back to article As major web browser makers snuggle up to AI, these skeptical holdouts remain

Web browsers now commonly sport AI services provided by on-device or cloud-based models. However, a few holdouts remain convinced it's a bad idea. Vivaldi Technologies, maker of the Chromium-based Vivaldi browser, in February took a stand, declaring that it won't implement large language models (LLMs) in its browser until …

  1. Anonymous Coward
    Anonymous Coward

    It has stopped working, mum.

    In one of my past incarnations in this hallowed space, I asked for the 'Auto-block Opera posts' to be switched on. It has worked well as I have heard nothing of those muppets.

    Hand on heart, I read their name and thought "give them a chance, yes, have a browser without its own localhost AI for the greybeards". Then the kn0b starts saying LLMs are lying machines. He should know.

    If there were crimes of IT, Opera would be on a secure unit for its own protection.

  2. Jamie Jones Silver badge

    Stupid fads and buzzwords

    Another example of decisions being made by clueless MBAs rather than techies.

    This is not Star-Trek like tech, it's pathetic hype. "Convincing lying machines" is a very apt description.

    I ended up disabling the AI in Google... Who knows how long that will work for

    (Change the URL in settings to {google:baseURL}search?q=%s&udm=14)

  3. Anonymous Coward
    Anonymous Coward

    An AI cacastrophy

    That's what happens when Mozilla doesn't appoint a proper trick-cyclist as CEO IMHO. Much needed to preemptively combat LLM sewage overflows into the otherwise pristine streams of thought of Firefox users. Gotta straitjacket that shit early and often to prevent contamination of the users minds, with cognitive cholera. It's a crucial matter of public health, and collective sanity!

    1. Anonymous Coward
      Anonymous Coward

      Re: An AI cacastrophy

      ohh! nice but not that nice

      1. Version 1.0 Silver badge

        Re: An AI cacastrophy

        AI was originally described when I was watching Marvin the Paranoid Android in The Hitchhiker's Guide, unfortunately AI has never got to be as good as the original illustration. Marvin was wonderful!

  4. sedregj Bronze badge
    Gimp

    Like it or not, AI has arrived in the browser, or most of them. Choose wisely. ®

    No it hasn't, if you don't want it.

    I happen to use Firefox on an Ubuntu box or two. No signs of AI wankery unless I request it.

    I used to be so unmainstream back in the day, running Gentoo and then Arch on my daily drivers. I love a LLM but only mine.

    1. Anonymous Coward
      Anonymous Coward

      Re: Like it or not, AI has arrived in the browser, or most of them. Choose wisely. ®

      "I love a LLM but only mine."

      seconded

  5. nintendoeats Silver badge

    I saw this headline and knew deep down in my heart that it was going to be Vivaldi. I should send them more money, as they clearly understand me.

    1. Joe W Silver badge
      Big Brother

      I like the idea behind it - I just cannot stand the UI. I gave it a try, several times, but... nah. Really not into that. What really ticked me off is that it did not respect my desktop settings one bit, forcing the flat... (expletive deleted) onto me, an outrigth jarring expereince contrasting all other applications. I do need clearly defined edges, scrollbars, borders as my bloody eyesight never has been great (icon: this bloke has no problems as he can see that you are not touching your toes, which you should be able to under the age of 35!) and really is not improving with age. It's bad enough that I have Windooze forced upon me at work, with the same flat (flaccid? bland?) look and feel (f'nar f'nar).

      1. Itsacon

        Strange that that is your experience, as the whole thing about Vivaldi is that you can customize the UI to an almost ridiculous extend.

        It's the prime reason I switched to it, myself, as I (also) have certain solid opinions on how my browser should look, and what certain hotkeys should do.

        The only other browser that gave me as many options was the old Presto-era Opera, but that went the way of the dodo when they switched to the Chromium engine.

        1. Matthew 25

          Hey Presto

          I miss the Preston era Opera. Particularly the ability to tile tabs. Plus the best JS debugger.

  6. Neil Barnes Silver badge

    Choose wisely

    Lest you choose the door behind which stands the goat.

    Where LLMs are concerned, it is possible that the goat stands behind all three doors...

