Amy Hoy

the anthropic dweebs have "discovered" that claude* uses a sort of swap file/memory. and that if you tell it "white elephant" this swap "space" includes the tokens "white elephant." & that means it's basically conscious *the entire pkg, bc without the harness, the llm is useless

A global workspace in language models \ Anthropic anthropic.com

src:

We Asked the ‘Future of Truth’ Author to Explain How He Used AI. It Didn’t Go Well wired.com

kane has figured out why the america.gov chatbot behaves so weird when told to “play minecraft”

America.gov america.gov

indeed the numbers are bad but THIS was my favorite bit about the anthropic s1 www.reuters.com/business/fin...

reuters.com

bunch of total fuckwits in my replies claiming that our eyeballs and hands are no different than a react app i’ve blocked them but genuinely this is an absolutely braindead argument and it’s a sign they’ve been spending too much time talking to chatbots and not enough to humans with brains

me: so it’s obvious that llms aren’t conscious bc they have no memory, no persistence, each turn is totally fresh, it’s fakery random dude: yeah well your perception is also an illusion bc of saccades and stuff …THAT IS STILL THE HUMAN BRAIN. my eyes aren’t an app written in react.

if artificial intelligence can feel pain i’m gonna need to put some in my thermostat so i can shock it when it decides to ignore my settings and just tells me it’s “waiting” WAITING FOR WHAT

“they’re torturing the llms, which tract as if they’re being tortured. it’s psycho behavior” clearly you have not spent enough time around players of the global smash hit game, the sims

ftr i did not repost the anthropic lost $80b thing bc i suspected there was an asterisk in there somewhere, and there is i do not fully understand the asterisk but i do know it’s not “they spent this money.” beware vague skeets that confirm your priors etc etc

waooow llms are so useful for summarizing