His point was clear: LLMs can read everything ever written, but they still don’t really understand the world. They don’t learn like we do,by seeing, doing, experiencing.
At yesterday’s Google I/O, this line of thinking came up again. But this time, not just in theory.
Demis Hassabis said: “We are working to extend Gemini into a World Model—one that can make plans and imagine experiences by simulating the world, like the human brain does.”
And now this concept looks very real.
We saw:
– Project Astra, which Google now calls the start of a Universal AI Assistant. It can see your screen, listen to your voice, remember what you asked and respond in context.
– Gemini Live, which already talks, listens, and acts almost like a companion.
– Veo and Genie, which take video, images, and simple prompts and turn them into immersive, reactive environments.
For many of us who’ve seen AI as just text or chat, this is a big change.
The model is no longer just responding, it is observing, interpreting, and in some cases, acting.
It’s early days, but you can see the shift:
From tools that assist… to systems that understand.
From search and chat… to context and memory.
From “what do you want?”… to “how can I help?”
This could change how we design products, write code, take decisions and how we expect technology to work around us.
And as always, it’s moving faster than we think.
Originally posted on LinkedIn on May 21, 2025.
Srinivas Jain 