microsoft is a death match that they seem to be in the process of losing. The market demands a response. And Sam Altman knows it. So you can expect a pre-emptive strike from OpenAI.
So it’s game on. The next two months are going to be interesting.
Nor are the other rivals idle. Today Meta announced that their own virtual assistant, cleverly named MetaAI, will now be integrated into all of their messaging apps. Meta claims to be testing “more than 20 new ways generative AI can improve your experience across Facebook, Instagram, Messenger and WhatsApp, spanning search, social discovery, ads, business messaging and more.”
The whole MetaAI announcement is worth reading, but what I found notable is a new feature called “Imagine” which seems to play to Meta’s strength in group collaboration. Imagine makes it possible for anyone using Messenger or Instagram to generate an image and share it with friends. The friends can press and hold the image to “reimagine” it by adding a new text prompt. This kind of shared image creation sounds like it could be fun, and if successful, could add fresh vitality to social media because it has the potential for viral growth, which Meta can and will amplify.
Like I wrote last week, the AI arms race is intensifying and accelerating. Which means every company will be tempted to jettison safety and regulation and caution because those things add drag coefficient at a time when speed is necessary for survival.
So, as we approach 2024, watch for the following:
The arms race in AI will continue to intensify, and the entire software industry is going through a period of self-disruption on a global scale. #interestingtimes.
OpenAI needs to do something significant to stay ahead of Google ... otherwise they will lose their premium as the market leader. We will learn quite a lot if and when Sam Altman closes the next round of funding. If, in the next two months, he manages to close a new round of financing for OpenAI that preserves the pre-soap-opera valuation of $86 billion, that will be a strong indication that OpenAI has another trick or two up its sleeve. On the other hand, if he raises less, or at a lower valuation, then that’s an indication that Gary Marcus may be right about LLMs hitting the performance ceiling.
We can count on Microsoft to continue to push the legal, ethical, operational and safety envelope, thereby increasing the pressure on the entire field to cut corners and move faster.
The voices of caution have lost. AI Doomers will continue to gnash their teeth and wail; the dog may bark but the caravan moves on.
Investors are fickle. They will continue to reward the most aggressive and reckless firms until recklessness leads to a disaster, perhaps something catastrophic. Then the investors will punish the firm that screwed up, and momentum will shift to the other players. This means the next round is largely about execution. Which of the big tech companies can execute this kind of product deployment and scaling process faster and better than all the others? We are about to see two of the biggest and most powerful companies on the planet locked in a brutal competition where blunders will be penalized severely.
Google has an awful lot riding on a successful launch of Gemini Ultra. Considering their lackluster sequence of botched launches and day-late-dollar-short disappointments so far in the AI race, I remain skeptical, even though I would personally prefer that Google win the next round.