Conversation

Perhaps a 150-dimensional vector space for representing others, with eigenbasis defined by stereotypes

Quote Tweet

Grognor

@Grognor

Nov 14, 2016

I do not know what the human Dunbar number is, but it is much greater than 150

Replying to

How many dimensions does our world representation have?

Replying to

word2vec has roughly 200 iirc?

Replying to

yes but

is asking about world2vec

Replying to

and

actually, while you will see me rant on the limitations of language, I think there's not that much of a difference

Replying to

and

some thoughts from the alt, epistemic status explorative

Replying to

and

wearing the boring hat for a second, dimensionality of word2vec comes from optimizing for particular encoding architecture

Replying to

and

I'd actually be interested in the details, the word2vec idea just seemed like a really convenient but sloppy shortcut

7:55 PM · Nov 14, 2016Twitter Web Client

Replying to

it's not sloppy, it's actually quite clean- just seems unlikely to be analogous to the way humans actually encode words

Replying to

and

word2vec just tries to predict a word from immediately surrounding words, this rep makes them small enough to put in an LSTM

Show replies

Replying to

Not word2vec, but world2vec @The_Lagrangian

Replying to

now all of us have come up with that one at least once :)

Quote Tweet

'(·)

@allgebrah

Jul 31, 2016

world words, holographic carriers of meaning from which you can reconstruct the entire book, language, society, universe