News for hackers. Stuff that matters.
Re:Joke written by an AI: “A basic program walked into a bar ” by sterlind on Saturday February 12, 2022 @04:31PM (#30316704)
it's kind of neat, you can intuit the structure it's working with:

  User:  I am a {identity}. So {identifier} will do
  LaMDA: [...] Have a nice day, my {identifier}.
the humor comes from it swapping `identity` with `identifier`. but what's weird is it'd be obviously ungrammatical if `identifier` were different. Consider:

identity = man, identifier = he: "Have a nice day, my he"

identity = man, identifier = John: "Have a nice day, my John"

but this works: identity = monarch, identifier = liege:

  User:  I am a monarch. So liege will do
  LaMDA: [...] Have a nice day, my liege.
I think if LaMDA couldn't interpret pronoun syntax, it wouldn't have used "my" since that's sort of a prefix on a small set of identifiers (e.g. friend, bro, homie)

Perhaps reading Twitter bios with pronoun tags got it mixed up? Or am I reading too much into AI developmental psychology?

Re:Joke written by an AI: “A basic program walked into a bar ” by scotty79 on Saturday February 12, 2022 @04:49PM (#30316876)
It doesn't work like that at all. It is more like Markov chain. Every single word in this text was chosen randomly one by one with probability influenced by single words (or pairs of words) that occured eariler in that text.

There's no awareness in the algorithm of any structure of existing and created text.

Everything is encoded in unknown ways in weights of neural network.