Google is back
Gemini is the AI interface with the most active users after ChatGPT, and Google now has: The best image model, the best video model, the best small model (Flash), a top SOTA model (2.5 Pro).
Finally some AI news worth writing about. Let’s start with this video (sound on):
That was made with Google’s new video generation tools (Flow and Veo3) that introduce a number of key new features to video generation: ‘solved’ character consistency, and speech. In addition, google announced:
Imagen4: A new image model, and (pending test driving) probably the best one right now.
A whole bunch of new developer features for building Gen AI stuff
Live translation in Google Meetings that maintains the original voice and tone (finally this science fiction staple exists!!)
Google also shared that Gemini has 400 million Monthly Active Users, making it the world’s second most-used AI Interface after ChatGPT (and only slightly ahead of Magicdoor ;)). In the mean time, Google Search AI results have gradually gotten quite good.
After a much ridiculed false start with overly woke AI that generated of black Nazi German soldiers and Asian ‘Founding Fathers’, among many other examples, many people thought Google had just lost the ability to innovate and was a hopeless sitting duck in AI.
Well, it seems that was not the case. They were clearly caught off guard by what OpenAI achieved with the Transformer Architecture that was invented at Google, but with their huge pile of money, talent, and insane distribution advantage they just put their heads down and got to work. The results are impressive. I am impressed.
Google is probably the new Bell Labs
One of the best books I read in 2024 was The Idea Factory about Bell Labs. Here are my notes on the book.
It’s funny that a friend
posted this on LinkedIn about Bell Labs:I’ve been obsessed with Bell Labs ever since reading "The Idea Factory."
I kept wondering: how did one place produce so many breakthroughs, and why haven’t we been able to recreate it?
This essay breaks it down well:
- It started with founders who were builders. Bell and Kelly were scientists first, managers second. That shaped everything.
- The principle: hire smart, obsessive people and get out of their way.
- No micromanagement. No KPIs. No endless decks. Just time, space, and trust.
- Management’s job wasn’t control, it was taste. Find great people, give them important problems, and let them run.
- People weren’t chasing promotions. They were driven by curiosity and the desire to do meaningful work.
- The culture selected for an internal drive. If you needed managing, you weren’t hired.
- The result: transistors, information theory, and other ideas that reshaped the world.
Bell Labs didn’t die because of money. It died when the culture shifted—from trust to control, from exploration to efficiency.
Most institutions today optimize for safety, not upside. That’s why we get incrementalism instead of paradigm shifts.
You don’t get genius by tracking hours. You get it by trusting smart people to wander—and to build something big when they find it.
If we want another Bell Labs, we don’t need more structure. We need a critical mass of talented people and a lot of autonomy.
Suddenly I had a realization that I had not had before: Google is the new Bell Labs, so I commented:
You forgot state sanctioned monopoly resulting in fairly low competitive pressure and high margins.
Actually, Google was in a very similar position in the late 2000s and early 2010s and they also invented a lot of stuff around that time and in a similar way. Remember the famous play time, or do whatever you want for some part of the workweek?
Viktor replied with:
A good point on the monopoly.
On the second one, I agree that Google is the most suitable contender, but it is still not quite comparable. After all, Bell Labs introduced:
Transistor (1947): Sparked all of modern computing and electronics.
Information Theory (Shannon): Literally invented the math behind all digital communication.
Laser and Solar Cell: Fundamental technologies used in medicine, manufacturing, energy.
C++ Precursors, UNIX, C: Infrastructure of today’s computing stack.
Charge-Coupled Device (CCD): Digital photography, space telescopes.
Telstar Satellite: First active communications satellite.
Submarine Communications Cables: Made the world permanently connected.
Cellular Networks: Invented the idea, protocols, and tech stack.
Digital Signal Processing, Fiber Optics, Microwave Transmission…
Whereas Google's punchy areas are mainly transformer architecture and search.
And I felt I had to come to Google’s defence on this. First of all, Google has only been going for 27 years, about as long as Bell Labs was when they invented the transistor. In that time, Google has already invented:
Google Earth, Google Maps and Street view, which we take for granted but took monumental effort and commitment to create.
Chromium: By far the best core web browser code that powers every modern browser, even Internet Explorer!
Android: By far the best light mobile OS that powers a majority of smartphones and smart 'devices'
Free, cloud-based word, excel, powerpoint
Waymo: the leading self-driving car project in the (Western?) world
Google translate
Golang, Dart and Flutter
The 'Material Design' UI system
Also, to reduce Google's basic research and innovation in AI to just 'transformer architecture' feels a bit easy. Before that there was AlphaGo, AlphaFold, and before that there was other stuff. Google is arguably the place where the modern AI revolution was invented, powered by Nvidia.
There is much, much, much more actually: a cancer detection pill, solar powered smart contact lenses, robots, drones, ultra-fast fiber, various genome projects.
There is much talk about breaking up Google for allegedly having a monopoly. If the precedent from Bell Labs is worth anything, that is probably one of the dumbest things the FTC could possibly do. Humanity can be very good at wishing for something while it’s already there right under their noses.
