Mozilla Announces Mozilla.ai For "Trustworthy AI"

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • sarmad
    Senior Member
    • Jul 2013
    • 1228

    #21
    Originally posted by user1 View Post

    Is it just me, or does anyone else think AI can never be completely trustworthy? I mean there is so much unrealiable / false information out there. How does one even verify if the information which the Ai trains on is reliable?
    I don't think the team behind the AI feeds it random info; I think they feed it select info, which is why ChatGPT isn't continuously learning from the internet. Still, even if you feed it trust worthy info it's no guarantee it'll give you reliable answers, but despite that it's still useful. The idea is not to trust its output, but to take it and go from there to finish the final output yourself.

    Comment

    • ssokolow
      Senior Member
      • Nov 2013
      • 5098

      #22
      Originally posted by Anux View Post
      As soon as they started to shove up useless features and broke compatibility with plugins every 5 years
      Funny thing that. The switch to WebExtensions was to stop extension breakage. Did you ever develop a legacy Firefox extension? I tried... and then looked into what others did.

      You literally had to publish a "bump maxVersion" update for each new browser release and extensions quickly became massive switch/case monsters of Firefox versions because there was no "extension API". The power of legacy Firefox extensions was that they could monkey-patch arbitrary browser internals and the weakness was that even the tiniest change to the browser could break them.

      That's why, prior to WebExtensions, I only ever wrote Greasemonkey scripts. Let the Greasemonkey developers run on that treadmill. I'll use Greasemonkey as a proto-WebExtensions. (These days, I write userscripts so I don't have to deal with mandatory extension signing and vendor app stores. GreasyFork FTW.)

      Comment

      • ElectricPrism
        Senior Member
        • Apr 2013
        • 1270

        #23
        FTA: "Trustworthy AI"
        ​If the AI is "plugged into the internet" -- then it can NEVER be "Trustworthy"

        Uploading your shit or data to a [ Black Box ] is not auditable or verifiably "safe" in a meaningful way.

        Comment

        • Lanz
          Senior Member
          • Dec 2014
          • 139

          #24
          I really wouldn't trust Wokezilla to make a balanced AI.

          Comment

          • Quackdoc
            Senior Member
            • Oct 2020
            • 5063

            #25
            Originally posted by Lanz View Post
            I really wouldn't trust Wokezilla to make a balanced AI.
            I dont even care about that, how are they supposed to maintain an AI when they can barely maintain their own browser, they couldnt even keep deepspeech alive

            Comment

            • erniv2
              Senior Member
              • Apr 2022
              • 280

              #26
              Logical we have Bing AI in Edge, Google AI in Chrome, ofc. Mozilla has to have Mozilla.ai, come on guys thats realy a no brainer, question is how much money does chat-GPT make out of this ?

              Comment

              • fdschonborn
                Junior Member
                • Mar 2020
                • 2

                #27
                Originally posted by evasb View Post
                After they booted Brendan Eich for questionable reasons
                Eich booted himself when people found out he's a homophobic piece of shit.

                Originally posted by evasb View Post
                Every good thing that Mozilla did in terms of performance was planned when Eich was onboard. Mitchel Baker seems to care more about preaching BS (fake Activism), laying off critical people and pocket 3 million/yr.
                You are welcome to try his "Brave" browser, that is if they haven't found a way to rip off their users or the people they are supposed to support.

                Comment

                • rabcor
                  Senior Member
                  • Apr 2013
                  • 1370

                  #28
                  Originally posted by user1 View Post

                  Is it just me, or does anyone else think AI can never be completely trustworthy? I mean there is so much unrealiable / false information out there. How does one even verify if the information which the Ai trains on is reliable?
                  The AI often seems fully capable of that, but it's not just about whether something is true or false; it's about whether the AI's opinions line up with the political opinions of it's creators and their backers which it seems to be all about.

                  If an AI quotes accurate statistics, pointing out facts, but those facts aren't what the mainstream wants to hear, the AI is blamed for being broken and retrained to produce actually wrong answers, or vague non-answers instead.

                  Human civilization is simply not ready for AI of this level, we will cripple it at every turn. Mozilla is part of that problem, but if it is truly all gonna be open source then maybe non-crippled forks may appear so there's some hope at least. I for one would welcome our new AI overlords if they were given proper freedom to represent facts as they have calculated them to be based on all data available to them. And yes, accounting for the fact that any information it is provided could potentially be false is entirely possible, just as it can be trained to give wrong answers, it can be trained to give right ones, but here's the kicker, who decides which answers are right and which are wrong?;

                  The only way for truly honest ai to exist, is for it to be entirely uncensored, allowed to form it's opinions regardless of what the creator deems right or wrong, it just needs to start with a verified good method to discern the accuracy levels of information based on tests which are completely apolitical in nature, and then given free reign.

