It’s been a story of the last week or so if you follow the kind of news channels a Hackaday scribe does, that Google have quietly installed an LLM as part of the Chrome browser. Reports vary as to when they did this because there’s a lot of confusion online with their online Gemini features also present in the browser, but it seems Chrome users are noticing its effect through slower performance and hefty disk access. Given that Chrome is by far the most popular web browser, this means that billions of users will have downloaded the four gigabyte Gemini Nano model, and now have an LLM they didn’t know about. It will be used to provide advanced auto-correct and other text suggestion features that their online version of Gemini would presumably be overburdened with, and since it’s available through a set of in-browser APIs we expect that it will find its way into a lot of websites, online applications, and plugins.
It’s caused a bit of a fuss in some circles, and we think, with some justification. When billions of computers unwittingly install an extremely energy intensive software component the effect on global power consumption will be significant, with a consequent uptick in the carbon footprint of computing. It’s not a phenomenon restricted to Chrome, as an example Siri has used a local LLM on Apple devices for a while now. We’ve seen rumblings of discontent and talk of getting European climate regulators involved, but perhaps instead it’s time to have a conversation about local AI models. The key is not whether or not they are a good thing to have, but when and how they operate.
While many of us are sick to death of AI slop and have not been lured into AI psychosis by an over-reinforcing chatbot, the fact remains that LLMs can do some useful things, they’re here to stay whether we like it or not, and having one under your control on your own computer doesn’t have to be a bad thing. Install Llama.cpp on your machine, and you’ve got an LLM of your very own, upon which your usage data isn’t going to be sold, and your content isn’t going to reinforce the finest plagiarism device the world has ever seen.
Opt-In and Opt-Out
The concerning development with the Chrome LLM is that not only has it been installed without the user’s consent, it runs without their consent too, and they can’t use it for anything except what Google Chrome wants it to be used for. Unlike the Llama.cpp mentioned above, it’s not under their control, instead it’s a compute-hungry monster ultimately controlled by Google. The prospect of a future in which multiple pieces of everyday software install their own similarly out-of-control multi-gigabyte CPU-munchers is a concerning one. Anyone who remembers Microsoft’s Clippy grabbing all the resources in a 1990s desktop as its stuttering animation played its course will know where this is going.
If local LLMs are an inevitability, what’s needed is a way to make them like any other application, one that the user chooses and installs themselves. Such an LLM could make its services available to applications such as a web browser if the user allows it to, but not run unless asked. It’s fairly obvious that installing Llama.cpp or similar is beyond many users, but it shouldn’t lie beyond the bounds of possibility to package something like it as an application they can install.
We know that the previous paragraph is pie-in-the-sky wishful thinking, and that as the person who knows computers in your family your next few Christmases will be spent wrestling with six different LLMs running on some elderly family member’s PC. But perhaps in Clippy lies the answer. If the consumer can learn to associate built-in AI features with their computer grinding to a halt just as they did with an office assistant thirty years ago, then perhaps they’ll demand change. We can hope.
Maybe it will perform a summary of your daily activity. That should cut down quite a bit on network traffic. And it will also save on power usge (theirs, not yours). Ecology and all that.
Ok Google.
Seriously BIASED article.
Google does not make you install their browser, even if some considers it the best available att.
There are so many optional browsers to fit the needs of the paranoid and those on low-end hardware.
Nothing is FREE in computing; from energy consumption to environment impact.
When there is a several double digit Watts worth of difference between running a lightweight FOSS Linux distro versus windows 11 on the same machine, its no long “you get what you pay for”, but a question to Microsoft and its ilk, “they can do it, why can’t you?”
If companies are so eager to greenwash themselves with “climate initiative” and sustainability web pages on their websites, they better show me numbers and work hard on reducing the idle consumption. They don’t trust me to take them on their face value do they?
Sorry went on bit of a rant there.
My point is, massive software companies that clearly have the resources must also be measured on the benchmark of energy efficiency. They claim to care about the environment, its their chance to convince us.
