This is exactly it.

  > the cost of having a full LLM integrated into every browser install

There's no fucking way there's going to be a "full LLM integrated into every browser." You really think they're going to drop in a 20GB-200GB model with every browser? Mind you, Llama-8B is over 15GB.

Nah. So far they are doing about 50MB per language translation that you ask for[0]. You have to explicitly install languages to translate.

There's neither "a full blown LLM" (whatever that means) nor forcing AI onto you. You still have to download the language packs, they are just offering an extension that more seamlessly integrates with the browser.

And we know what they're building too! Go look in the "Labs" tab and you'll see an opt-in for testing a semantic search of your history. That doesn't take an LLM to do, that takes a vector embedding model. What next? Semantic search in page? How terrible of a feature! (But seriously, can we get regex search?)

[0] https://hacks.mozilla.org/2022/06/neural-machine-translation...

LLMs are sequence-to-sequence like language translation models, were invented for the purpose of language models, and if you were making a translator today it would be structured like an LLM but might be small and specialized.

For practical purposes though I like being able to have a conversation with a language translator: if I was corresponding with somebody in German, French, Spanish, related European languages or Japanese I would expect to say:

  I'm replying to ... and want to say ... in a way that is compatible in tone

  I didn't expect to see ... what does that mean?

  When the story says ... how is that written in Japanese?

  > most people interpret that as "LLM integration", not the page-translation feature that's been around for ages.

I mean a Firefox download is 150MB, not 16GB...

Plus, we know what Firefox is looking to do. In their labs tab they let you opt into trying out semantic search of your history. So that's a vector embedding model, not an LLM.

Edit:

Okay, they have "Shake to summarize". But that's a shortcut to Apple Intelligence. Nothing shipped with the browser. Similarly I don't understand how the chatbot window is so controversial. I̶'̶m̶ ̶n̶o̶t̶ ̶s̶u̶r̶e̶ ̶w̶h̶a̶t̶ ̶t̶h̶e̶ ̶s̶h̶o̶r̶t̶c̶u̶t̶ ̶i̶s̶ ̶o̶n̶ ̶l̶i̶n̶u̶x̶ ̶o̶r̶ ̶w̶i̶n̶d̶o̶w̶s̶,̶ ̶but are people really pressing <C-x> on a mac or ctrl+alt+x on linux/windows? It's not a LLM shipped with the browser, it is just a window split ("shortcut") to literally one of the most popular websites on the internet right now (ChatGPT is literally the #5 most visited website and you all think AI is unpopular?!). Adding shortcuts isn't shoving AI down your throat. Are they shoving Wikipedia down your throat because you can do "!w hacker news"? Give me a break guys

>Starting today, Google Translate uses advanced Gemini capabilities to better improve translations on phrases with more nuanced meanings like idioms, local expressions or slang.

https://blog.google/products/search/gemini-capabilities-tran... [Dec 12, 2025]

I think it's simpler than that. AI is fast becoming synonymous with something being force fed and generally unwanted.

AI is a broad term going back to 1955. It covers many different techniques, algorithms, and topics. The first AI chess programs (DeepBlue, et. al.) were using tree search algorithms like alpha-beta pruning that are/were classified as AI techniques.

Machine translation is a research topic in AI because translating from one language to another is something humans are good at while computers are not traditionally.

More recently, the machine learning (ML) branch of AI has become synonymous with AI as have the various image models and LLMs built on different ML architectures.