At Mozilla, we work hard to make Firefox the best browser for you. That’s why we're always focused on building a browser that empowers you to choose your own path, that gives you the freedom to explore without worry or compromises. We’re excited to share more about the updates and improvements we ha...
We’re looking at how we can use local, on-device AI models – i.e., more private – to enhance your browsing experience further. One feature we’re starting with next quarter is AI-generated alt-text for images inserted into PDFs, which makes it more accessible to visually impaired users and people with learning disabilities. The alt text is then processed on your device and saved locally instead of cloud services, ensuring that enhancements like these are done with your privacy in mind.
IMO if everything’s going to have AI ham fisted into it, this is probably the least shitty way to do so. With Firefox being open source, the code can also be audited to ensure they’re actually keeping their word about it being local-only.
Nope, they can use your NPU, GPU or CPU whatever you have… the performance will vary quite a bit though. Also, the larger the model the more memory it needs to run well.
With it being local it’s probably a small and limited model. I took a couple courses on machine learning years ago (before it got rebranded as “AI”), and you’d be surprised at how well a basic image recognition model can run on the lowest-spec macbook from 2012.
Tbh the inversion of typical intuition that is LLMs taking orders of magnitudes more memory than computer vision can mess people unfamiliar up on estimates of the hardware required
While I dislike corporate ai as much as the next guy I am quite interested in open source, local models. If i can run it on my machine, with the absolute certainty that it is my llm, working for my benefit, that’s pretty cool. And not feeding every miniscule detail about me to corporate.
I mean that’s that thing. They’re kind of black boxes so it can be hard to tell what they’re doing, but yeah local hardware is the absolute minimum. I guess places like huggingface are at least working to try and apply some sort of standard measures to the LLM space at least through testing…
Sums up my thoughts nicely. I am by no means able to make sense of the inner workings of an llm anyway, even if I can look at its code. At best i would be able to learn how to tweak its results to my needs or maybe provide it with additional datasets over time.
I simply trust that an open source model that is able to run offline, and doesnt call home somewhere with telemetry, has been vetted for trustworthiness by far more qualified people than me.
I tried one of their test builds. Seems like the AI part just means the browser can integrate with llamafile (Mozilla’s open source solution for running open source llm’s with just one file on any platform)
Sure, okay.
To each their own.
Whatever, it’s fine.
HOLLUP
IMO if everything’s going to have AI ham fisted into it, this is probably the least shitty way to do so. With Firefox being open source, the code can also be audited to ensure they’re actually keeping their word about it being local-only.
Don’t you need specific CPUs for these AI features? If so, how is this going to work on the machines that don’t support it?
Nope, they can use your NPU, GPU or CPU whatever you have… the performance will vary quite a bit though. Also, the larger the model the more memory it needs to run well.
The feature will obviously just be disabled on machines that don’t support it.
Running AI models isn’t that resource intensive. Training the models is the difficult part.
With it being local it’s probably a small and limited model. I took a couple courses on machine learning years ago (before it got rebranded as “AI”), and you’d be surprised at how well a basic image recognition model can run on the lowest-spec macbook from 2012.
Tbh the inversion of typical intuition that is LLMs taking orders of magnitudes more memory than computer vision can mess people unfamiliar up on estimates of the hardware required
You only need lots of precessing power to train the models. Using the models can be done on regular hardware.
While I dislike corporate ai as much as the next guy I am quite interested in open source, local models. If i can run it on my machine, with the absolute certainty that it is my llm, working for my benefit, that’s pretty cool. And not feeding every miniscule detail about me to corporate.
I mean that’s that thing. They’re kind of black boxes so it can be hard to tell what they’re doing, but yeah local hardware is the absolute minimum. I guess places like huggingface are at least working to try and apply some sort of standard measures to the LLM space at least through testing…
I mean, as long as you can tell it’s not opening up any network connections (e.g. by not giving the process network permission), it’s fine.
'Course, being built into a web browser might not make that easy…
Sums up my thoughts nicely. I am by no means able to make sense of the inner workings of an llm anyway, even if I can look at its code. At best i would be able to learn how to tweak its results to my needs or maybe provide it with additional datasets over time.
I simply trust that an open source model that is able to run offline, and doesnt call home somewhere with telemetry, has been vetted for trustworthiness by far more qualified people than me.
I tried one of their test builds. Seems like the AI part just means the browser can integrate with llamafile (Mozilla’s open source solution for running open source llm’s with just one file on any platform)
Focus on “local”. Mozilla is working since a while on that.