He compared scaling current systems to building taller towers.
"Come see our artificial bird!" "Impressive, but that's a tower." [Later]"What about this bird?" "A fine tower." [Later]"This one reaches the stratosphere, higher than any bird." "Still a tower, not a bird." "Bah! Stop moving the goalposts! How high must it reach convince you?"
Positive users believe LLMs already demonstrate thinking and will reach AGI soon, while negative users argue they fail to show true understanding according to Deutsch's tower-bird analogy.
Airplanes do not need flapping wings in order to carry people and goods through the sky, but they *would* need flapping wings in order to convince someone that humanity actually understands how birds work in enough detail to build an accurate fully functional bird from scratch.
"Come see our artificial bird!" "Impressive, but that's a tower." [Later]"What about this bird?" "A fine tower." [Later]"This one reaches the stratosphere, higher than any bird." "Still a tower, not a bird." "Bah! Stop moving the goalposts! How high must it reach convince you?"
In the case of frontier AI, humanity knows less about what we’ve built than we know about airplanes *or* birds. LLMs almost completely fail to satisfy the primary *original* motivation for AI research, which was to advance the science of human cognition.
But that won’t stop near-future LLMs from automating your job.
@davidad Heard that great writer Ted Chiang has a story about this. I mean the tower. The one that reaches the stratosphere.
@reiofw @davidad Is this a Pagliacci joke
@quasicoh @davidad Man trains model on a book. Model says it's conscious. He goes to doctor and the doctor says, don't worry, great writer Ted Chiang is in town. He will convince you it isn't. Man says, "But doctor..."
@DavidDeutschOxf More like an airplane vs bird.
If the argument is it’s not a mind - we should specify what’s a mind, else its non-falsifiable by Popperian standards - I can think of a simple definition - mind is something which can make novel deductions or theory not latent in its training.
@davidad i would say they are far from completely failing, it's just a matter of time
@DavidDeutschOxf I don't think people are changing the goalposts. We just think LLMs really are thinking. I believe they can lead to AGI, they contain real intelligence and creativity. You disagree, I don't understand why but that's a separate thing. I don't see anyone changing the goalposts.
@DavidDeutschOxf Still no AGI
@DavidDeutschOxf @ToKTeacher Lol
@gnostic_snakes Fair, I should have said “so far almost completely”.
Meanwhile, someone else watching: "OMG. Bird has been achieved!"
"Come see our artificial bird!" "Impressive, but that's a tower." [Later]"What about this bird?" "A fine tower." [Later]"This one reaches the stratosphere, higher than any bird." "Still a tower, not a bird." "Bah! Stop moving the goalposts! How high must it reach convince you?"
@reiofw @davidad Man goes to doctor, says he's depressed. He's grasping for an argument that LLMs aren't conscious. He's afraid he'll lose his humanity if they are. Doctor says, don't worry, great writer Ted Chiang is in town. He will tell you how to cope with alien intelligence. Man says, ...
He compared scaling current systems to building taller towers.
"Come see our artificial bird!" "Impressive, but that's a tower." [Later]"What about this bird?" "A fine tower." [Later]"This one reaches the stratosphere, higher than any bird." "Still a tower, not a bird." "Bah! Stop moving the goalposts! How high must it reach convince you?"