My book, GAMES: AGENCY IS ART is out! It's about: . How game designers sculpt agency. How games let us record, transmit, and explore new forms of agency. How real games make us more free. How gamification undermines our freedom.

The Nightmare Before Christmas as a meditation on cultural appropriation.

RT @SigalSamuel: Congratulations to @add_hawk, one of the coolest philosophers around today, who's on the @voxdotcom.Future Perfect 50 list….

RT @joynessthebrave: Brat summer is out, attic wife autumn is in. We’re hissing at people. We’re withdrawing from society. We’re growing ou….

RT @Bethlehemtekola: Interesting paper on the mental health of people doing qualitative research:. Historically, those doing qualitative wo….

RT @jonlpan: from "Food assembly and the idea of a dish" by @add_hawk.

RT @NoahGarfinkel: A key point missing in a lot of post debate analysis is that Trump’s claim about immigrants eating pets almost perfectly….

RT @kareem_carr: Infographics of this dataset have been kicking around on the internet for years. It is an insult to real scientists everyw….

RT @jason_koebler: Beloved Apple blog TUAW was shut down in 2015, sold to private equity, then sold to a company in Hong Kong. It recently….

@KathleenACreel

PS @KathleenACreel is a trained philosopher and software engineer who’s appointed as a prof in both phil and CS, and she has proposals for technical solutions to partially ameliorate these problems which are above my pay grade.

Anyway, somehow the naming artifacts problem lets me understand, in my gut, what I've already heard many times but never fully processed: exactly the degree to which LLMs are not thinking, but are mass averaging parrot devices.

And, she says, this is a tough problem to get out of, because it's hard to train different models on different data sets - because there's only one Internet.

And she says there's a bunch of distinctive new LLM biases, like that naming artifact. Which is because LLMs are trained on the internet, and the internet is, like, 60% by weight Reddit. So resume algorithms are all systematically biased against anybody named Donald or Hillary.

@KathleenACreel says: there are a bunch of different corporate products, but they're usually adaptations of one original model. Or even if they aren't, they're different models trained on the same base data set.

This creates a distinctive new social harm: the possibility that you'll get shut out of *every* job search because your resume doesn't fit that one model.

I learned this from @KathleenACreel and her AI ethics work on "the algorithmic monolith". Like when the majority of HR departments use a handful of resume-filtering algorithms for picking who to interview, which are all drawn from one original model.

The example that helps me intuitively understand how LLMs "work, more than anything else: "naming artifacts":. It turns out, LLMs are systematically biased against anybody with the following names: Donald, Barack, Hillary, Bernie.

RT @jordanbhx: If you’ve ever wondered why we can’t build places that look like Italian hilltop towns, one of the reasons is Edwardian prud….

RT @car0linehaskins: New/exclusive: .African workers who moderate platforms like Instagram + train AI for places like OpenAI (often for >$2….