Google has been pushing to have something similar to IPA standardised for a while but main browser vendors (who don't have models they can ship to their users) are against it because it creates an unfair competitive advantage for Chrome. https://developer.chrome.com/docs/ai/prompt-api
In fact, modern Chrome already ships this API and allows you to run inference from JS using gemma. They also have specialized APIs for translation or summarization that are able to use smoler, less general-purpose models that run on device.
> What I really don't get is where will the person ask the AI, if there is no browser extension and no chatbox in the webpage itself.
browsers like Brave have a built-in AI assistant. Chrome has the prompt API and has a bunch of AI integrations in DevTools for instance. new browsers like Aside call themselves "agentic" since they have this stuff built in. i think it's likely that incumbents will all have some sort of in-browser, on device inference at some point.
In fact, modern Chrome already ships this API and allows you to run inference from JS using gemma. They also have specialized APIs for translation or summarization that are able to use smoler, less general-purpose models that run on device.
> What I really don't get is where will the person ask the AI, if there is no browser extension and no chatbox in the webpage itself.
browsers like Brave have a built-in AI assistant. Chrome has the prompt API and has a bunch of AI integrations in DevTools for instance. new browsers like Aside call themselves "agentic" since they have this stuff built in. i think it's likely that incumbents will all have some sort of in-browser, on device inference at some point.
