Context preview

Google I/O, World Models, I/O Spaghetti (Stratechery Update

A limited preview of the context Palate can cite when answering questions from Trevor McFedries.

User uploadsMay 20, 2026Public source

Document preview

Google I/O put AI everywhere, for better and for worse. Meanwhile, is DeepMind aligned with Google's business objectives? View in browser LrqY6CNprwNX2_4u6yG6tEp8ZNHIgXfQHNtkgAh2hhzEmg4AU4nlYCglv84NXZmvrb89eRwEuY-H9KW5JZB5D7YDr0FdfmY9na6g5Ph34wAprHRfYN3d-Vgi4YbD64XFrH25HdBplYODq7uvkFm1Ql-nXMlZm8fPeeXmy9P5Olvl-55n1xNrekxDXXV625pjvE9lImDsBgW_rAGIIw6E9rD0Yw1h_15iahK22xyNz0REVf5qlGrqUOGyYbFMyUIowV4WU18Q3Oplacj5eY8biRB3Vo3Igz1lbz4as7wbwTtykBHP5g8F1vGPwDrUw3Vbr2k-AMP1siR9-VA__fF8Nw> | Listen in your podcast player online%2Ffeed%2Fpodcast%2F4wXG68hJAKW44aGqcgGJo2> [image: STRATECHERY] com> Google I/O, World Models, I/O Spaghetti LrqY6CNprwNX2_4u6yG6tEp8ZNHIgXfQHNtkgAh2hhzEmg4AU4nlYCglv84NXZmvrb89eRwEuY-H9KW5JZB5D7YDr0FdfmY9na6g5Ph34wAprHRfYN3d-Vgi4YbD64XFrH25HdBplYODq7uvkFm1Ql-nXMlZm8fPeeXmy9P5Olvl-55n1xNrekxDXXV625pjvE9lImDsBgW_rAGIIw6E9rD0Yw1h_15iahK22xyNz0REVf5qlGrqUOGyYbFMyUIowV4WU18Q3Oplacj5eY8biRB3Vo3Igz1lbz4as7wbwTtykBHP5g8F1vGPwDrUw3Vbr2k-AMP1siR9-VA__fF8Nw> Wednesday, May 20, 2026 ------------------------------ Good morning, This coming Monday is Memorial Day; there will be no Update. On to the Update: Google I/O From Bloomberg com/news/articles/2026-05-19/google-revamps-youtube-docs-with-artificial-intelligence-tools>: Alphabet âs Google is redesigning its iconic search box and adding new artificial intelligence coding tools, the latest steps in the companyâs multibillion-dollar campaign to expand influence in the age of AI.

Google overhauled the search box to better handle the longer, more complicated queries people bring to chatbots. The firm is also planning to update the search engine with agents that can help people track topics of interest, make reservations and monitor their health, among other things. Some features will initially be limited to paying subscribers. From the Wall Street Journal com/tech/ai/google-unveils-new-gemini-ai-agent-for-personal-tasks-b8093197>: Google is supercharging its Gemini artificial-intelligence model to become more competitive in the era of agentic AI. The company has started rolling out what it calls Gemini Spark, a personal agent it says is capable of navigating a userâs digital life and acting on his or her behalf.

The agent will work across many of Googleâs products and run on the companyâs cloud infrastructureâ¦Google also announced the rollout of Gemini Omni, a tool to create videos from various inputs including text, audio and imagesâ¦Google on Tuesday introduced Gemini 3.5 Flash, a model it says is well suited for coding and agentic work and runs four times faster than other frontier models. Gemini 3.5 Pro, a stronger but slower model, will roll out in the coming weeks. This yearâs Google I/O was so sprawling that I decided to use both of my usual sources for these Updates, because they helpfully made my point by choosing entirely different things to focus on!