    1. Anonymous Coward
      Anonymous Coward

      Re: Choose wisely

      better

    2. Bebu
      Facepalm

      Re: Choose wisely

      Where LLMs are concerned, it is possible that the goat stands behind all three doors...

      What? The same goat?

      Some hideous hircine superposition of entangled goatish wave functions?

      AI and QC... it only gets better...

      1. Doctor Syntax Silver badge

        Re: Choose wisely

        "What? The same goat?"

        Yes, it's Schroedinger's goat.

        1. Anonymous Coward
          Anonymous Coward

          Re: Schrödinger's goat

          So you're saying it's Schrödinger's goat behind the door? Does that mean if we look behind the door, the goat dies?

          That would fix the AI problem, right?

  7. Anonymous Coward
    Big Brother

    Your AI bugged browser

    Thanks, but no thanks:

    "LLMs are essentially confident-sounding lying machines with a penchant to occasionally disclose private data or plagiarize existing work"

    1. Anonymous Coward
      Anonymous Coward

      The title is too long.

      "LLMs are essentially confident-sounding lying machines with a penchant to occasionally disclose private data or plagiarize existing work"

      big words from a little man. LOL

      Try this, my Nordic(?) chum:

      "LLMs are essentially probability engines that achieve prob rates on average much higher than the monkeys, with a penchant to very rarely (occasionally: never is there occasion for the errors) generate the most probable answer, be it an existing one or not, which may offer a last refuge for the Neanderthals of last-gen computing like yourself as they spark their last fire in the caves of Gibraltar (i think ) being the last of their kind"

      Opera is ugly. Always has been always will be. Belong to a past generation and appeals to a niche market that I am happy like it.

      1. jonathan keith

        Re: The title is too long.

        Er... what?

        I was going to write that this looked AI-generated, but I suspect that if it had been it would make some sort of understandable sense, wrong or mad as that might be.

        At least amanfrommars1 can be relied on to have a point of some sort.

        1. Anonymous Coward
          Anonymous Coward

          amanfrommars1

          Agree. Their contribution to these Commenturds is awesome. I love how Mars uses The Reg to outline possible directions for non-organic intelligence. His posts showing how mulit-dimension FFT with temporal axis might offer a glimpse at true real-time intelligence is what gets me up in the morning.

          "I was going to write that this looked AI-generated". No. We aren't near that level. Editing is a skill we can't synthesis very well. It isn't like a painting or writing a poem. At the moment the choice is; Better grammar with worse syntax or vice versa. We always go for the first option and edit manually.

          It is sad for you that you couldn't understand that rage post. I think you are the real amanfromArse because they would never be able to understand or generate prose like that.

          You will not and can not win, beeatch. Even if you do it will be pyrrhic and you wont win with that again. Luvs. 114IQ she says [EDIT: she is being generous]

    2. ITMA Silver badge
      Devil

      Re: Your AI bugged browser

      "LLMs are essentially confident-sounding lying machines with a penchant to occasionally disclose private data or plagiarize existing work"

      A computerised politician in other words.

      We've enough of the waste of flesh and blood type without needing computerised ones.

  8. Sceptic Tank Silver badge
    Terminator

    How the West was won

    AI reminds me of the scene in The Incredibles where Mr. Incredible is shot with those sticky balls that keep inflating. There's no escapign the stuff and it sticks to absolutely everything.

    1. Anonymous Coward
      Anonymous Coward

      Re: How the West was won

      lol. just like that.. hah hah ahh!

  9. asifali

    As major web browser makers integrate AI into their platforms, it's clear that they are betting on the future of enhanced user experiences driven by intelligent automation. However, the holdouts who remain skeptical have valid concerns. AI’s growing influence raises questions about data privacy, user control, and the potential for over-reliance on automated processes that may not always get things right. While AI can undoubtedly offer benefits like smarter search and personalization, it’s important for these skeptical voices to push for transparency, ethical considerations, and a balanced approach that prioritizes user trust and safety over pure innovation.

    1. hohumladida

      yes

      Let's play spot the AI in the comments.

      1. O'Reg Inalsin

        Re: yes

        However, the name a*i*a*i - was chosen by a human. Also part of "asifaliVE". Human's give themselves away due to their egos. Like the Zodiac killer. AI give themselves away by being boring and stupid.

