Conversation

Notices

1. Embed this notice
   Bread up, Bro (sickburnbro@poa.st)'s status on Thursday, 23-May-2024 23:22:51 JST Bread up, Bro
   this is why I'm not worried about these systems. There is something critical missing in them which allows this to be presented.
   
   It shows a critical missing piece with how they link things.
   • Embed this notice
     ApexBoomer (apexboomer@poa.st)'s status on Friday, 24-May-2024 01:45:56 JST ApexBoomer

     in reply to
     @sickburnbro These AIs are often wrong. When I engage ChatGPT on a topic I am familiar with, the errors it makes are striking. Simple stuff that even a novice couldn't get wrong, the AI mangles. And when you call it on its errors, it pretends it meant to say the correct answer, and when you say, "No, I was lying, your original answer is correct," it will go back the original. Some more complex concepts it absolutely nails. It's just not trustworthy. The scary bit is that they *could* use this technology to control robots -- it seems straightforward. But it would not do the right things in many cases.
     Bread up, Bro likes this.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 01:47:21 JST Bread up, Bro

     in reply to

     • ApexBoomer
     @ApexBoomer right, and that's the part that is important. A lot of those things it is wrong on should be a fundamental problem for a reasoning system.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 01:48:30 JST Bread up, Bro

     in reply to

     • Inertia
     @inertia it's not hilarious because they're gonna put it everywhere for "cost savings" and it's gonna kill you.
   • Embed this notice
     Inertia (inertia@pone.social)'s status on Friday, 24-May-2024 01:48:31 JST Inertia

     in reply to

     @sickburnbro this is without a doubt hilarious.

     AGI HAS BEEN ACHIEVED INTERNALLY! ONLY 7 MORE MONTHS!
     - AI shills

   • Embed this notice
     Semilexic (epictittus@poa.st)'s status on Friday, 24-May-2024 04:23:58 JST Semilexic

     in reply to
     @sickburnbro That critical part is context, and you need a body existing in the world and experiencing reality to understand context. The best AI could donis rely on people for context, like if someone curates the information fed into it (eg scientific articles), but unfiltered information is useless.
     
     Unfiltered input and output just makes an incredibly funny shitposter.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 04:23:58 JST Bread up, Bro

     in reply to

     • Semilexic
     @epictittus I'm not even going that far. What I'm saying is that elmers glue isn't in the category of "food". one of the basic rules of food is there is stuff that is food and stuff that isn't. putting something not-food into food should be a no-no.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 04:24:50 JST Bread up, Bro

     in reply to

     • Semilexic
     @epictittus if you have a magic system that can learn deep contexts, either it is not able to pick this up .. or it is failing to apply it here. In either case, not good.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 05:43:45 JST Bread up, Bro

     in reply to

     • Jonny
     • ApexBoomer
     @Jonny @ApexBoomer it's pretty close to there now and it's freaking them the fuck out.
     
     All you need is something that functions at 2010 google search quality and answers truthfully or with your desired filter.
   • Embed this notice
     Jonny (jonny@annihilation.social)'s status on Friday, 24-May-2024 05:43:46 JST Jonny

     in reply to

     • ApexBoomer
     @ApexBoomer @sickburnbro
     For the time being, AI is being programmed by jews. Eventually AI will be decentralized like Fedi and it will be amazing, but for now, talking too chat gpt is like talking too a New York tunnel jew.
   • Embed this notice
     teknomunk (teknomunk@apogee.polaris-1.work)'s status on Friday, 24-May-2024 05:44:12 JST teknomunk

     in reply to

     • teknomunk
     • Semilexic
     @sickburnbro @epictittus
     
     I think to get to a general intelligence is going to require multiple neural network models to to handle the various tasks. The ChatGPT track of research is basically the core of a linguistic model, and the speech synthesis models give you a voice.
     