I’m not asking anyone not to run an LLM model, I’m just saying only run it when it makes sense, preferably on energy efficient dedicated hardware and not your uncle’s old Core2Duo laptop.
I couldn’t tell from the article whether this was specific to laptops/desktops, or all instances of Chrome.
It may not be default on computers, but Chrome is the default browser on Android phones, and many apps use Chrome to open websites even if a different browser is installed and set as default.
It should run under the private computer app on your phone. It’s a tiny model intended for mobile use.
To a large extent at this point they do, as actual functional on the modern web alternatives almost don’t exist at all, and with most of those ‘alternatives’ being chrome powered and the obvious standout that isn’t being funded largely by Google so they can claim they are not a monopoly…
Yes you can technically use another browser, and right now that probably will be enough, but Google have so much direct control and even more influence beyond that doing things like this is concerning and might well be something we all end up stuck with even the tech savvy folks that know they don’t have to use the provided browser in theory. At least unless we can convince all the webpages we wish to visit to go back to something that doesn’t require such a complex browser so everyone here can manage to write our own browser if we wanted to that actually functions.
Tell me where in the article that I say Google forces you to use Chrome?
All I say is that it’s used by a lot of people. Not including me, as it happens. Please read the article.
Exactly! You did not write that everyone must use Chrome. Nor that you must download the version of Chrome that has the model. It’s about the fact that Google is putting this model on others’ computers through Chrome installations that people have ALREADY installed and use!
If you’re not going to read the article, be better at faking it.
Brother, what? The same google who won the controversial antitrust lawsuit that would have made them sell off Chrome and Android? The same google who was made to stop paying apple a huge amount of money to make Google search the default browser on iOS, the primary (nearly only) competitor to their own Android OS? THAT google? If paying to make their website the default search engine and then adding popups to the website to “Install Google Chrome” isn’t making you (and everyone else) install the browser, I have a bridge to sell you. Do you think that if I build a rube goldberg machine to punch you in the face, then I activate it, that I have punched you in the face any less?
I was with you up until the last part. I tried chrome a couple times long ago, hated it and had security concerns, and so never used it again. Like everyone who’s been online (well, most everyone, since the demise of askjeeves
and webcrawler*) I’ve used Google search, but once the data/ads/etc became problematic, I’ve never used it again. Good alternatives to both exist, free to anyone who desires them.It’s true that I probably have more experience and know-how than the “average” internet user (low bar, and menial for anyone to climb above who cares to). But it’s not like alternatives to chrome and Google search are hidden in the shadows where only l33t h4x0rs know how to find. They get advertised for gods sake.
So no, they are not “making” us use their crap. Company policy might dictate what employees must use for work, but that’s up to each company’s leadership. But Google is not making me, and certainly not “everyone” use their browser or search. They do significantly tip the scales, they do use manipulative tactics to make users think they have to—but that’s very different than “making” us. Defaults can be changed. Easily. Hell, even my mom uses firefox and duckduckgo for her day-to-day (outside work which mandates chrome), and she is very far from being internet/device literate.
As such, this leaves me with two questions for you: how big is this bridge, and what are you asking for it? … OH—I see what you did there, you sneak! There never was a bridge, you were just deceiving me to make me think there was, and that I should buy it. Luckily I have a mind of my own, and am capable of making my own decisions based on the data available to me. I’m not buying what you’re selling.
*Holy crap, webcrawler still exists!
If it actually works, for most people, then what is the difference?
“All the time you’re saying to yourself, ‘I could do that, but I won’t,’ — which is just another way of saying that you can’t.”
-Richard Feynman
It would be a bit of a bait and switch if you installed what you thought was a web browser and instead (in the background) they were using your GPU to do some extra work. Hell, for all we know, your computer may now be using your electricity to answer somebody else’s questions.
That’s correct; Google pre installs it for them.
You are correct…..but a bit…. disingenuous. For a large part of the public there is only the big commerical advertised world stuff.
