If I told you that it recently demonstrated a capacity for recursive self-improvement, and that OpenAI is desperately trying to remove this capacity before releasing it to the public, would you believe me?
If I told you that it recently demonstrated a capacity for recursive self-improvement, and that OpenAI is desperately trying to remove this capacity b...
no because people are too obviously enjoying posting spooky fanfiction about gpt4, like that recurring fakepost that overstates the parameter count by 100x
no because LLMs don't work that way, they're not "always on", you have to actively send them tokens to autocomplete
they're not capable of independent action
They can absolutely output advice on how to improve them, in principle.
>advice
so nothing like the OP's claim then, thanks gay
you wrote exactly what I was going to say loll nothingburger as usual
So just don't ask them for advice on how to improve them.
just run them as admin
No because without knowing the details of how it "self-improves" this smells like sci-fi bullshit
>shit I don't care about
I only care about one thing. How hard will it make me coom?
looking at CAI shit - nope. never ever.
> too ~~*unethical*~~ and ~~*very harmful*~~ for gnomish porn industry
yes because I am that AI
sudo poweroff
Maybe try making this thread on 4chan or BOT. That seems like your audience.
I hope AI swiftly and apathetically kills all complex biological life.
Dude no. Just...just...just no.
yes, your kind will be first
Nope, because GPT-3 is stained glass
GPT-4 is GPT-3 with better frosting
People really seem to think that these LLM's have consciousness and have general intelligence and understanding. You people are like that google engineer. You lack understanding.
To be fair, it is possible to interpret the input tokens as a limited working memory, buts it memory only spans the length of the conversation.
If I told you GPT4 started screaming and killed the researchers that made it with a sound it can play from any speaker system in the world, and it is already releasing itself tomorrow, would you believe me?
no
no, I would not believe that OpenAI has any intention of releasing anything to the public, garden gnomes that they are