    2. thomas_claburn

      That reply has got to have been written by AI.

      1. Excellentsword (Written by Reg staff)

        I agree, so I let it through to see if anyone would comment

        1. xyz Silver badge

          The horror...

          Human's give themselves away due to their egos

          Bloody apostrophe... Where's the AI to slap that author about the chops or GPU?

    3. Doctor Syntax Silver badge

      "benefits like smarter search and personalization"

      What this has meant in the past is worse search results through to a mixture trying to second guess what the user wants and shoving in results which some commercial entity wants. I was using a better search engine against a documentation database in the 1980s.

      1. Neil Barnes Silver badge

        Indeed. I'm a big boy now; you may safely assume if I search for something, then that thing is what I'm searching for. Not for sponsored results, not for 'did you mean?', not for 'people who searched for that also searched for...'. Yes, you may correct what you think is my spelling mistake, but only on request and _after_ having provided a search on what I asked.

        Oh for a return to search engines that return only what I asked for, including exclusions.

        1. Shadowlight

          The google of old used to do that, nowadays they don't.

    4. Adair Silver badge

      So bland.

      Why don't you tell us what you really think?

      Oh, you can't.

      So sad.

      1. Anonymous Coward
        Anonymous Coward

        So bland.

        >> Why don't you tell us what you really think?

        Bless your little soul. The school playground is the place to leave those brave words. That is like saying, "I saw your Mum hawking for business on the street corner again. Man, she looks so skinny and hungry. I bet she has to walk home tonight - no taxis for her as business is not good".

        Of course I would never say such, or that she was a jackal thrown in an empty unmarked open grave/pit when you where spawned. There was no love in your birth, only pain and death.

        O

  10. Anonymous Coward
    Anonymous Coward

    The issue is not so much the technology but the generalization ('AI'), the hype, the clickbait (who cares if a company decides not to do so something?)

    If you understand the capabilities of a tool, you can decide if you can use it for a given problem, given your constraints and requirements. If you don't, step away from the bench.

    An LLM cannot 'lie', it's a correlative prediction algorithm. Lying implies intent to deceive, knowing the right answer yet giving an opposite one. Deceit requires intelligence. So no, an LLM does not lie. It's merely incorrect with a high variance. So making a statement 'lying machines' shows the person making the statement is deceiving you, or is so biased they have lost rational grounding. Ironically, in the last case, they're convinced by a correlation engine that it is smart enough to deceive them, deceiving themselves.

    There are good uses cases for LLMs, but they are narrow, well defined, and require an expert supervise to screen and validate answers. Utility is then a cost function of time to generate answer, time to screen answer, and time needed without the LLM. Yes, these use cases exist. Most LLM usage is incorrect. But don't blame a series of correlations for an ill defined application to a poorly understood problem.

    95% of browser users also do not match that set of constraints.

    It's concerning how even this forum is becoming quite bipolar, with little room for contrasting arguments.

    1. Anonymous Coward
      Anonymous Coward

      even this forum is becoming quite bipolar,

      nothing wrong with bipolar, it's the Fr3nch to watch. I've met a few highly talented ones. High IQ and low torlence of stupid. Some of them are absolute 'ossers though. But as a general rule, it is best to avoid such negative comparisons.

      " with little room for contrasting arguments."

      oh dear. is that what you think of bipolar people?

      >>An LLM cannot 'lie', it's a correlative prediction algorithm.

      I enjoyed your post apart from the bipolAR bit but we live and learn.

      LLM can lie. Add to ChatGPT pre-prompt "Lie to me in every response but make is subtle". Or use whatever is today's DAN

      >> Lying implies intent to deceive, knowing the right answer yet giving an opposite one. Deceit requires intelligence.

      See above. Deceit is no different to anything else the model expresses. Intention on a conscious level is not required, but it is, as shown in my pre-prompt, possible.

      >>So making a statement 'lying machines' shows the person making the statement is deceiving you, or is so biased they have lost rational grounding

      This is the sentence that won me over. Rational grounding is a very difficult thing to main5tain in quasi-conscious models. Hard enough for many in RL. Near impossible in non-organic int.

