So it’s entirely optional, you only encounter it by interacting with the browser in an atypical way, and the thing it does is a thing that AI is actually pretty decent at (summarizing text)? Sounds like they couldn’t stop themselves from joining the dick hammering bandwagon, but decided not to hit it too hard.
This isn’t my wheelhouse so take it with a grain of salt, but an argument against link summarizers that I’ve heard is that it takes views away from websites that could be generating revenue for themselves. Instead an LLM scraped their content and fed a summary directly to the user.
Do people making that argument also find ad blockers even ten percent as horrible as this? They both ultimately have the same effect, which is your web browser not maximizing someone else’s profits by denying them a revenue opportunity.
I’d be curious if the link summarizer in Firefox runs a model locally or calls some remote API. Most current machines ought to be able to run an appropriate LLM model for that task.
That’s a good question. I’d personally argue that it’s different in that the adblockers are not an inbuilt part of Firefox, they’re made by extension developers. This is built right into the browser.
It’s more of a concern with google summaries that show up at the top of search results because it completely removes the need to click on any of the websites it pulls from. Ideally a link summary just lets you figure out which link you need without clicking on and looking at each one.
Thanks for adding that detail, makes sense. I was thinking the link previews were yeah more like the google results. Hover over and not have to go to the site.
No, it’s on by default, it has access to everything you’re browsing and doing fuck knows what with it, and you need to know that it is doing it and unless you’ve read it somewhere you don’t know that it’s there.
thing that AI is actually pretty decent at (summarizing text)
That’s so not true it’s not even funny.
I mean yeah, it could be worse, everything could be worse. Still, not good, not good at all.
you only encounter it by interacting with the browser in an atypical way
Leaving my pointer on a random place that happens to be a link is atypical? I don’t think it is. I had this pop up to me a couple of days ago and I didn’t even understand what could’ve triggered it, I was wondering if I clicked something or pressed a key unconsciously.
and the thing it does is a thing that AI is actually pretty decent at (summarizing text)
Pretty decent? Just passable, if the text is about some run-of-the-mill topic.
Doesn’t do that for me. I have to hold left click on a link for over a second to trigger it.
And yeah, pretty decent. It can produce a basically summary of a fair amount of text pretty quickly and generally accurately. It’s not an expert wordsmith, it won’t give a deep and thoughtful analysis of the poem you pointed it at or anything, but that’s not the use case. The use case is “give me the key bullet points of this article so I can decide if I should give it more attention.”, and it does that job pretty well.
This is absolutely, demonstrably, documentedly not true. It is accurate sometimes, and sometimes it shows you absolute bullshit. When will it be accurate? Who knows. So you can use it only when you don’t care about the truth, in which case why even bother, just imagine the article said what you wanted it to say and be done with it
Depends on model tuning. Basically, you can tune a model to hallucinate less, or to write more human-like, but not really both at the same time, at least not for a model you could expect most users to run locally. For this sort of application (summarizing text), you’d tune heavily against hallucination, because ideally your bullet points are going to mostly be made up of direct paraphrase of article text, with a very limited need for fluid writing or anything even vaguely creative.
Basically, you can tune a model to hallucinate less
You can tune it to hallucinate more, you can’t tune it to not hallucinate at all, and that’s what matters. You need it to be “not at all” if you want it to be useful, otherwise you can never be sure that it’s not lying, and you can’t check for lies other than reading the article, which defies the whole purpose of it.
Doesn’t do that for me. I have to hold left click on a link for over a second to trigger it.
I misunderstood the previous comments, actually, yeah, it’s not triggered the way I assumed.
The use case is “give me the key bullet points of this article so I can decide if I should give it more attention.”, and it does that job pretty well.
I’ll put aside all the other complaints I have on my mind, because we’ve both probably gone through similar discussions, I don’t want to get bogged down in yet another, and just say that I honestly can’t imagine this being such a useful or time-saving thing in the first place. Like, did it use to be a frequent problem to you to start reading an article, realise you’re not interested, and give up on it?
