23:32:01

been working on an interview bot. it extracts important phrases from your text and explores the surrounding vector space. then it asks you stupid questions

🗨️ 1 ♺ 0 🤍 3


23:32:17

how could i make this more useful? nobody knows

🗨️ 1 ♺ 0 🤍 3


23:38:38

really I'm building the baseline prototype for the actual AI to come. this is wisdom from Emmanuel Ammeisen i think. build the shell of the whole app before your build your model, using a simple heuristic in place of a complicated neural net or what have you

🗨️ 1 ♺ 0 🤍 2


23:45:10

something like that. so if the stupid questions bot is amusing enough to keep me typing into it, it will serve it's purpose. As long as it keeps looking for new words in the near reaches of semantic space, and remembers not to ask about a given word twice. the human does the work

🗨️ 1 ♺ 0 🤍 1


23:48:58

later i can test various Transformer based solutions or other vector transformations. right now it has a gradient based on the difference between the word "simple" and the word "uncomplicated". it tries to complicate your words. (did simplifying at first, but this was funnier)

🗨️ 1 ♺ 0 🤍 1