I was down on Google at the end of last year asking what they were doing. Now, the tools are so impressive. You left out one of their best tools, imo, NotebookLM. But like you said, they've had enough winning.
The thing is Google is winning on paper, but they have a hard road (including building a strong enterprise culture) ahead to succeed. And success here doesn’t look like dethroning ChatGPT.
That's why Google's story these past years is so amazing. Because they've learned that lesson. Like a person that goes through a near death experience. The kind of insight you get from that gives you for life protection on these issues. You never make the same mistake again - either of not valuing every instant of life or of not being bold in your business opportunities
Even Google's failed products are part of their learning curve. For example, after failing Plus they brought a lot of the insights to YouTube and Maps, which are arguably very successful Social Networks.
That's exactly right, imagine the learning that's emerged out of those initiatives. When you operate at Goggle's scale, high-risk initiatives are at that level. It's easy to think that you can sit down and make a GANTT chart and some slides to foresee whether these initiatives will succeeded, it's tad more complex than that 😊
True! Google is so big that they can afford to have this "Venture Capital" INSIDE the company. One of the reasons I've always invested in Google is that it's not 1 company. They're actually many, and quite diversified. Alphabet is a freaking hedge fund, just like Berkshire.
For years I've been telling people to not bet against Google. They don't make trending products, but, damn, they know how to make it useful. People use Google stuff without even realizing it's a product. We fell like Google is just "the internet", it's infrastructure, roads for driving. And they take you effectively to places, quite literally sometimes.
People will be using Gemini without even realizing "it's AI". Google knows how to integrate its innovations into their services. The low-key release of NotebookLM is one example like that. While all AI companies were struggling to make AI more useful for actual productivity, NotebookLM feels like Google just put some intern to plug Google Keep with their cheapest LLM - and it worked just fine.
And now, for the cutting edge stuff: have you read about Alpha Fold models? And GenCast, for weather prediction? What about GNoMe, their model who discovered "2.2 million new crystals – equivalent to nearly 800 years’ worth of knowledge". Or their recently breakthroughs in Geothermal Energy, or Quantum Chips.
Google is a behemoth, man. And people absolutely underestimate it, just because it's an quiet monster.
Are you sure this isn’t just benchmark maxxxing? Inspired by this post I popped over to Google Gemini and asked it for some code and my results so far are Claude Sonnet 3.7 > ChatGPT o3-mini > Gemini 2.5.
Maybe I need to learn a new way of prompting to get the most out of it?
I think so, yes, every model needs your to spend some time understanding its idiosyncrasy. Also, Google's models seem very robust to that kind of over fitting
Super weird that it doesn’t have support for reading uploaded .py (Python) or .tsx (TypeScript) files. It’s got an “upload code folder” button but can’t read code!
I’ll play around with it a bit more and see if I can work around that, but that’s a pretty crazy limitation. Maybe you’re right and it DID get tired of winning. 🤣
From the Golden Retriever: OK, I get the tech analysis, thanks to you! Bill Gurley has also often pointed out the sheer range of G's businesses, which you also cover. So...what is the risk that (as before) MANAGEMENT will somehow find a way to snatch defeat from the jaws of victory, as we move forward? For me, that's really the Gillion Dollar question--because it does seem like it's their game to lose. Thoughts?
This is a good question and I think I know the answer: they have learned that lesson already. What happened to them these years is the lesson. They're back with something they didn't have before ChatGPT - they know what it is to be on the edge of defeat.
That’s my hope, as well. It’s kind of existential, actually: Search is being supplanted by AI, not immediately but irresistibly, and it’s imperative that G owns it, too. In spite of my personal enmity towards Sam Altman, I see OpenAI as a formidable competitor. Hell of a Battle Royal!
Yup, this echoes much of my own thinking. "Google being widely seen as #1 in generative AI by the end of 2025" was also among my AI predictions for this year. For the past several weeks, Google has been shipping almost non-stop.
So true. I wasn't expecting them to be so clearly my favorite so early, but here we are. I guess it was a matter of time before they figured out how to approach this new generative AI situation against quick-paced startups. An extremely good execution by the leadership as well.
