From personal experience, most behavioural/mental/psychological disorders are just that, but condensed and amplified to extremes.
Having ADHD is like riding a rollercoaster and forgetting to get off every time it stops, then spending the rest of the loop distractedly wondering why and trying to figure out a way to get off it, during which you miss the exit again.
Why did I know it was a NileRed video from the tiny thumbnail? 🤔
“Remind me later…”
It’s a vibe.
The post is about the new AI Window feature, and here’s its description from Mozilla’s own announcement:
Now, we’re excited to invite you to help shape the work on our next innovation: an AI Window. It’s a new, intelligent and user-controlled space we’re building in Firefox that lets you chat with an AI assistant and get help while you browse, all on your terms. Completely opt-in, you have full control, and if you try it and find it’s not for you, you can choose to switch it off.
So, definitely agentic.
And I think a lot of the “anti AI folks” here are also angry about recent non-agentic AI additions that Mozilla added, such as e.g. tab recommendations for tab groups.
I’ll speak honestly: I’m not happy with that either. But that’s incidental to this topic. The agentic component is my biggest concern.
That’s a fair point. But search algorithms fall under the “recommender system” umbrella, which are a very different family from the agentic AI we’re discussing here.
Both are AIs, but with very different use cases. In the same way, you could technically classify both your gaming PC and phone as computers, and you’d be correct.
Edit: links
A disabled service will not consume any RAM. People just look for excuses to hate on it because it has AI in the name.
First it was that it would consume 4GB of RAM in the background. Then that it might reserve the memory just in case. Now the issue is that even if it didn’t, it would make the binary larger lol.
A larger binary maps 1:1 to a larger memory footprint in most cases.
Disabled code is still part of the process address space, which is pre-allocated on startup for every instance of the Firefox process that’s launched[1]. This affects both startup time and memory usage, and during startup, it depends on disk i/o speed as well.
AI is not a trivial feature. It will need a lot of dependencies, each library will also be loaded to the address space on startup, and will recursively load its own dependencies into memory. Factor these as well.
Now, how many subprocesses does Firefox launch during a typical session?
The Firefox multi-process model launches a minimum of 4 background host processes by default last I checked (Quantum), maybe more for extensions and other things, and a dedicated GPU rendering process if DMA-BUF is enabled if I remember correctly. (Can someone verify?)
Multiply that process count by the size of the new code and dependencies they pull and you’ll get a not-so-insignificant number, especially considering the complexity of the feature in question.
And aside from the aforementioned issues: a disabled service at runtime can still mean thousands of CPU instructions wasted everywhere checking whether it’s enabled or not. Depends on how tightly integrated it is. [2]
I sure hope you also complain about all the disabled drivers that ship with your Linux kernel.
I certainly would if they all got loaded every time my system started up.
[1] Unless the code is linked dynamically and loaded on demand at runtime via dlopen or the equivalent on Windows. I won’t claim extreme familiarity with the codebase, but I’ve worked with Firefox sources in the past while testing SpiderMonkey in an old project, and from my experience, that’s not the typical method used in the code and is only used to load some system and third-party libraries (such multimedia codecs, etc) that aren’t referenced at link-time.
[2] if it’s not very tightly integrated then it shouldn’t be in the main binary to begin with. If it’s as useful as they portray then it will have checks sprinkled everywhere.
Are you purposefully trolling or just that level of ignorant? I’m done responding here regardless, just curious.
What does being opt-in or opt-out have to do with resource consumption? It will still consume resources even if you’re not using it. It might even reserve memory while inactive for faster startup by default.
Even if you completely disable it, just having it built into the default binary already consumes resources.
So much whining… Please get to know the things you attack, before you embarrass yourself.
I’m a senior full-stack engineer at a recruitment startup. I’m the one responsible for the AI side of things. We do document analysis of résumés and all kinds of identification documents and photos. We also run various assistant chatbots that interact with users and accept file uploads, do job interviews and much more. I think I know enough not to embarrass myself.
Do you know what a “build in AI” is? It’s just some algorithms that makes your day easier.
No. A “built-in AI” is not a collection of “magic algorithms” that suddenly make life better. Working with AI is some of the worst hours I’ve ever spent of my life. Writing increasingly convoluted prompts to make it follow explicitly stated instructions for the N-th time in order not to misuse tools or output the correct structured data is not magic. It’s an exercise in futility most of the time.
And no one steals your data… sigh
It’s a local collection of algorithms, and you are quite safe.
So… You’re introducing an autonomous agent between you and the web you’re browsing. With the sole purpose of making decisions and taking actions on your behalf. And this thing is already proven to have the capability to HALLUCINATE.
Aside from all the privacy concerns and assuming that the models actually run locally; all code can have bugs, and worse: this “magic tool” is an amorphous collection of floating point numbers and you can’t debug a neural net’s weights.
If this thing has permission to freely interact with web pages, then sorry, your data will never be safe.
You probably already use it in so many ways you can’t count them - it’s just not called “AI” which I guess is the trigger word for you…
I think you’re the one confusing the terminology.
I exclusively use https://noai.duckduckgo.com/
So no.
I don’t see any technical background about it. Just a marketing spiel
I haven’t heard about it anywhere. But considering Firefox’s history with memory handling and considering the feature in question, I’m calling it. I’m reasonably sure that’s how it’s going to turn out
Ok. Then please answer this because it’s driving me crazy: what problem does adding AI to Firefox solve?
Except if AI Window turns out to be another resident subprocess that eats 4GB of ram in the background
Why not just distribute a separate build and call it “Firefox AI Edition” or something? Making this available in the base binary is a big mistake. At least doing so immediately and without testing the waters.
You had me you bastard
Isn’t this basically the banana cultivation problem of computing? Linux has pretty good genetic diversity with mutations and speciation happening on the regular, MacOS doesn’t have that variety, but is a genetically engineered abomination that’s regularly gene-edited to patch problems out.
As for Windows, I’d say it’s the Cavendish banana of operating systems.
Fair point. Although it made me feel called out more than most. :|