This AI Doesn’t Search Google, It Imagines the Internet

Hey Friends,

This week I watched something that completely messed with my sense of how AI will function in the near future. It wasn't another chatbot or pixel-perfect image generator. It was a deeper shift, systems that skip the old methods entirely and operate on a whole new logic. This episode had robots feeling packages, AI hallucinating web pages, and video models simulating physics more accurately than cameras. What’s coming isn't a better version of what we had before, it’s something different, and honestly, a little wild.

Let’s start with the most brain-bending one. Alibaba unveiled an AI model called ZeroSearch, which doesn't look anything up in a search engine. It doesn't need Google, Bing, or links. It imagines what a top 10 list of links should be based on the way you ask a question. Instead of pulling real results from the internet, it predicts what should show up, and shockingly, it’s often better. The cost savings are massive, but more importantly, this kind of model becomes independent of the web’s actual architecture. It doesn’t search, it hallucinates relevance

Meanwhile, Apple is in the middle of a quiet shift too. Eddie Cue, one of Apple’s senior leaders, testified that more users are skipping Google in Safari altogether and going straight to AI tools. Apple gets $20 billion from Google every year to make it the default, but Cue admitted that this may have slowed their own innovation in search and AI. That’s huge. Apple is now looking at ways to bring AI-driven answers directly into Safari, maybe even from partners like OpenAI, Anthropic, or Perplexity. If they move forward with this, the biggest pipeline of search traffic could reroute overnight, and Google knows it. Their stock fell almost 9 percent after that testimony.

Amazon has also quietly introduced a powerful piece of tech. It’s called Vulcan, and it’s a robot that can feel. Not visually detect, but physically sense the shape and texture of what it touches. That’s a huge leap forward in industrial automation. It can sort Skittles from a plush toy, and even measure how hard it’s gripping. It’s already running in Amazon warehouses, and nobody knew until days ago. This is the kind of tactile intelligence that most roboticists didn’t expect to scale so soon. Now that it’s here, dark warehouses with no lights or people might become the standard.

On the entertainment front, Google DeepMind dropped a new update to its table tennis robot. It now plays at the level of an amateur human. The real breakthrough isn’t the gameplay though, it’s the way it thinks. The robot doesn’t have a fixed program. It draws on a collection of micro-skills, switching strategies based on your personal play style. It learns from you while it’s playing you. That kind of live adaptation isn’t only for fun. It’s the future of how machines will handle real-time challenges, whether in robotics, customer support, or autonomous control.

The part that really twisted my brain was a model called MagicTime. It was trained on time-lapse video, flowers blooming, bread rising, slow chemical reactions. The researchers taught it how time works, not in seconds, but in changes. That’s why the model can now simulate processes that cameras can’t even capture in real time. It outputs videos with frame rates in the hundreds of thousands per second, capturing bullets smashing metal or explosions unfolding millisecond by millisecond. It’s not animation, it’s a simulation of physics so accurate that it might replace sensors and cameras in research labs.

All of this points to a bigger theme. AI is no longer building on top of old systems. It’s replacing the foundations entirely. It doesn’t need Google, manual touch programming, or even real-world video to understand and replicate how the world works. Whether you're excited, nervous, or skeptical, one thing’s clear. The platforms of the past are becoming optional.

Warmly,

Dylan Curious