Indeed, if you wanted a positive spin on Googleâs plethora of announcements, itâs that the company is clearly fully committed to putting AI into anything and everything; if you want to put a negative spin, well, itâs the exact same thing. One of the enduring critiques of Google is that the company is unfocused and unmanageable, which, to the extent this keynote was a manifestation of the company it represents, the shoe fits. World Models The most curious aspect was the role of Google DeepMind CEO Demis Hassabis; Google CEO Sundar Pichai posted this tweet on Monday hyping up the event:

[image: X Post] com/i/web/status/2056524502746747048> Hassabis is one of the giants of AI, so featuring him in the promotional tweet makes sense; Hassabis didnât, however, feature for very much of the keynote. The first time he was on stage was to talk about Gemini Omni, a world model whose manifestation â Gemini Omni Flash â is focused on video editing and generation. [image: VideoPress Video Player] com/embed/C9CLTDeW> Over the past year, AI capabilities have leaped forwards. We now have agents that can plan and act on our behalf, and artificial general intelligence is just a few years away.

Today, Iâm excited to share the progress weâve made towards building AGI. Last year, I outlined our vision of extending Geminiâs incredible multimodal capabilities to become a world model: AI that can understand and simulate the world. This is a crucial aspect of achieving AGI and will be important for everything from building AI assistants to training robots. Now weâre taking the next big step. Iâm excited to announce Gemini Omni, our new model that can create anything from any input. It combines Geminiâs intelligence with the best of our generative media models for a new level of world understanding, multimodality, and editing.

Models like Veo, Nano Banana, and Genie are able to create extremely realistic videos, images, and interactive simulations. Although not perfect, they already demonstrate some impressive notions of intuitive physics. With Omni, weâve now made even more progress. Itâs a step change in simulating things like kinetic energy and gravity. Previous systems would have found these concepts difficult. Geminiâs world knowledge and reasoning really shine in Omni. It can translate complex ideas into highly accurate videosâ¦Weâre starting with video, but over time, Omni will be able to generate any output from any input. This was always our goal with Gemini, and why we built it to be multimodal from the very start.

It was a harder path, but the foundation is now paying off. Hassabis has been making this argument about the importance of world models for a long time; for example, this is a post on X com/demishassabis/status/1926085891715412175> after the release of Veo 3 last year: [image: X Post] com/i/web/status/1926085891715412175> He laid out the case for world models most explicitly last December on the Google DeepMind podcast v=PqVbypvxDto>: [image: VideoPress Video Player] com/embed/5p0EHmJu> *Tell me why you care about simulation. What can a world model do that a language model * *DH:* Itâs probably my longest-standing passion is world models and simulations, in addition to AI, and of course itâs all coming together in our most recent work, like Genie.

And I think language models are able to understand a lot about the world â I think, actually, more than we expected, more than I expected â because language is actually probably richer than we thought. It contains more about the world than even linguists maybe imagined. And thatâs proven now with these new systems. But thereâs still a lot about the spatial dynamics of the world â spatial awareness and the physical context weâre in and how that works mechanically â that is hard to describe in words and isnât generally described in corpuses of words.

And a lot of this is allied to learning from experience, online experience. Thereâs a lot of things which you canât really describe something. You have to just experience it. Maybe the senses and so on are very hard to put into words, whether thatâs motor angles and smell and these kind of senses. Itâs very difficult to describe that in any kind of language. So I think thereâs a whole set of things around that. And I think if we want robotics to work or a universal assistant that maybe comes along with you in your daily life, maybe on glasses or on your phone and helps you in your everyday life, not just on your computer, youâre going to need this kind of world understanding, and world models are at the core of that.

So what we mean by world model is this sort of model that understands the causative and effect of the mechanics of the world â intuitive physics, but how things move, how things behave. Now, weâre seeing a lot of that in our video models, actually. And one way to show, how do you test you have that kind of understanding? Well, can you generate realistic worlds? Because if you can generate it, then, in a sense, you must have understood â the system must have encapsulated a lot of the mechanics of the world.

So thatâs why Genie and Veo and these models, our video models and our interactive world models, are really impressive, but also important steps towards showing we have generalized world models. And then hopefully, at some point, we can apply it to robotics and universal assistants. And then, of course, one of my favorite things Iâm definitely going to have to do at some point is reapplying it back to games and game simulations and create the ultimate games, which, of course, was maybe always my subconscious plan.