Firefox will have an option to disable all AI features

>"Shut up with the talk that makes people switch to Chrome".

No this is silly and you should dispense with it. Theres no cow so sacred it cant be criticised.

>Do you realize how big an LLM is? Clearly you don't.

Yes I do, I have run qwen 4b locally.

>The browser isn't going to fit on a lot of people's computers if they shove an LLM in.

The point I was making. Anything down this track is going to cause resource headaches, browsers are already a resource headache.

>We've been grabbing our pitchforks for years every time Mozilla even slightly steps out of line

I have been cool with literally everything else they have done, and even spent time pointing out that the T&C changes were relatively normal practice. Lumping me in with every other criticism you don't like is a genetic fallacy.

>It's a literal footgun.

Its community feedback. They ignore it at their peril.

> But don't listen to me look at what's happened in the past. Look at the comments here. Look at the comments in the past. FFS people were equating Mozilla accepting crypto donations with shipping a miner in the browser. It literally takes place in the Mastadon thread we're all talking about. Those things are wildly different and it is wildly a disingenuous interpretation.

I dont see how you are equating these. My post isn't in the Mastodon thread.

>Saying they're shipping LLMs

You literally quoted my "if", not reading and understanding how that word modifies the language around it is on you.

Theres really 2 states until there's an OS level LLM layer.

1. Shipping a low parameter/ otherwise compressed LLM.

2. LLM features are non local, with all the headaches that entails.

> And hey, if you feel I'm wrong here go jump on a fork that isn't going to add those things like Mullvad or Waterfox

I am required as a matter of employment to use an approved web browser 9-5. We have strict rules against AI use, only permitting copilot with enterprise data protection. (Using copilot without that stupid green tick, and letting customer data into it, is an instant RGE, same with any other LLM)

Another leg of this, is that I have 1000 profiles on 1000 computers for quite a few different customers. The nature of my work does not permit a single profile that roams to every computer I use. I cannot cross the streams between multiple customers.

I am not going to be asked to go to every PC and server, and opt out of AI features. I am likely to be asked to completely remove Firefox as an approved browser, pull it off every machine, and push only Microsoft Edge going forward. Because perversely, it auto signs in to Copilot with EDP via 365, if EDP has been purchased. So it can (sadly) do whatever the fuck it wants.

We are super sensitive to supply chain stuff, so are unlikely to be permitted to use any fork. We recently dismissed an addon for a product that was developed, by one of the core developers of the product, on his own time as an extra feature. Why? The guy didnt spend a lot of time maintaining it and seemed overwhelmed with pull requests. I haven't looked at the Firefox forks, but I know our security posture and its probably a non starter.

Even the idea that an unapproved LLM could scoop up what I am browsing is going to poison the well.

And like I have repeatedly stated, I like firefox. I want to use firefox. Part of the reason I want to use firefox is that it has no LLM instead of the approved LLM that I detest.

Even if Firefox goes "aha, we will implement our service to connect to Copilot with EDP" it would have to do so in a fashion where it is completely seamless, and zero touch, with no failback to regular copilot or another service. I cant conceive of Microsoft being that friendly.

We aren't alone in this. Lots of organisations are making similar decisions. I speak to other service providers going through the same headaches. Our contracts make it our responsibility to actively prevent customer data exfiltration, which is a basic feature and requirement of a lot of LLM implementations. As you stated, the browser LLM will want the browser window context. Microsoft (sadly) understands this, and knows that users will desire path to an LLM if one isn't available, and they understand what words we need written into the service agreement to make EDP attractive. Copilot is terrible, its also perfect for enterprise.

Firefox has over the last few years become a starkly friendly safe haven for privacy focused work. I feel, or at least felt, completely safe performing sensitive work with their product. Work being the keyword in enterprise. Things that become a toy or a risk go away. Chrome use in similar orgs to mine is decreasing for various similar reasons.

I am not sure if you think losing a decent chunk of enterprise users with sensitive requirements is worth it but at least you spent your time defending Firefox from criticism that might have retained those users? And when Firefox is just feature for feature exactly doing what chrome is doing, you might see new users show up for some reason? I don't see that as a winning strategy.

I really don't think they should play the ice cream vendor game at all. Let Edge and Chrome play in that space. Give firefox room to not suck. Let firefox be an actual alternative with significant points of difference not just an also ran.

You are right, machine learning models usually improve with more data and more parameters. Open model will never have enough resources to reach a compatible quality.

However, this technology is not magic, it is still just statistics, and during inference it looks very much like your run of the mill curve fitting (just in a billion parameter space). An inherent problem in regression analysis is that at some point you have too many parameters, and you are actually fitting the random errors in your sample (called overfitting). I think this puts an upper limit on the capabilities of LLMs, just like it does for the rest of our known tools of statistics.

There is a way to prevent that though, you can reduce the number of parameters and train a specialized model. I would actually argue that this is the future of AI algorithms and LLMs are kind of a dead end with usefulness limited to entertainment (as a very expensive toy). And that current applications of LLMs will be replaced by specialized models with fewer parameters, and hence much much much cheaper to train and run. Specialized models predate LLMs and we have a good idea of how an open source model fares in that market.

And it turns out, open source specialized models have proven them selves quite nicely actually. In go we have KataGo which is one of the best models on the market. Similarly in chess we have Stockfish.

At one point, paying for Mozilla's mullvad rebadge would give money to the corp. If you were already going to pay for a VPN, then it's effectively a donation.

Though, just because money goes to the corp, doesn't mean it will contribute to Firefox' development either.

After using projects named like "slurp", "eww" (combined with "yuck"), "yay", "honkers railway", "jason" and many many others, I personally kind of gave up on any attempts to judge projects by their titles. Partly due to many developers being whimsical nerds, partly because even marketable names say nothing about the product half the time so what's the point anyway

You and your argument are reductionist.