A follow-up on Firefox pushing a "labs" feature that gives you a sidebar to talk to an AI chatbot.https://mastodon.social/@mcc/113098293672475886
-
Safari recently announced an "Apple Intelligence" feature where the web browser, rather than being either
* A viewer for a formatted document (what browsers mostly are today)
* A presentation layer for the display of semantic content (what browsers originally were, what Safari becomes in Reader Mode)
The browser is a new, third thing
* A cuisinart that turns other people's writing into content slurry in-place
By way of a "summarization" feature.
Firefox has a tracking issue for just this.
-
Consider context. There are a variety of problems with LLM "AI". One of the many problems is that LLMs are, almost by necessity, grounded in cloud services, which creates a variety of secondary issues around surveillance, un-auditability and user dependency. In the past, when Mozilla has worked on machine learning, they have stuck to (and been widely praised for) a "local model" approach that avoids cloud services. See their translation feature, their (discontinued) text-to-speech engine, etc.
-
I have one contact from a Mozilla dev and a detail-free tracking issue to work from here. I can't know how this subject is being discussed within Mozilla and you should be clearly aware I'm speculating. But what I see here is, they view "summarization" as a component of the "AI chatbot" model which they can tear off and move into a local model, thus "solving" the privacy/safety problems.
Which alarms the heck out of me, as the other issues— licensing, environmental impact, and
️ lying ️
remain. -
This might never ship; most Mozilla R&D projects don't. But seeing this glimpse of potential next steps, my despair compounds.
As a person who writes things and hosts websites, I now must view Safari users as potentially the enemy, since I can't know if they're *reading* my website or using me as raw mulch to power the "Apple Intelligence" slop feed. I am staring down a future where Mozilla becomes a vendor of "Privacy-Preserving and Cost-Effective" lies, and Firefox users are the enemy too.
-
Let me state this in more direct terms: When you add the "AI summarization" feature, it is no longer a web browser. It is some other thing, some new thing, some new idiom of consuming content, some thing I do not want to be creating "content" for.
-
@mcc that's the same guy who's active in the "feedback on this feature" thread, using a lot of the same rhetorical tactics that the chrome team uses to dismiss criticism when they deliberately break some part of the web. "oh you think it's unethical? do you have any ethical LLMs you could recommend that we put alongside [the unethical LLMs we added without asking anyone]?" - that insistence that any critics need to be ready with their own alternative solution, obviating "y'all just shouldn't've"
-
@mcc as i said in that thread, i think they're just opening themselves up even further to be crushed by the entities with capital, who are the entire reason the current LLM hype wave exists in the first place (ie it's a way for them to monetize their data hoards) (and i don't buy the naive fossbro argument that open models will eventually "win" in this space, ever)
-
@jplebreton @mcc I think open models are important because they’re not about “winning” anything.
-
@polotek @jplebreton I think I would *potentially* find this argument compelling, but I find it hard to apply it because I have yet¹ to see a model under a license I would consider actually "open". It all looks like Hashicorp/Cockroach to me.
¹ With admittedly a non-exhaustive perusal of options
-
@mcc @jplebreton I hear you. I think the licensing can be solved as long as the tech is available and well understood outside of these corporate structures. And not yet patented.
That is the only thing we've got going for us tbh. The terrible version will still happen. But it's possible for us to build alternatives.