Post

In response to our objection to the Prompt API, some folks replied "well, fix it. Propose something better". This is something we end up doing a lot at Firefox…

…I'm currently finishing a redesign of popover="hint" github.com/whatwg/html/..., we're proposing an alternative to the translation API after finding security issues in the Chrome API github.com/mozilla/stan.... We provided a lot of fixes to the design & spec of anchor position & the nav API…

Also, the lack of an alternative proposal is not a good reason to ship the half-baked thing.

Genuine question: are you happy with the role that the web plays on mobile devices today in comparison to native apps? If not, is it possible that the industry shift to AI might be an iPhone-level moment and put the web on the same position on desktop too (say, if openclaw and MCP takes on?)?

I'd like the mobile web to be better in the future than it is today. I don't think the Prompt API achieves that.

I would bet lurking in the background somewhere is a director or VP wringing their hands about adding more AI

“But we came up with something using our unlimited resources. Surely you can just make an effort now it’s announced and we’re pushing to resolve it”

The Mozilla objections are not to the API. The API is pretty boring and looks like what any LLM has looked like for over half a decade. The opposition is to there being an LLM at all. The objection is to Google shipping an LLM, that too many people would target that LLM. This isn't resolvable!

standards are standards due to being deterministic (hence the word standard), and while some approaches (like translation) can be coerced into that standard, llms seem far too chaotic for me to imagine there is a way to integrate them, the prompt api is probably better as a library rather than

The web is a good 15 years behind native platforms. "Going slow" is just a recipe to become even farther behind, and failing users in the process. I remember the B2G days, and Mozilla's position was not to go slow then. I missing that Mozilla.

Do native platforms have first class prompt APIs?

Windows ML is a native local AI inferencing framework based on ONNX runtime (it’s more about letting devs ship an app and then the framework makes that run well on whatever hardware and models you have so not exactly the same but native is getting these kind of features)

I use AI for: - Coding assistance - Audio cleanup - Speech to text - Searching/research - Translation Most of these are cloud. The local models are custom for the task. None are using the OS's model.

God forbid the web is ever first to something!