Comment by ComputerGuru
20 hours ago
This might be worth it if Gemma4 E2B were a good model, but honestly it's absolutely useless in all our testing without further training and finetuning, and those aren't usecases that are fit for normal web browser use such that one would care to support it by adding such overly broad and expensive infrastructure to make it happen.
Gemma 4 E4B is a much better model, but it's too large to simply download and run everywhere.
IMHO, this is jumping the gun. Google's going through a lot of effort to release a model that will give everyone a very poor first impression of what on-device models are capable of, souring it for everyone for a long time afterwards. It would be better to wait until a smaller, better model ships before doing this.
Most users aren't even going to know that this is here. Web developers will expose this capability to the user. The devs will have to determine if the model is delivering what they need.
It's good to have something to work with if these Web APIs are going to be part of a standard. I suppose this means that ALL the browser vendors are likely to implement something
> I suppose this means that ALL the browser vendors are likely to implement something
Mozilla has taken a strong stand against the prompt api.
I had to dig around to see where they took the stand.
https://github.com/mozilla/standards-positions/issues/1213#i...
Mozilla makes great points. Even if the API is model agnostic, which it ought to be designed as from the very beginning to even be considered a spec, models can act vastly different.
Mozilla didn't say this but the user should at least be presented an option to choose which model (at least once) starting from day one, even if your browser only has one option available. That's assuming a universe where Google plans on actually being concerned about standards adoption.
Mozilla classically has taken a very strong stand against ever holding true to their values so we'll probably get one from them in a few months.
> Google's going through a lot of effort to release a model that will give everyone a very poor first impression of what on-device models are capable of, souring it for everyone for a long time afterwards.
I wonder what that will do for the competition between hosted genai and local models...