3 ms·

> When fixing broken layout, it is always easier to add stuff that you forgot than removing stuff that your framework did for you and you that can’t see. This

> When fixing broken layout, it is always easier to add stuff that you forgot than removing stuff that your framework did for you and you that can’t see. This paragraph reminded me of this old Jargon File koan: --- In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6. “What are you doing?”, asked Minsky. “I am training a randomly wired neural net to play Tic-Tac-Toe” Sussman replied. “Why is the net wired randomly?”, asked Minsky. “I do not want it to have any preconceptions of how to play”, Sussman said. Minsky then shut his eyes. “Why do you close your eyes?”, Sussman asked his teacher. “So that the room will be empty.” At that moment, Sussman was enlightened.

Izkata 7y ago
I've never understood this one. Initialized randomly and then trained is accurate; Minsky is the one who comes off as ignorant, when the last line is supposed to indicate the opposite.
- TuringTest 7y ago
  That's not ignorant on his part, it's just a silly analogy to get the point across; and I think you're missing the point by thinking of what happens after the network is trained; the point made is about the initial state of the network, not the outcome. The neural network will have preconceptions on how to play no matter how, even if you wire them randomly; the difference of making them random is that you'll have no idea what those preconceptions are, not that they don't exist. The original story on which the koan is based might make it clearer. [1] https://simple.wikipedia.org/wiki/Hacker_koan#Uncarved_block https://simple.wikipedia.org/wiki/Hacker_koan#Uncarved_block If you still want this to be about the final state, the point would be that, if the training process works well, it shouldn't make any difference whether the net was wired randomly or orderly.

Izkata 7y ago

I've never understood this one. Initialized randomly and then trained is accurate; Minsky is the one who comes off as ignorant, when the last line is supposed to indicate the opposite.

TuringTest 7y ago
That's not ignorant on his part, it's just a silly analogy to get the point across; and I think you're missing the point by thinking of what happens after the network is trained; the point made is about the initial state of the network, not the outcome. The neural network will have preconceptions on how to play no matter how, even if you wire them randomly; the difference of making them random is that you'll have no idea what those preconceptions are, not that they don't exist. The original story on which the koan is based might make it clearer. [1] https://simple.wikipedia.org/wiki/Hacker_koan#Uncarved_block https://simple.wikipedia.org/wiki/Hacker_koan#Uncarved_block If you still want this to be about the final state, the point would be that, if the training process works well, it shouldn't make any difference whether the net was wired randomly or orderly.