ai

From Google’s DeepMind announcement today:

…because a typical AI tool lives in its own window, users need to drag their world into it. We want the opposite.

Okay so. Google has spent two years getting beaten up about losing the AI race, and the framing has always been the same: OpenAI took Google’s own famous research paper and beat them to the punch.

And so far all AI products are basically a chat window you go to in one form or another.

Google’s pitch from the post is that the AI should come to you. Baked into Chrome, into the OS, into the cursor itself.

It’s confidently saying: the version of AI that wins is the one that’s already where you are. And, by total coincidence, the company that’s already where you are is Google.

I don’t know if they’re right. Maybe people genuinely prefer the destination model. There’s something to I am now Doing AI as a discrete activity, in the same way Google search remained a destination for twenty-five years even as search-as-a-feature kept getting bolted into everything else.

But if you take the post seriously, the implication is sort of unkind to a lot of companies whose entire product is the chat window. If the chat window is a phase, then the chat window companies are a phase.

The word I keep coming back to from their post is detour. They’re calling the dominant interaction pattern of the current AI generation a detour. In a research blog post. That is not a neutral word choice. That is the company that owns the road telling you the place you’ve been driving to is out of the way.

An essay of mine made it to the front page of HN today to spur an interesting discussion. I genuinely believe the puck is heading to an increased amount of local AI usage.

What I’m particularly excited by is that in a few years, local hardware will be even more capable of running highly intelligent and efficient models that are more powerful than we think possible right now.

I still believe frontier cloud hosted models will always be better from an intelligence point of view pound-for-pound, but the underlying point I was trying to make in my essay is not everything needs that level of intelligence.

In fact, most things don’t.