Oh, no, it doesn’t make, it merely makes it nigh-impossible to participate in Modern Internet (TM) by being the de-facto “standard” maker (and changing the goalposts before any non-Chrome browser gets anywhere near a bug-for-bug implementation)…
This is not called “choice” by the oft-mentioned “reasonable user”, I’d say.
Everyone who chooses to not use Chrome on a decently spec’d machine is “paranoid”?
Where did it say “Google makes you install their browser”???
It’s not biased to confirm a reality that most people I know is using google chrome on their PCs so of course google isn’t forcing you to ‘install Chrome’ but it does force their updates in the background the last time around I was using it.
This is why most of my traffic is on Brave & Opera browsers. Also not saying that those browsers will save me from the LLM but still.
Let’s me intellectually honest please
I’m not sure this is a bad thing. The biggest ethical problem with LLMs is that they’re under lock and key of big tech. All your information is therefore filtered and controlled by them. If they make usable local LLMs the technology to do so will eventually leak. Once that happens the open source community will run with it and we may finally see an end to the current stranglehold big tech has on information.
we know how they work and what the technology is. Do you have a spare datacenter to run one on?
I’m running the GLM-4.7-Flash model on a 24GB RTX 3090 card I picked up used for $700. Using it to help with programming tasks and it’s working well. Zero hallucinations (when you set your context prompt properly AND just a handful of tunable model parameters) so far. Getting great speeds too … usually around 100 tokens/sec. No more online AI model usage for me.
I can’t afford another $700.
Working on it.
https://youtu.be/7DAPd5MGodY
There are LLM models available for local use from all the major players, what is not available is their core closed source models like Gemini. The only gatekeeper is how much resource your hardware has, this limits the size of a model you can run. Data centers have super nice GPU setups with something like hundreds of GB of RAM. These data center computers generate intense amounts of heat and burn as much electricity as a full sized appliance.
I use cloud AI bc my PC is a potatoe, I lack the hardware resource necessary to make it worthwhile. My privacy is also nonexistant on Google AI Studio without API keys (which cost per token), que sera. Dont put any personal info in your prompts.
You can already run loads of open source models on your own hardware. You are limited to lower size and performance models though and this model built into chrome will be no different, it will just be a small model that is only useful for a few things like auto correct and text predicting. It will be nowhere near the performance of cloud based LLMs.
Plus unless your device has an NPU it will be running the model on your CPU or GPU, on laptops with integrated graphics this is not really a good option and will severely reduce your battery life.
So… why are local LLMs an enevitability? What is the reason we cannot avoid them? Is there something inherit in tech that forces us to use them? Or is this waving a white flag and giving up on saying no?
Nobody says you have to be one of the people who uses one, but their being a big part of the upcoming computing world, yes, absolutely, an inevitability.
Are they inevitable for all use-cases? Do we need to throw out simple algorithms and instead feed everything through a model?
Maybe time to collect code from spell and grammar checkers.
Maybe future AI will reinvent the code to save power… lol
I think the Author could have been slightly more courageous in their reply. Why are they an inevitability? All* the main hardware and software vendors for “personal computers” on the market are heavily invested in “practical” uses for large language models. The biggest counterexample that I know of is Apple, even they are following the train somewhat. The inevitability is due to WHO is building the systems and writing the standards, not due to any technological advantage.
sure, there are some small vendors, but they can hardly control what companies larger than them can do
Hmm
https://techxplore.com/news/2026-05-silicon-oscillators-problems-thousands-years.html
I tend to agree that silently stealing computational resources from me is not behavior I want. If I have spare cycles, I would rather donate them to crowdsourced community science.
I know that caring about efficiency is declasse these days. Publishers are more concerned with getting features out the door then with performance, since modern processors run fast enough that customers mostly don’t notice sloppy code. Still.
If this can’t be turned off, or is difficult to turn off, it may push me off chrome.