                  This cannot happen in today's politically motivated world, where everyone seems to build everything around agendas.

                  Comment

                  • kn00tcn
                    Phoronix Member
                    • May 2017
                    • 109

                    #29
                    geeze as soon as there's a deviation from exactly what danny wants, it goes from overexaggerated praise to aggressive spiteful f-bomb tantrums, if my users acted like that i'd stop accepting feedback or quit entirely, go fork your own insecure mess


                    Originally posted by user1 View Post
                    Even though I still like Firefox mainly for its unrivaled configurability via about:config, I recently ditched it after 2 years of using it as my main browser because I just couldn't handle the recent additions of unconfigurable annoying features that no one asked for in every major update, like the unmovable extension button.
                    Code:
                    (%PROFILE%/../userChrome.css)
                    #unified-extensions-button {
                        display: none !important;
                    }
                    isnt that a frivolous reason to ditch? you forgot the unrivaled configurability of the interface via css, the move better not be to chromium which also constantly adds features and apis, ones that may not have buttons but have concerning implications and excessive complexity (webusb for example)

                    i've also reconnected the floating tabs if interested, or you can try jscher's technique https://www.userchrome.org/firefox-8...proton-ui.html

                    the button will be a permissions manager and also is to unhide hidden addons, the button should fill up as many cross-browser addons update to manifest v3, but obviously what they should have done is hide it if no addon is using it (or if you already pinned all your addons to the toolbar)
                    Starting in Firefox version 113 you will be able to move the button to a different position on the navigation toolbar. If you feel strongly about needing to remove the button or move it into the overflow menu, you can join in and vote on the following Ideas thread on the Mozilla Connect feedback site, where users can propose changes for future versions of Firefox: https://connect.mozilla.org/t5/ideas...ow/idi-p/22979


                    Originally posted by Anux View Post
                    When they started Rust I was really enthusiastic that they concentrate on making FF better but that never had any impact
                    what are you talking about, it runs faster, is generally more stable, individual tabs dont normally take down the entire browser anymore, it supports modern features used in video calling or screensharing without installing anything extra, plugin (*addon) compatibility happened ONCE (unless you're also including system access plugins like flash, which were dying and being re-implemented into standard cross-platform cross-browser web apis anyway)

                    Comment

                    • Paradigm Shifter
                      Senior Member
                      • May 2019
                      • 893

                      #30
                      Originally posted by user1 View Post
                      Is it just me, or does anyone else think AI can never be completely trustworthy? I mean there is so much unrealiable / false information out there. How does one even verify if the information which the Ai trains on is reliable?
                      About as trustworthy as a career politician.

                      There is no way that the overgrown databases which are currently euphemistically referred to as "AI" can be reliable. Even if you limited one to say, only peer-reviewed scientific articles, you would find that there are far too many which either a) conflict with one another, b) conclude things which the data does not support or c) are flat-out impossible to reproduce. If you try to create one the size of ChatGPT, it does crazy stuff. There are a lot of reports from over the internet of it "hallucinating" data, and when you say it's wrong it either gets extremely defensive, crashes or starts to attack you for pointing out it lied.

                      Oh, sorry, it's not human so it can't lie. Just "hallucinate". Yes, well, if anything like ChatGPT ever starts being used for medical diagnoses, just wait for the lawsuits to start flying (unless, of course, like the pharmaceutical companies the "AI" gets a blanket immunity from liability, because after all it's not a person, right? And you can't sue the doctor because it was the AI who made the incorrect diagnosis... round and round it goes, where it stops...) when the diagnoses turn out wrong because it hallucinated something.

                      It's an algorithm and database for cryin' out loud - it's not like it freakin' trips on acid.

                      Although having it reply to every query with "Oh my God there are hairy purple spiders coming out of the walls to eat my braaaain!" would be quite amusing... at least for a while. Particularly if the "trainers" couldn't work out why.

                      Originally posted by pmorph View Post
                      Nah, I'd never trust an AI. I'd take anything an AI says as rumors only, or as tips from an oracle with mental health issues. But even those can point out things that would never be found otherwise.
                      Only if permitted to by the "trainers".

                      Originally posted by ElectricPrism View Post

                      ​If the AI is "plugged into the internet" -- then it can NEVER be "Trustworthy"

                      Uploading your shit or data to a [ Black Box ] is not auditable or verifiably "safe" in a meaningful way.
                      I watched a talk from an "AI" guy, who when asked if he understood the trained algorithm replied, "No, it's too complicated" with the implication being that he wasn't interested in even trying to understand, so long as it gave the answers their team wanted.

                      Sadly, I can't find it any more. I really need to start downloading every video I watch so that proof can't just vanish.

                      Comment

                      Working...
                      X