Google generally suck at marketing, but this kind of works to their advantage in the overhyped AI space. They simply quietly ship increasingly powerful models and products that win people over because they work.
They won’t be able to charge what they do for model access once their search business starts to crater. I’ll admit, I don’t know when that will happen and I think it will take a long time. Nonetheless, AI is an existential threat to their (by-far) most profitable business unit. Only time will tell if they can successfully pivot
Google is certainly the favorite to win the long game...except that AI eats software. Your personal AI will replace all software in the next 5 years (or less). AI will replace all of the "noise" (email, calendar, phone calls, entertainment etc) in our lives and provide a curated experience in its place. Google's entire ecosytem is based on its software continuing to be relevant. Who wins the long game is still very much undecided.
Yep, Alberto, great article! By the way, many don't understand the hardest and biggest problem with AI alignment: AI companies will probably align their AIs somewhat but hackers won't.
There will be a giant AI botnet, bigger than that 30 millions one that was never fully eradicated. We have more bots and hacks each year, with AI agents it's gonna be much worse. We recently shared a solution to this hardest problem: it's a actually a unicorn startup you can start
I was down on Google at the end of last year asking what they were doing. Now, the tools are so impressive. You left out one of their best tools, imo, NotebookLM. But like you said, they've had enough winning.
So true. I forgot about Notebook! (despite I reviewed it right when it was launched). I hope they integrate it with all the rest.
Agreed NotebookLM is a real differentiator.
The thing is Google is winning on paper, but they have a hard road (including building a strong enterprise culture) ahead to succeed. And success here doesn’t look like dethroning ChatGPT.
Still, I’m happy for them.
Thomas Kurian is crushing it. I think they are coming on strong in the enterprise space.
Okay, I believe all that. And I love my Pixel phones. But there are a few caveats:
- History is littered with superior technologies outmaneuvered by scrappy upstarts. IBM vs MS-DOS, Betamax vs VHS, Dvorak vs. Qwerty, etc.
- Remember Google Wave, Google Plus, Google Hangouts? The Google graveyard of failed products might be the largest in the world.
- And judging by Google Workspace, Google hires the absolute worst UX designers in the world, bottom of the barrel.
Still, I hope you're right. OpenAI doesn't deserve it anymore. And anything but xAI.
That's why Google's story these past years is so amazing. Because they've learned that lesson. Like a person that goes through a near death experience. The kind of insight you get from that gives you for life protection on these issues. You never make the same mistake again - either of not valuing every instant of life or of not being bold in your business opportunities
Even Google's failed products are part of their learning curve. For example, after failing Plus they brought a lot of the insights to YouTube and Maps, which are arguably very successful Social Networks.
That's exactly right, imagine the learning that's emerged out of those initiatives. When you operate at Goggle's scale, high-risk initiatives are at that level. It's easy to think that you can sit down and make a GANTT chart and some slides to foresee whether these initiatives will succeeded, it's tad more complex than that 😊
True! Google is so big that they can afford to have this "Venture Capital" INSIDE the company. One of the reasons I've always invested in Google is that it's not 1 company. They're actually many, and quite diversified. Alphabet is a freaking hedge fund, just like Berkshire.
You must have never spent any time on an Amazon product. Prime video, AWS, even Amazon.com. They’re toxic to my soul, lol.
You’re right about the Google graveyard though. Ever seen https://killedbygoogle.com/ ?
Good points. Though I must admit, Prime video showing scene-by-scene links to actors on IMDb is one of the niftiest features nobody ever asked for.
For years I've been telling people to not bet against Google. They don't make trending products, but, damn, they know how to make it useful. People use Google stuff without even realizing it's a product. We fell like Google is just "the internet", it's infrastructure, roads for driving. And they take you effectively to places, quite literally sometimes.
People will be using Gemini without even realizing "it's AI". Google knows how to integrate its innovations into their services. The low-key release of NotebookLM is one example like that. While all AI companies were struggling to make AI more useful for actual productivity, NotebookLM feels like Google just put some intern to plug Google Keep with their cheapest LLM - and it worked just fine.
And now, for the cutting edge stuff: have you read about Alpha Fold models? And GenCast, for weather prediction? What about GNoMe, their model who discovered "2.2 million new crystals – equivalent to nearly 800 years’ worth of knowledge". Or their recently breakthroughs in Geothermal Energy, or Quantum Chips.
