hckrnws

Wrong outrage clickbait. It has nothing to do with AI features (edit: in the "LLM sense") but comes from the new vector search in the address bar. It was first rolled out to a tiny subset of users with an English Firefox in Great Britian and Canada, x64 and at least 8 GB RAM in Fiefox 141.0.2. This issue has been fixed in Firefox 143. The original source is NeoWin based on a dubious report from Reddit: https://www.neowin.net/news/mozilla-under-fire-for-firefox-a... They now corrected their article:

> Update (August 13, 2025, 03:57 GMT): While the community correctly identified a performance issue, their attribution of the cause was mistaken. A Firefox spokesperson provided the following statement, clarifying the situation:

> We're working to improve client-side matching in the address bar, which makes it possible for users to recall previously visited websites without remembering exact keywords in the URL or page title.

> We unintentionally shipped a performance bug during the phased rollout of this feature, which processes information privately on-device. After receiving reports of issues that hadn't come up in our testing, we reversed the rollout, and the performance issues should be resolved.

The article from The Register cites this too but still comes to the wrong conclusion it's for naming tab groups. The author has no idea what he is writing about.

The author is me, and I dispute that. I expressed my uncertainty in the story.

Mozilla have been building local AI for years, and it have been great.

We got tons of local AI services of acceptable quality thanks to them: Translation, text to speech, speech to text.

Let them cook, I'm sure a few of theses features will be really useful.

Better searchability of your history sounds very practical.

Local translation has been pretty useful for me. It's obviously not as good as fat commercial models used by e.g Google Translate, but get the point across often enough that I almost stopped using those commercial engines. What's interesting is that this only useful ML service came out from collaboration with several European universities and was mostly an academic project, polished by a commercial interest into a useful service. Makes an interesting comparison with those pipe dreams conjured up by Mozilla business suites that set fire to millions of dollars and ended up with nothing.

Agree, the local translation is borderline unusable. I've had to change browsers as a result.

You say you agree, but the content of your comment seems to imply you're saying the opposite of GP?

Article author here.

We (El Reg) did include a comment from Mozilla itself about this in the story.

Now a German site claims the story is all wrong:

https://www.soeren-hentzschel.at/firefox/richtigstellung-cpu...

I have disabled all the `browser.ml` functions I can find, and translations do seem to still be working. I just read that link, translated DE->EN, in Firefox 141! (Ich kann ein bisschen Deutsche verstehe, aber hier, es genug nicht war.)

For me this means some uncertainty remains. Mozilla told me that the "inference" process is for translation, among other things. But with ML off, translation still works. Odd.

Personally I am deeply opposed to genAI and would prefer Mozilla to ignore and remove all this kind of functionality.

Is it possible that translation runs on the inference process, but is controlled by flags other than the browser.ml ones?

> ...but it does look like Mozilla's product management has maintained its laser-like aim at its own feet.

Sadly, this feels very true. I used to donate money to Firefox back in the day, but since Mozilla seems to focus on anything but making Firefox a good -browser- - those days feel very distant.

Silver lining: It's local. They're not using it for slurping your keystrokes. It seems to me their stated goals, ie. searching history without remembering keywords or urls, feels well-suited to small LLMs running locally. I think though given the immense security/privacy/censorship concerns regarding AI that this ought to be done a bit more loudly and openly.

I’m not outraged as a FF user. This seems like a decent tradeoff where the entire inference is run locally, as it should be.

I would be more concerned about the RAM usage than the CPU power if it is using inference on a model of any notable size. If the model is small enough that the RAM cost is negligible, then it shouldn't be using a great deal of CPU unless it is continually running.

I'm curious how Mozilla is preventing their inference from being too greedy. Are they doing something with `nice`? If I ask it to translate or summarize, I obviously want it done pronto, but otherwise it's almost certainly the case that slower inference that doesn't interrupt anything else is preferable. Since inference is almost always intensive, this seems like table stakes so I can't imagine they did nothing.

Yeah — that’s been a recurring complaint in certain Firefox builds.

I had to disable yet another feature in about:config last week, something related to AI, not sure if it's the same. Startup and first load of many pages were painfully slow.

That's why Firefox was pulling CPU last night! I'm mainly a FF user since the FF+NoScript combo is most of the time fairly reliable in not using any CPU (I run "apps" in Chrome, but "article pages" that requires JS don't interest me too much).

Seems the issue is for 141, I am on 140.1 and I have been seeing odd issues of late.

I figured someone is hammering cloudflare, or did cloudflare make another change that is killing some sites ?

I wonder if that is part of the problem. FWIW, this is on Linux.

"People are noticing Firefox gobbling extra CPU and electricity, apparently caused by an "inference engine" built into recent versions of Firefox. "