So it’s entirely optional, you only encounter it by interacting with the browser in an atypical way, and the thing it does is a thing that AI is actually pretty decent at (summarizing text)? Sounds like they couldn’t stop themselves from joining the dick hammering bandwagon, but decided not to hit it too hard.
This isn’t my wheelhouse so take it with a grain of salt, but an argument against link summarizers that I’ve heard is that it takes views away from websites that could be generating revenue for themselves. Instead an LLM scraped their content and fed a summary directly to the user.
Do people making that argument also find ad blockers even ten percent as horrible as this? They both ultimately have the same effect, which is your web browser not maximizing someone else’s profits by denying them a revenue opportunity.
I’d be curious if the link summarizer in Firefox runs a model locally or calls some remote API. Most current machines ought to be able to run an appropriate LLM model for that task.
That’s a good question. I’d personally argue that it’s different in that the adblockers are not an inbuilt part of Firefox, they’re made by extension developers. This is built right into the browser.
It’s more of a concern with google summaries that show up at the top of search results because it completely removes the need to click on any of the websites it pulls from. Ideally a link summary just lets you figure out which link you need without clicking on and looking at each one.
Thanks for adding that detail, makes sense. I was thinking the link previews were yeah more like the google results. Hover over and not have to go to the site.
No, it’s on by default, it has access to everything you’re browsing and doing fuck knows what with it, and you need to know that it is doing it and unless you’ve read it somewhere you don’t know that it’s there.
That’s so not true it’s not even funny.
I mean yeah, it could be worse, everything could be worse. Still, not good, not good at all.
Leaving my pointer on a random place that happens to be a link is atypical? I don’t think it is. I had this pop up to me a couple of days ago and I didn’t even understand what could’ve triggered it, I was wondering if I clicked something or pressed a key unconsciously.
Pretty decent? Just passable, if the text is about some run-of-the-mill topic.
Doesn’t do that for me. I have to hold left click on a link for over a second to trigger it.
And yeah, pretty decent. It can produce a basically summary of a fair amount of text pretty quickly and generally accurately. It’s not an expert wordsmith, it won’t give a deep and thoughtful analysis of the poem you pointed it at or anything, but that’s not the use case. The use case is “give me the key bullet points of this article so I can decide if I should give it more attention.”, and it does that job pretty well.
This is absolutely, demonstrably, documentedly not true. It is accurate sometimes, and sometimes it shows you absolute bullshit. When will it be accurate? Who knows. So you can use it only when you don’t care about the truth, in which case why even bother, just imagine the article said what you wanted it to say and be done with it
Depends on model tuning. Basically, you can tune a model to hallucinate less, or to write more human-like, but not really both at the same time, at least not for a model you could expect most users to run locally. For this sort of application (summarizing text), you’d tune heavily against hallucination, because ideally your bullet points are going to mostly be made up of direct paraphrase of article text, with a very limited need for fluid writing or anything even vaguely creative.
You can tune it to hallucinate more, you can’t tune it to not hallucinate at all, and that’s what matters. You need it to be “not at all” if you want it to be useful, otherwise you can never be sure that it’s not lying, and you can’t check for lies other than reading the article, which defies the whole purpose of it.
I misunderstood the previous comments, actually, yeah, it’s not triggered the way I assumed.
I’ll put aside all the other complaints I have on my mind, because we’ve both probably gone through similar discussions, I don’t want to get bogged down in yet another, and just say that I honestly can’t imagine this being such a useful or time-saving thing in the first place. Like, did it use to be a frequent problem to you to start reading an article, realise you’re not interested, and give up on it?
*yet
The new CEO is an AI True Believer, and I don’t doubt this will last.
They’ll remove the options to turn it off, and make it full of the AI features, with no opt-out or opt-in
Holding my click on a volume slider triggered the feature so it quickly became really annoying