AI will be ‘bigger than the internet’ - Google CEO
Diffusion models challenge GPT as next-generation AI
Gemini Diffusion an ingenious AI text model
Need a newsletter for your business, but short on time?
Newsletters are the highest performing recurring email that businesses can send, but they're also extremely time-consuming to build.
Potions takes this responsibility off your plate for just $100/month.
You’ll get a polished, on-brand newsletter sent weekly - built by a team with a track record of 42% average open rates and 5% CTR - and fresh leads delivered to your inbox each week.
Get your first newsletter free
***
AI will be ‘bigger than the internet’ - Google CEO
"I think it comes from the depth and breadth of the AI frontier we are pushing in a more fundamental and foundational way. We spoke a lot about this theme called research becomes reality, but it is… We’ve always felt we are a deep computer science company, and we’ve been AI-first for a while. So putting all that together and bringing it to our products at the depth and breadth is what I think is really pleasing to see. For example, people may not have noticed it much. It was so quick. We spoke about text diffusion models in the middle of the whole thing, but we are pushing the frontier on all dimensions"
This is the only platform where I think the actual platform is, over time, capable of creating, self-improving, and so on. In a way, we could have never talked about any other platform before, so that’s why I think it’s much more profound than the other platform shifts. It’ll allow people to create new things because, at each layer of the stack, there’s going to be profound improvements. And so I think that virtuous cycle you get in terms of how you can unleash this creative power to all of society, be it software engineers, be it creators — I think that is going to happen in a much more multiplicative way. So I think you’re going to see a new wave of things, just like mobile did. Just like the internet did.
People are working on legal assistance, and there are all kinds of startups. I was recently in a doctor’s office, and they have an AI that transcribes the whole thing, puts it all in reports, and so on. That’s an enterprise application layer. It kind of works completely differently from when I went on a visit two years ago. So all that change is happening across the board, but I think we are just in the early stages. You will see it play out over the next three to five years in a big way.
Do you think in 2004 if you had looked at Gmail, which was a 20% project, which people were internally using as an email service, how would we be able to think about Gmail as what led us to do workspace, or get into the enterprise? I made a big bet on Google Cloud, which is tens of billions of dollars in revenue today. And so my point is that things build out over time. Think about the journey we have been on with Waymo. I think one of the mistakes people often make in a period of rapid innovation is thinking about the next big business versus looking at the underlying innovation and saying, “Can you build something and put out something which people love and use?” And out of which you do the next thing, and create value out of it. So when I look at it, AI is such a horizontal piece of technology across our entire business. It’s why it impacts not just Google search, but YouTube, Cloud, and all of Android. I gave the Waymo example. The sentiment on Waymo was quite negative three years ago. But actually, as a company, we increased our investment in Waymo at that time, right? Because you’re betting on the underlying technology and you’re seeing the progress of where it’s going.
I think AI is going to be bigger than the internet. There are going to be companies, products, and categories created that we aren’t aware of today. I think the future looks exciting. I think there’s a lot of opportunity to innovate around hardware form factors at this moment with this platform shift. So that aha moment of robotics, I think, when it happens, that’s going to be the next big thing we will all grapple with….
***
Diffusion models challenge GPT as next-generation AI
POST FROM MARCH 13th, important to read for next post:
Inception Labs, a startup founded by researchers from Stanford, recently released Mercury, a diffusion-based language model (dLLM) that refines entire phrases at once, rather than predicting words one by one. Unlike traditional large language models (LLMs), which use an autoregressive approach—generating one word at a time, based on the preceding text—diffusion models improve text iteratively, through refinement.
“dLLMs expand the possibility frontier,” Stefano Ermon, a Stanford University computer science professor and co-founder of Inception Labs, tells IBM Think. “Mercury provides unmatched speed and efficiency, and—by leveraging more test-time compute—dLLMs will also set the bar for quality and improve overall customer satisfaction for edge and enterprise applications.” IBM Research Engineer Benjamin Hoover sees the writing on the wall: “It’s just a matter of two or three years before most people start switching to using diffusion models,” he says. “When I saw Inception Labs’ model, I realized, ‘This is going to happen sooner rather than later.’”
Diffusion models don’t play by the same rules as traditional AI. Autoregressive models like GPT build sentences word by word, predicting one token at a time. If a model is generating the phrase “To whom it may concern,” it predicts “To,” then “whom,” then “it,” and so on—one step at a time. Diffusion models flip the script. Instead of piecing together text sequentially, they start with a rough, noisy version of an entire passage and refine it in multiple steps. Think of it like an artist sketching a rough outline before sharpening the details, rather than drawing each element in order. By considering the whole sentence at once, diffusion models can generate responses faster, often with more coherence and accuracy than traditional LLMs. Hoover points to Inception Labs’ Mercury as a prime example of how diffusion models are closing the gap. “That model proved diffusion could hold its own and is actually faster and more efficient than comparable autoregressive models.”
Diffusion models are often more efficient(5-10x) because they refine entire sequences in parallel rather than generating each word step by step like traditional LLMs, reducing computational overhead. “Our customers and early adopters are developing applications powered by dLLMs in areas including customer support, sales and gaming,” Ermon says. “They’re making their applications more responsive, more intelligent and cheaper.”….
***
Gemini Diffusion an ingenious AI text model
Gemini Diffusion is a new text model that employs the diffusion approach typically used by image generators, "converting random noise into coherent text or code," The result, Google says, is a model that can generate text far faster than other approaches.
The latest research model, Gemini Diffusion, is a state-of-the-art text diffusion model that learns to generate outputs by converting random noise into coherent text or code, like how models in image and video generation work. The experimental demo of Gemini Diffusion released today generates content significantly faster than our fastest model so far, while matching its coding performance. If you're interested in getting access to the demo, please sign up for the waitlist….