Conversation

Notices

1. Embed this notice
   iced depresso (icedquinn@blob.cat)'s status on Thursday, 07-Nov-2024 17:39:52 JST iced depresso

   • THOT POLICE
   @s8n when the AGI cometh
   all will become cute girls
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 07-Nov-2024 17:45:59 JST iced depresso

     in reply to

     • THOT POLICE
     @s8n that's because the dingbats are looking in the wrong fucking places.
   • Embed this notice
     THOT POLICE (s8n@posting.lolicon.rocks)'s status on Thursday, 07-Nov-2024 17:46:00 JST THOT POLICE

     in reply to
     @icedquinn I was watching an interview Michio Kaku was giving for some reason even though I'm not a fan of his oversimplifications that err on the side of wrong too often
     
     but he was talking about how our most powerful, most complex, most super secret military AI is, right now, about as intelligent as a retarded cockroach. Because if you put it in the forest it will just get fucking lost and die. But an actual cockroach would mate and reproduce and all sorts of crazy shit. We're not even at a measurable fraction of a percentage of the intelligence of a cockroach and everyone thinks AGI is happening in the next hundred years
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 07-Nov-2024 17:51:20 JST iced depresso

     in reply to

     • THOT POLICE
     @s8n numenta was actually on to some stuff before they got gates'd. they had got a cortical based attention model working (but they didn't realize this; they just called it a sensorimotor model) right before they gave up and sold out to bill gates and are now doing some other stupid shit instead of the biological models. what numenta gleamed the surface of is that enforcing sparsity is necessary.
     
     schmidgel showed spike networks exhibit reliable stability on reinforcement training, which the google labs never achieved.
     
     it turns out the spike coding is absolutely mandatory for intelligence to work. googlers et all are still using some cope that "well the weights are kind of the same thing as pulse train coding averaged over ..." :blobcatstop: NO. BAD. do it like nature did it.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 07-Nov-2024 17:55:43 JST iced depresso

     in reply to

     • THOT POLICE
     @s8n the paper for e-prop showed you can unroll the derivatives as in backpropagation through time, but then you can actually approximate over the derivatives through time, and replace them with basically a single protein marker. which enables backprop to operate online (and the same protein traces appear to exist in real brains.) it doesn't work in super deep contexts, which is fine because the brain doesn't actually have those.
     
     its looking like the way intelligent systems work is actually nowhere near most ANN research; it seems to be mostly dumbly self organizing maps and a very thin layer that actually adapts to inputs, but
     
     i stopped editing my reference manual when factorio space age came out. its still on the drive i just haven't been cramming the papers all night.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Thursday, 07-Nov-2024 18:01:05 JST iced depresso

     in reply to

     • THOT POLICE
     @s8n i've never really cared or asked what kovarex's opinions were. i remember a youtuber protest deleted all his content because kovarex said the social justice nerds could eat a dick or something lmao
     
     i've been putting off doing some of the tests. i have the paper stack mostly read but some of this shit has to be custom rigged because its literally like published six months ago level shit
   • Embed this notice
     THOT POLICE (s8n@posting.lolicon.rocks)'s status on Thursday, 07-Nov-2024 18:01:06 JST THOT POLICE

     in reply to
     @icedquinn I didn't buy it because the developers were huge shiteating faggots about ukraine