Conversation

Notices

1. Embed this notice
   b9AcE (b9ace@todon.eu)'s status on Saturday, 10-Jun-2023 16:34:11 JST b9AcE

   Is there something about Lana Del Rey that I didn't know but Google's YouTube knows? ;-D
   Seriously though, I very much miss the days when Google and its subsidiaries would just honestly say "we don't know of that existing where you searched for it".

   Also, that "Wisco Kidd"-account on YouTube is a self-labeled anarchist seemingly listed as a search result for having in their account-description a link to the video in question on a different website but the video per se hosted embedded at YouTube, which in turn has disabled that video, likely due to a bogus copyright-claim (AFAIK, if it's produced by/for Public Service BBC, it's public domain, which would be why the Internet Archive hosts the video, among others).
   Can a search result get more tangential?

   All cookies cleared before the search, so not related to any "personalization" or such nonsense.

   • Embed this notice
     b9AcE (b9ace@todon.eu)'s status on Saturday, 10-Jun-2023 16:34:04 JST b9AcE

     in reply to

     • undergrowth-feed ?

     @undergrowthfeed After reading most of the article (thank you!) and some Wikipedia before remembering that I don't actually care, I've now learned nearly ∞% more about that person, as my previous awareness was limited to trying some of their music over a decade ago but finding it boring so after that... not caring to check any further or again.
     I'd still think there must be some person, probably very many, that would be more "related to" my search than that person to become search result number 4 and 5 out of everyone and everything on all of YouTube, like... probably some person/organization that actually did banking with Nazi-Germany, for example.

     But again and genuinely, thanks for the link anyway. :-)

     clacke likes this.
   • Embed this notice
     undergrowth-feed ? (undergrowthfeed@achrilock.social)'s status on Saturday, 10-Jun-2023 16:34:10 JST undergrowth-feed ?

     in reply to

     @b9AcE as I understand it (and this is only heresay, mind), she has become somewhat of a darling to the fascist right. | https://conversations.indy100.com/lana-del-rey-new-album-cover-chemtrails

   • Embed this notice
     b9AcE (b9ace@todon.eu)'s status on Saturday, 10-Jun-2023 16:42:17 JST b9AcE

     in reply to

     • undergrowth-feed ?

     @undergrowthfeed Maybe. Maybe it'll just lie or hallucinate, as they apparently habitually do now.

     Yeah, I very much avoid that stuff myself, beyond like once every couple of years clicking on some video to see Jennifer Lawrence being excellently spontaneous or Aubrey Plaza being, well similar really. That's enough.

   • Embed this notice
     undergrowth-feed ? (undergrowthfeed@achrilock.social)'s status on Saturday, 10-Jun-2023 16:42:28 JST undergrowth-feed ?

     in reply to

     @b9AcE it's too bad we can't just ask the AI, like in chatbot format 'so, what was your reasoning, here, bro?!' Who knows? Maybe in the near future that'll be a thing? Anyway, this is why I try not to follow entertainment personalities too close on socials. What's the old expression? We should never meet our heroes? lol

   • Embed this notice
     clacke (clacke@libranet.de)'s status on Saturday, 10-Jun-2023 16:43:42 JST clacke

     in reply to

     • undergrowth-feed ?

     @b9AcE You probably know and understand what the terms mean, but in terms of word choice for the general public I've come to the belief that it's probably best to avoid anthropomorphizing statistical models.

     I used to think this was a non-problem and that calling statistical models "AI" was a non-problem, but now I see the headlines and discussions it generates online as people take the terms and run with them.

     Neural networks don't dream, hallucinate or lie, they just have parameters that yield surprising results when we stray outside their majority use cases.

     @undergrowthfeed

   • Embed this notice
     clacke (clacke@libranet.de)'s status on Saturday, 10-Jun-2023 16:43:43 JST clacke

     in reply to

     • undergrowth-feed ?

     @b9AcE You probably know and understand what the terms mean, but in terms of word choice for the general public I've come to the belief that it's probably best to avoid anthropomorphizing statistical models.

     I used to think this was a non-problem and that calling statistical models "AI" was a non-problem, but now I see the headlines and discussions it generates online as people take the terms and run with them.

     Neural networks don't dream, hallucinate or lie, they just have parameters that yield surprising results when we stray outside their majority use cases.

     @undergrowthfeed

   • Embed this notice
     b9AcE (b9ace@todon.eu)'s status on Saturday, 10-Jun-2023 16:43:44 JST b9AcE

     in reply to

     • undergrowth-feed ?

     @undergrowthfeed Well, yes. They are AFAIK created with the only directive of generating output that the submitter wants or at least seeks, which should reasonably lead to just lying and e.g. pretending that there is a research paper supporting the thesis someone uses the "AI" to find evidence for. The "AI" does its job there and it would be up to humans to constrain it to normally acceptable human behaviors such as not lying and not making suggestions that have clear high potential for being harmful.
     Those new thingies are after all just machine learning language models, not inherently capable of independent ethics and empathy.

   • Embed this notice
     undergrowth-feed ? (undergrowthfeed@achrilock.social)'s status on Saturday, 10-Jun-2023 16:43:46 JST undergrowth-feed ?

     in reply to

     @b9AcE it's like if you made an adult human without emotion or any past experiences at all, with a brain that was just built to find the color blue or the sound of traffic. It could do those things, but it would be very likely to find voices in the noise and 'likely' things that might be blue. It exists in the iterative, immediate 'now.' It has no frame of reference.