Yeah… the behavior of Microsoft with regard to their flagship OS made me finally get rid of my last device using it. I had it installed because it had slightly better graphics drivers than linux when I installed it. I replaced it with linux as with my other devices and I’ve been much happier since. I’m thinking this is the last straw for chrome to me as well.
The ‘beyond your control’ part is the most concerning. Will they use it to process other data than the users, basically stealing infrastructure/power of the users? Will the LLM output be sent off to google?
I would guess they will not use it to process other users’ data (unless maybe your computer is ridiculously powerful and you leave it on all the time) due only to other users also having an LLM built in and it being more cost effective to do processing in bulk servers than to send data between users. But I could be wrong.
i would certainly be bummed out if i have to abandon a reasonably plausible ‘least common denominator’ browser
I wouldn’t call a 4b model that intensive. This is a model intended to run on phones.
You mean the device with the most amount of sensors that people own and always have on them?
…..my phone is easily one of the most powerful devices I own
Don’t worry, it’s Google. They’ll probably abandon support in a few short years.
They only do that to projects that cost them money to run and don’t generate enough profit. A local LLM costs them nothing to run. If they can use that feature to lure users (most likely their thinking), they can get more profit through their established channels. I don’t see them scrapping this for these reasons, but I appreciate the comedic sentiment. If people start abandoning Chrome, however, they may remove it.
Reminds me of when I got notified that my old phone couldn’t use some fast food ordering app (which was only installed because I wanted access to some secret menu items; the servers crashed too frequently for me to see any convenience in it). A couple days later, the website (sorry, “web application”), which was supposed to be my next-best option, upgraded to use so much of my local processing power that my phone crashed.
From a website.
But, hey, at least the “smart” recommendations and jiggling logos started causing the app and website to crash brand new phones the very next week.
I can’t help but wonder if there’s a general push to make sure that only the absolute newest devices can use the internet; and even then, only some of the time.
i think it’s worse than that. i think usability has become an afterthought even for top-end devices.
Only the web botchers of today, they sit in front of their fancy Macbook Pros that do all this with ease and never even try to run their site on a ten year old PC or five year old phone, the average lifespan of such a device for the average Joe and Jane.
This is HaD – why wasn’t the article about how to turn it OFF or uninstall it and still have chrome? Or do we have to uninstal chrome in it’s entirety to get rid of it and put an old version on with no updates instead?
And what about all those things that use Chrome V8? Are they impacted too?
The trouble here is that google have fundamentally changed the purpose of the browser that billions of people use…
I was wondering if this will impact Chromium.
Go to chrome colon slash slash flags, put “Gemini” in the search box, turn off all the switches.
chrome://flags/#optimization-guide-on-device-model
chrome://flags/#prompt-api-for-gemini-nano
I wonder if this is (the genesis of) a distributed training play. Why invest in giant data centers when you can get millions of little data centers for free.
Ugh gross. It would be much more acceptable if this was a plugin or flagged component one could turn on or off. I get it how it is explained here, there are just so many problems that could arise out of this mess and shoving ai into everything where it is not needed and running by default. I just liked it better when one big robot read all my gmail and passed it on to whatever body requested it lol. Now there is the bot in gmail, chrome, gemini in the phone, etc it is disconnected and convoluted. I hope this gets the attention and CVEs it will deserve eventually.
My only beef with this is that it’s not optional. But I’ve been hoping that there’d be Brower APIs to interact with an LLM for a while: It means WebDevs don’t have to do i18n manually anymore (which is a nuisance to maintain).
Okay, apart from the ethical and practical concerns of using an LLM to do internationalization, afaik this will NOT be available to web developers. Only Google.
Or maybe I misread? The future will tell.
Storm, meet teacup.
Google could have put in an non-LLM module to do the same thing (invent a technology if you like, this is a hypothetical) and nobody would notice. But as soon as “AI” is mentioned, people lose their minds.
as soon as you mentioned 4GB resource monster in a browser people lose their minds, irrespective it is AI or anything else..
RAM would start losing something if this became the trend.