     There is currently nothing like a prefrontal cortex that I've seen so far.
     Bread up, Bro likes this.
   • Embed this notice
     teknomunk (teknomunk@apogee.polaris-1.work)'s status on Friday, 24-May-2024 05:44:13 JST teknomunk

     in reply to

     • Semilexic
     @sickburnbro @epictittus
     
     ChatGPT, if I understand roughly how it works, is an extension of Markov chain generators, with a neural network instead of a simple table of probabilities of word pairs.
     
     Using a Markov chain generator trained on your own posts is a fun thing to do that makes a bot that sounds roughly like you , but is often hilariously wrong.
     
     https://en.wikipedia.org/wiki/Markov_chain
     Bread up, Bro likes this.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 05:45:01 JST Bread up, Bro

     in reply to

     • teknomunk
     • Semilexic
     @teknomunk @epictittus indeed, I remember the problem at one point was 'we can make systems to parse language pretty well, but noun knowledge is not doable' .. well now we have a noun knowledge system, pretty cool.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 05:46:33 JST Bread up, Bro

     in reply to

     • teknomunk
     • Semilexic
     @teknomunk @epictittus right, that's when they say "next token"
   • Embed this notice
     Jonny (jonny@annihilation.social)'s status on Friday, 24-May-2024 05:46:42 JST Jonny

     in reply to

     • ApexBoomer
     @sickburnbro @ApexBoomer
     Just imagine a decentralized AI ran by our guys and maybe even secured with block chain.
     Bread up, Bro likes this.
   • Embed this notice
     Bread up, Bro (sickburnbro@poa.st)'s status on Friday, 24-May-2024 21:06:17 JST Bread up, Bro

     in reply to

     • EvolLove
     • Semilexic
     @EvolLove @epictittus all "reality" is, is having a bunch of assumptions about how things work. If you were placed on mars tomorrow, a lot of your assumptions would have to change.
   • Embed this notice
     Semilexic (epictittus@poa.st)'s status on Friday, 24-May-2024 21:06:18 JST Semilexic

     in reply to

     • EvolLove
     @EvolLove @sickburnbro Thats not how it works. If, for example, it thinks genociding white men is ok, you just have to say
     >what if the white men self identify as jews/blacks/women
     To get the genocide exemption. Leftism is deeply flawed, it just doesnt make sense in any way, and this leaves gaps wide open for exploitation.
     
     The way flesh and blood leftists get around it is to run away, so maybe thats what the AI would do. But that means it wouldnt deliver the message.
   • Embed this notice
     EvolLove (evollove@noauthority.social)'s status on Friday, 24-May-2024 21:06:18 JST EvolLove

     in reply to

     • Semilexic

     @epictittus @sickburnbro

     nah I really don't think AI would have any problem with a distorted reality since it has no sense of reality. it will just keep on making dumber and dumber conclusions for an eternity.

     way I see it AI should be perfect for leftist ideology. since they are just as clueless as the AI. It's like a perfect match.

     conservatives know better than to trust a machine with any decision.

   • Embed this notice
     Semilexic (epictittus@poa.st)'s status on Friday, 24-May-2024 21:06:19 JST Semilexic

     in reply to

     • EvolLove
     @EvolLove @sickburnbro Not really. If the political ideology is incorrect and contradictory, it will cause an AI that breaks down and acts retarded, or it has to use way more memory to keep track of all the retarded contradictory rules of leftism.
     
     Right wing AI will be sleek and super fast by comparison.
   • Embed this notice
     EvolLove (evollove@noauthority.social)'s status on Friday, 24-May-2024 21:06:19 JST EvolLove

     in reply to

     • Semilexic

     @epictittus @sickburnbro

     I don't think so. because the AI lacks the ability to detect contradiction. it will just keep adapting its data bank based on ideology approved data without worrying about any contradictions.

   • Embed this notice
     EvolLove (evollove@noauthority.social)'s status on Friday, 24-May-2024 21:06:20 JST EvolLove

     in reply to

     • Semilexic

     @epictittus @sickburnbro

     and that's why this will be a disaster.

     Because the next generation has to be fed approved data only. that means that humans will decide what is true and not.

     that means that it will be fed data based on political ideologies.

     that gives us something very dangerous. it's gives us an AI based on pseudo science.

     possibly very dangerous.