Google is a behemoth, man. And people absolutely underestimate it, just because it's an quiet monster.
🤖🤖🤖🤖🤖🖖
They also own a considerable portion of Anthropic (afaik ~14%), so there's also that.
And you still didn't discuss Waymo which is winning robo-taxis.
True, I missed a few things, also the world models, notebookLM…
Are you sure this isn’t just benchmark maxxxing? Inspired by this post I popped over to Google Gemini and asked it for some code and my results so far are Claude Sonnet 3.7 > ChatGPT o3-mini > Gemini 2.5.
Maybe I need to learn a new way of prompting to get the most out of it?
I think so, yes, every model needs your to spend some time understanding its idiosyncrasy. Also, Google's models seem very robust to that kind of over fitting
Super weird that it doesn’t have support for reading uploaded .py (Python) or .tsx (TypeScript) files. It’s got an “upload code folder” button but can’t read code!
I’ll play around with it a bit more and see if I can work around that, but that’s a pretty crazy limitation. Maybe you’re right and it DID get tired of winning. 🤣
From the Golden Retriever: OK, I get the tech analysis, thanks to you! Bill Gurley has also often pointed out the sheer range of G's businesses, which you also cover. So...what is the risk that (as before) MANAGEMENT will somehow find a way to snatch defeat from the jaws of victory, as we move forward? For me, that's really the Gillion Dollar question--because it does seem like it's their game to lose. Thoughts?
This is a good question and I think I know the answer: they have learned that lesson already. What happened to them these years is the lesson. They're back with something they didn't have before ChatGPT - they know what it is to be on the edge of defeat.
That’s my hope, as well. It’s kind of existential, actually: Search is being supplanted by AI, not immediately but irresistibly, and it’s imperative that G owns it, too. In spite of my personal enmity towards Sam Altman, I see OpenAI as a formidable competitor. Hell of a Battle Royal!
How is Google doing on coding benchmarks? Programming is a different beast from general intelligence. (I know plenty of smart folks who can't program)
Quite good actually
Yes. Been a Google bull for a while, the only one in my friend group, and wrote this a lil while ago...
They still have a long way to go to productize their offerings properly, but now they have a chance.
https://www.strangeloopcanon.com/p/in-defense-of-gemini
The Google‘s Agent-to-Agent (A2A) Protocol enables AI agents to communicate. https://a2aprotocol.xyz/
Yup, this echoes much of my own thinking. "Google being widely seen as #1 in generative AI by the end of 2025" was also among my AI predictions for this year. For the past several weeks, Google has been shipping almost non-stop.
So true. I wasn't expecting them to be so clearly my favorite so early, but here we are. I guess it was a matter of time before they figured out how to approach this new generative AI situation against quick-paced startups. An extremely good execution by the leadership as well.
Google generally suck at marketing, but this kind of works to their advantage in the overhyped AI space. They simply quietly ship increasingly powerful models and products that win people over because they work.
They won’t be able to charge what they do for model access once their search business starts to crater. I’ll admit, I don’t know when that will happen and I think it will take a long time. Nonetheless, AI is an existential threat to their (by-far) most profitable business unit. Only time will tell if they can successfully pivot
Google is certainly the favorite to win the long game...except that AI eats software. Your personal AI will replace all software in the next 5 years (or less). AI will replace all of the "noise" (email, calendar, phone calls, entertainment etc) in our lives and provide a curated experience in its place. Google's entire ecosytem is based on its software continuing to be relevant. Who wins the long game is still very much undecided.
Yep, Alberto, great article! By the way, many don't understand the hardest and biggest problem with AI alignment: AI companies will probably align their AIs somewhat but hackers won't.
There will be a giant AI botnet, bigger than that 30 millions one that was never fully eradicated. We have more bots and hacks each year, with AI agents it's gonna be much worse. We recently shared a solution to this hardest problem: it's a actually a unicorn startup you can start
https://melonusk.substack.com/p/notes-on-euto-principles-and-tenets
Gemini doesn't perform great in ARC Prize though, so it's reasoning capabilities are limited:
https://arcprize.org/leaderboard