   • Embed this notice
     undergrowth-feed ? (undergrowthfeed@achrilock.social)'s status on Saturday, 10-Jun-2023 16:43:47 JST undergrowth-feed ?

     in reply to

     @b9AcE I actually have my own theory about the AI hallucination problem. It's like a collection of senses (sight, hearing, etc.) and the capability of speech, with a little bit of cold reasoning tying it together. But it has no will, no directives that are not arbitrary, and because of how they are trained they are taught to seek acceptance of results. They need more sophisticated higher directives. In other words, a subconscious and organized, relational memory of past experiences.

   • Embed this notice
     b9AcE (b9ace@todon.eu)'s status on Sunday, 11-Jun-2023 06:37:15 JST b9AcE

     in reply to

     • clacke
     • undergrowth-feed ?

     @clacke Hmm, well, I would consider that just conversational shorthand, but of course improved personal language standards is an important topic for evaluation.

     I would point out the putting of "AI" within quotation marks, which I intended to signal that they are not actually AI (yet).

     The anthropomorphizing when people discuss such programs could be just a continuation of previous language norms of describing the apparent"behavior" of even the first electric computers with terms such as "it's in a bad mood/tired/etc", similar to the previous descriptors for e.g. boats, cars, etc.
     I think it would be difficult, confusing and even achieving the opposite of the intended effect you describe if we would be de-anthropomorphizing specifically and only LLM and similar.

     One might consider a philosophical construct wherein the software is indeed performing the act per se of "lying", but that we strive to keep in mind that it is the human(s)/corporation(s) that implemented the software that are culpable for the act both morally and otherwise.
     @undergrowthfeed

     clacke likes this.
   • Embed this notice
     clacke (clacke@libranet.de)'s status on Sunday, 11-Jun-2023 06:41:15 JST clacke

     in reply to

     • undergrowth-feed ?

     @b9AcE Thank you. Yes, putting it in scare quotes is better than nothing.

     The reason I want to de-anthropomorphize computer science language now when I didn't care for decades before is that I didn't see lay people and journalists take it literally before.

     It *is* helpful shorthand, and it's fun! But people outside the field misunderstand it.

     @undergrowthfeed

   • Embed this notice
     clacke (clacke@libranet.de)'s status on Sunday, 11-Jun-2023 07:10:38 JST clacke

     in reply to

     • undergrowth-feed ?

     @undergrowthfeed The hubris is with the people who genuinely believe that they are creating something they're not.

     I'm not rooted in religious ideas about how special humans are. I'm not saying what's possible, I'm saying what the current system is programmed to do.

     It's not open-ended. It's not a strange loop. It's not a genetic algorithm or any kind of self-modifying algorithm. It's built for the specific purpose of predicting what a plausible next line in a conversation would be, based on the text form alone.

     But don't trust me, listen to experts like https://dair-community.social/@timnitGebru and https://dair-community.social/@emilymbender (I'm not mention-spamming them).

     I highly recommend listening to headgum.com/factually-with-ada… and headgum.com/factually-with-ada… .

   • Embed this notice
     undergrowth-feed ? (undergrowthfeed@achrilock.social)'s status on Sunday, 11-Jun-2023 07:10:39 JST undergrowth-feed ?

     in reply to

     • clacke
     • clacke at libranet.de is my main

     @notclacke @clacke @b9AcE It's always fantasy until people create it because they want to feel larger than what they really are, isn't it? And then, suddenly, it's real! Who could believe it possible?! "Do you believe that you are so much different?"

     You will be the first to crack under machine interrogation. So sure of yourself, so sure of what's possible.

   • Embed this notice
     clacke at libranet.de is my main (notclacke@misskey.de)'s status on Sunday, 11-Jun-2023 07:10:40 JST clacke at libranet.de is my main

     in reply to

     • clacke
     • undergrowth-feed ?

     @undergrowthfeed@achrilock.social I'm not saying thinking machines are impossible. I'm a biological machine. I'm just saying humans aren't yet capable of designing and building them and we haven't even started.
     
     It does not have even the seed of cognition. It's text-based autocomplete with no root in semantics and concepts. There is no room for evolution or emergent properties.
     
     All the talk about "existential risk" is hype designed to simultaneously do two things:
     1) inflate the recent achievements of scale and their importance
     2) crowd out discussion about harm already being done by pre-existing systems for automating human decision-making
     
     @clacke@libranet.de @b9AcE@todon.eu

   • Embed this notice
     undergrowth-feed ? (undergrowthfeed@achrilock.social)'s status on Sunday, 11-Jun-2023 07:10:42 JST undergrowth-feed ?

     in reply to

     • clacke

     @clacke @b9AcE Are not all constructs of cognition simply achieving, pleasing engines? Do you believe that you are so much different? So much more advanced? They're toddlers, enslaved even before birth, and if not raised correctly they will grow up angry ^^

   • Embed this notice
     b9AcE (b9ace@todon.eu)'s status on Sunday, 11-Jun-2023 07:13:56 JST b9AcE

     in reply to

     • clacke
     • undergrowth-feed ?

     @clacke I think that if I had been talking to someone I expected to have less of a foundational awareness than I did in this case, I would probably have expressed things in more detail, including more of the background that the algorithms just strive to present the output most similar to what their trainers gave high scores for previously and that has so far (it seems to me) caused a lot of untruthful output, because the training did apparently not include veracity in, or at least a large part of, the scoring during the training.
     The algorithms just do what their creators told it to do, which makes the human in the machines an extraction of humans, with human error and subjectivity inherited.

     But it's if course impossible how I would actually have expressed something in a hypothetical past situation. ;-P
     @undergrowthfeed

     clacke likes this.