Google Chrome is downloading a 4 GB Gemini Nano model onto users' machines without consent, with no opt-in, no opt-out short of enterprise tooling, and an automatic re-download every time the user deletes it. The pattern is identical to the Anthropic Claude Desktop case I wrote about last month, but the scale is between two and three orders of magnitude larger. This article does the legal analysis and, for the first time, the environmental analysis. The numbers are not small.
I’m not sure if I understand you, it sounds like you think running a small llm locally on your computer will suddenly make it use like 10x more power. That’s not how it works. It’s the servers used to run the full sized models that use that much power, as each one has tens of thousands of processors running at once. And local llms do have usage, especially for accessability. I use a local llm for my home assistant instance so I can use voice commands, which is very helpful as a disabled person.
I think their point is that regular web browsing will use less power than web browsing with local LLM calls. Your PC running an LLM is likely gonna hit its TDP limits, while browsing will be a fraction of that. Yes it’s less power than used by a trillion parameter model but I think their point is it’s vastly more than your non-LLM standard browsing would be
Your PC running an LLM is likely gonna hit its TDP limits
Debatable for a 4GB model, depending on the hardware. It’s also (most likely) not constantly running, so while yes, it will use more power than not having it, whether or not it is a significant change in the long run depends on many factors.
I’m not sure if I understand you, it sounds like you think running a small llm locally on your computer will suddenly make it use like 10x more power. That’s not how it works. It’s the servers used to run the full sized models that use that much power, as each one has tens of thousands of processors running at once. And local llms do have usage, especially for accessability. I use a local llm for my home assistant instance so I can use voice commands, which is very helpful as a disabled person.
You seem to have no idea how good modern computers are at idling
What does that have to do with anything?
I think their point is that regular web browsing will use less power than web browsing with local LLM calls. Your PC running an LLM is likely gonna hit its TDP limits, while browsing will be a fraction of that. Yes it’s less power than used by a trillion parameter model but I think their point is it’s vastly more than your non-LLM standard browsing would be
Debatable for a 4GB model, depending on the hardware. It’s also (most likely) not constantly running, so while yes, it will use more power than not having it, whether or not it is a significant change in the long run depends on many factors.