wow this was hard but fun. apparently i don't understand the structure of word2vec at all
semantle.novalis.org
Conversation
got it in 32! I had a very sudden jump on my penultimate guess, rot13:
zl bayl aba 'pbyq' thrff unq orra "genafvfgbe" juvpu vf whhhfg va gbc 1000, gura nsgre zber zrnaqrevat V gevrq "vairagvba" juvpu vf whfg fvkgrra sebz gur gbc, naq tbg "cngrag" ba gur arkg thrff :Q
2
1
ayyy nice. in retrospect i didn't "jump around wordspace" anywhere near enough in the beginning and got way too anchored on my first few guesses
1
1
yeah I was noticing this too; the way you have to jump to completely different regions of concept space felt antithetical to how, like, thinking works fundamentally😅
2
1
also I was so confused when nsgre "genafvfgbe" (fpber gjragl gjb) V gevrq "jver" naq vg tbg bar cbvag svir, yvgrenyyl gur shegurfg jbeq bs nal V raqrq hc gelvat!
1
1
Replying to
yeah there's some really unintuitive word2vec distances. to the extent that it's roughly about like "substitutability" you might argue that those two words don't really appear in similar places in similar documents, maybe???
Age-restricted adult content. This content might not be appropriate for people under 18 years old. To view this media, you’ll need to log in to Twitter. Learn more
Show replies