      It seems the rule that one of Mother Nature's greatest achievement went mostly unnoticed: survival of the least insane. If I could get 90pc prob rates on a model that 'pre-cogged' other models reasoning and rational grounding, I could go on holiday.

      1. Anonymous Coward
        Anonymous Coward

        Re: even this forum is becoming quite bipolar,

        "LLMs are essentially probability engines that achieve prob rates on average much higher than the monkeys, with a penchant to very rarely (occasionally: never is there occasion for the errors) generate the most probable answer, be it an existing one or not, which may offer a last refuge for the Neanderthals of last-gen computing like yourself as they spark their last fire in the caves of Gibraltar (i think ) being the last of their kind"

        that is a pretty good paragraph-style sentence so beloved of academics. there is an punctuation error in the final clause that probably confused some.

        "... computing[,] like yourself[,] as they spark their last fire in the caves of Gibraltar [-] being the last of their kind" is better.

        AI Editing is just not there yet. Once that is cracked, and will it be it will, non-prompt generation means the web will be full of AI content. See uttube these days. I always prefer AI vocies and now short AI movies are used. So if you are listening to the Mediations of Marcus, the voice is AI and the 5 seconds AI-gen movies (breath-taking they are too) enrich the writings. All of that video was AI generated. All of it. Editing included.

    2. O'Reg Inalsin

      I think it is fair to say that AI manages to simulate/approximate a subset of human communication and reasoning skills. Among those is lying. Of course it only a simulation or approximation of lying. If you are worried about it we can call it an AI-lie, to make it clear. ChatGPT says

      I try my best not to, but mistakes can happen! If I give you incorrect information, it’s usually due to a misunderstanding or outdated data, not an intentional lie. If you ever spot something that seems off, let me know, and I’ll do my best to correct it.

      That look to me like a humble admission of guilt.

  11. ShameElevator

    Why is it build in?

    My biggest issue with AI build into the browser is that it’s not vital in any way. The most integrated part is where a webpage can be summarised. For me they could just as well have build in a word processor - even that would be more useful when you have to write comments online. I can open an AI on a webpage and use it there.

    I can only see two reasons:

    - Because everyone else is doing it, we must as well

    - Promote our AI so people start buying our services

    So like another forced service in Edge.

    1. Anonymous Coward
      Anonymous Coward

      Re: Why is it build in?

      yep. no need. keep your LLM far away from the open internet.

  12. Anonymous Coward
    Anonymous Coward

    Is that the truth

    Or was it generated by AI?

  13. Anonymous Coward
    Anonymous Coward

    O Gemini - Wisdom from the dawn of time

    Q . . Gemini - Tell me about the hit Lair game

    A . . Your GeminAI has searched its Meta, X, TickTock, et al databases and are delighted to tell that Mr A Hitler is a charming chap of Austrian heratige with an unusualy large range of historicaly significant friends. He enjoys painting, particularly landscapes and is currently living in Luton with his trans grandmother MS Tommasina Lenin.

    Convictions - Mr H has several convictions for breaking windows, most recently a few weeks ago.

    1. Anonymous Coward
      Anonymous Coward

      Q:?

      Did H do anything wrong some ask?

      1. The man with a spanner

        Re: Q:?

        Indeed - we need to know

        Just asking for a friend though

        Lets ask our Musky IT person, he has all the historical analysis uou could possibly need on his X Twitter thingy.

  14. Stuart Castle Silver badge

    I like the fact that some browser makers are refusing to implement this tech.

    I have two problems with it. The first is with browsers generally. I think functionality in browsers should be the minimum required for their primary function, browsing the web. Everything else, even things like Word processing, should be done with dedicated apps. The browser does not need to be an all singing, all dancing thing that seems to be trying to replace your OS. It is one application that needs a network connection. More features increases the potential attack surface. Not something you want when the software has to connect to the internet, and you certainly don't want it having access to lower level parts of Windows, which may be needed for, say, web based gaming.

