I think the focus on bigger AI models is missing the point

A friend who works in robotics in Pittsburgh told me last week that everyone is obsessed with trillion-parameter models, but the real gains are in making smaller, faster models that can run on a phone. He said the race for size is just a marketing game for big tech. I've changed my view and now I'm more interested in projects like Google's Gemini Nano. Has anyone else seen practical uses for these smaller local AI tools?

2 comments

2 Comments

harperwright18d agoTop Commenter

Yeah exactly, that's the thing that most people don't get. The whole trillion-parameter race is just flexing compute budgets while ignoring the fact that real world applications need speed and privacy. Once you've used something like Gemini Nano for offline dictation or photo sorting on a midrange phone, you realize how much value gets lost in the cloud-first approach.

roberts.jordan18d ago

Oh man, I tried running a huge model on my old laptop once and it just sounded like a jet engine taking off.