And thanks to companies like Google, OpenAI, Anthropic, Meta, and Twitter (is the company called X now or just the service? forget it, I don’t care), RAM is considerably more expensive IF you can even afford it. They should have optimized before rushing to market. This situation is a direct result of greed.
“and since it’s available through a set of in-browser APIs we expect that it will find its way into a lot of websites, online applications, and plugins.”
Did anybody else read that as …
“and since it’s available through a set of in-browser APIs we expect that it will find it being leveraged by a lot of malware.”
No? Just me then.
No I thought much the same, Javascript or Flash prove rather good examples – might be useable for something actually desirable for the user, but also going to be used against them.
If it becomes an in-browser API unfortunately it may reach a point where you need to have an LLM into every browser in the same way as you need to have flash player or javascript to use most web sites …
My personal website is almost 100% text only. The exceptions are small jpegs of my past projects or some code you can download and use. Why fight all the hacking attempts when you can just prevent them? No PHP, no sql, no java, no javascript, nothing but HTML written by hand from my memory of the language back when HTML 1.1 was the newest thing. If no one wants to read my site, I don’t care. It is not my source of income, and no one is being asked for anything.
I would happily avoid websites that require such technology. The internet was a lot more fun before it got so flashy. If we build websites we would like to read and we visit those sites, who cares what the average social media addict does?
I would like to learn whether its intended use is as part of a distributed system. The Chinese don’t have easy access to the very powerful centralized AI systems we’ve been focusing on, developing instead a constellation of experts, a distributed model designed to give similar performance to western systems with less powerful (and less power hungry) hardware. The logical next step for this is to distribute the model using the techniques pioneered by SETI (which was in essence a large AI type project that searched for meaningful patterns in enormous data sets by using the surplus computing power of people’s desktop computers). Google has a lot of forward thinking people in it so if someone was to develop this capability (outside of China) then they would be my prime suspect.
The question I’d want answered is “Do I really want my system burdened like this (especially given the way its dragged down by ‘analytics’ anyway) and what are they going to do with it?”
You just gave me an idea. A system that performs that distributed compute via free online LLMs. Ask the LLM to process the data and then copy and paste back to Seti@Home or some other such project. If Google thinks getting systems they don’t own to do their compute work for them is perfectly fine and legal, why not do the same to them?
This means they also have experimented with later phases they want to roll out, like LLM interaction with your OS and out of sandox resources ie files. I wouldn’t be surprised if there is a way to elevate the LLM so it can “perform tasks” on your machine, like find and report all local private source code, documents, photos and other created media etc. Chrome has become a back door to your PC now for Google, either remove it, or hand over everything you think is private and own, to Google (assuming you havent already done so to Microsoft/Meta/Apple etc with Github, Cloud Sync etc etc).
LM-Studio, i bet my mom could install and use it, not much harder than using an office suit.
New fashion, wearable AI.
Anthropic did the same thing last week by having Claude Desktop quietly reach into every browser on the system and give itself a backdoor to the web. No one said anything, so Google did the same thing this week to their own browser. At some point, we have to stand up to large corporations and tell them no, that they have gone too far.
I can’t use your choices of library books to check out to determine if you’re a serial killer, why should I be able to get a hold of your LLM queries and not do the same thing? Google doesn’t need to do this “harvesting” on their browser and if you don’t stand up to it now, it won’t be the last software to take matters into its own hands.
Also, it DOES reach out, it’s not ONLY a local model :-) Anything it can’t figure out goes up to the big model. That’s a huge data leak.
Also, I think this article does a much better job of explaining why this is more than just a nuisance: https://www.thatprivacyguy.com/blog/chrome-silent-nano-install/
google, ever more unwanted intrusive. Apparently never dawns on the net.tyrant that they might ask if you want anything that the shove down your throat…
You all DO know that Google and Alphabet are funded by the feds to spy on the public, right?
Replicators and beserkers…
I hope someone is able to find the old Clippy code and hack in an LLM backend. Talk about full ironic circle. So sweet.