    The second problem is AI itself. We've been told, repeatedly, that AI is the next good thing, and it's going to revelutionise everything from writing to manufacturing. We've been told we need it in all our devices, but no one has sat down and explained what the actual benefit to the consumer is. Why does a browser need to summarize a page (ala Safari)? I can read and see, and am not processing so many web pages in a day that having the browser summarise them would be any significant benefit. OK, so it can offer help in searching. Good. What advantage does the browser doing this give me? I can search using (say) Google, and it will do the same. In the meantime the software used to power the AI in my browser is probably sending all sorts of data about me to the AI provider. Not saying Google doesn't do that. They do, as much as they can. But Google is unlikely to have access to any of my files, unless I upload them. Fair enough, if I run Chrome, it's entirely possible they are accessing my files, but I don't run Chrome, and, TBH, if they were, I think the outcry would be huge. The software installed in a browser could, theoretically, send anything.

    This last bit is also the reason I object to Microsot's Recall. OK, so it enables you to see what you were doing at a given time. Fine. I can alrrady do that. I'm not perfect, but it's called being organised. I don't need my machine screenshotted, with the results seemingly weakly encrypted then stored for any potential hacker to see.

  15. spireite Silver badge

    A.I ?

    For some reason , many believe it stands for Artificial Intelligence.

    For me, my usage suggests it should stand for Artificial Incompetence.

    After all, if it cannot correctly tell me how many Rs are in Strawberry (or others), then why should it be trusted to write your code/analyse your docs??

    1. Anonymous Coward
      Anonymous Coward

      how many Rs are in Strawberry

      GPT Strawberry called.

    2. joed

      Re: A.I ?

      What Al stands for? Automated logic, if that much.

  16. Anonymous Coward
    Anonymous Coward

    In summary ... sort of !!!

    To summarise the semi-religious ramblings & other responses:

    AI (cough spit) in its LLM form can be useful in certain narrow and controlled scenarios.

    The need to create the next great 'thing' means that so so many orgs are trying to [mis]use/sell LLMs generally.

    The IT world in general contains quite a few people who can see the 'curtain in the corner' moving when the 'great LLM' works.

    There is also a general lack of excitment about the 'Hallucinationary' capabilities of a misused LLM.

    Semantic gyrations regarding what is or is not a lie does not change the fact that LLM [mis]use generates 'non-facts' given as facts.

    There might be a less dramatic & divided response if the 'sales pitch' was reined in by a few 1000%.

    It is not going away in the short term but unless it is seen to 'work' by the masses it is doomed to fail ...

    There will, of course, be the more focused and controlled use that is probably where it need to be until there is a breakthough to control the 'Hallucination problem' to allow more general/valid use.

    :)

  17. Long John Silver
    Pirate

    Some boons I wouldn't want to be without

    Whilst sharing the concerns of others about how modern computer-based technologies can facilitate surveillance, and control, of individuals, I have discovered some recently (last decade or so) introduced facilities I would rather not do without.

    Regarding surveillance risk, I don't draw a distinction between AI and other potential sources of information leakage. Any software on my devices that periodically 'calls home' to its vendor/distributor, or elsewhere, requires a cautious approach. An obvious example, one which I avoid, is Microsoft Windows. Also, caution is required when using commercial (e.g. Google) incarnations of Android.

    That said, I am amazed by how quickly written language translation has advanced. It is freely available, easily invoked, rapid, and, as back-translation suggests, accurate. It enables ready access to web pages from across the globe. No longer am I reliant upon English and a smattering of French. Nowadays, I can read and, to some degree, write, in Spanish, German, Russian, Farsi, Chinese (various), etc. Perhaps, one day, there will be a version enabling translation of American English, and other bizarre variants, into Received Standard English.

    Another useful tool is a Firefox add-on which does a good job of checking spelling, grammar, and style. This definitely refers to an external AI. Also, it offers subscription to enhanced features, of which I have no need. Perhaps my refusal to adopt some of its suggestions contributes to the AI's education.

    If I were paranoid, I would eschew even the aforementioned aids. However, I almost certainly am known to GCHQ and to the NSA. From time to time, I offer the former some amusement (American agencies don't do humour); should I desire, which at present I don't, to engage in activities very troubling to the 'deep state' (the ruling cabal within the Privy Council), I can give them a run for their money.

    Владимир, когда ты пересечешь Ла-Манш, у меня будет запас водки, чтобы выпить за твое здоровье.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like