Source: Leniolabs
That damn 37 again
reply
reply
37 signals
reply
Makes sense given it’s trained on human data
reply
On the number 37 by Veritasium
reply
surprisingly this seems to be an area where it actually is more human than one might expect. humans are notoriously bad at generating random numbers and often pick favourites that "feel random".
as an example, if asked to pick a random number from 1 to 10, most people will (unconsciously) reason the following way about each possibility:
  1. can't be random because it's at the beginning of the range.
  2. can't be random because it's the smallest even number.
  3. might be random because it's prime, but it's the smallest odd prime so don't pick it.
  4. not random, it's even.
  5. not random, it's the middle of the range.
  6. not random, it's even.
  7. fairly random, it's prime, and not near the beginning, middle or end.
  8. not random, it's even.
  9. not random, it's three 3s and near the end of the range.
  10. not random, it's the end of the range.
the result is that most people pick 7, even though none of the "reasons" are anything to do with randomness, and any of the numbers could be just as "random" as any other.
I wonder what ChatGPT's reasons are for favouring the numbers it has chosen? 42 could be because of the Hitchhiker's Guide to the Galaxy reference, but what about the others?
reply
Lack of 69 smh
there is clearly an alignment problem
reply
69 is clearly not random enough
reply
Interesting that it forms a bell curve
reply
Shocking
reply
Wouldn't be surprised if AI advises us to start the counting from 42 and not from 0.
reply
Quite interesting.
reply
42: for Hitchhiker's guide obviously.
reply
42 of course
reply