darkpatterns

Totally agree. 2/3 is wild, especially given it seems like you could mitigate most of the risk just by removing vowels from the dictionary.

Literally just cancelled mine today a couple of hours before reading this. It's great for music discovery but when already paying for YouTube premium which includes YouTube Music, which although has a much worse UX ultimately has the same content. Even if I wasn't getting music with the YouTube subscription when I added up the 50 odd tracks I regularly listen to on playlists and divided that by the cost of a yearly subscription, the economics make no sense.

One domain where this concept of handling ambiguous left and right in natural language is generative image AI models like StableDiffusion. What I wouldn't give for a clear syntax to define what I mean when I say "left side" or "to the right" in a way that captures a character or object which may be rotated in any number of positions. Running 100 generations with the same prompt and different seeds demonstrates how hard this is to solve using words.

My only point that as with anything (and especially with crime statistics), "it's just not that simple" and nuance matters, in the sense that (in the modified analogy), to say at least some systemic racism doesn't exist in the way the prison stats play out seems naive at best, it seems similarly naive to assume there's no bias in this case as well.

Don't read anything into this, but can you see the next step here if you replace "men" with "black men"? Is this actually the point you want to make?

Good community called Scraping Enthusiasts on this topic here: https://discord.gg/4fGEPZzs Plus curated list of research papers here if you want to go deep on the subject matter: https://github.com/prescience-data/dark-knowledge