Piatti del giorno 86

Been a weird couple of months in which ChatGPT has been utterly unavoidable as a topic. I’m skeptical, so I kind of hate it. But as a topic it’s also so all-encompassing in what it contests–socially, technologically, in design terms, how it keeps coming up in every setting including with my non-techy family–it is fascinating and never boring.

• What Is ChatGPT Doing … and Why Does It Work - Long but good from Stephen Wolfram, and makes effective use of small multiples visualizations (eg. on loss functions).
• Amazon Begs Employees Not to Leak Corporate Secrets to ChatGPT
• GLAZE: Protecting Artists from Style Mimicry by Text-to-Image Models: A method for poisoning image files so that they still look the same to humans but cause problems in training text-to-image models.
• ChatGPT, Galactica, and the Progress Trap
• Vector Symbolic Architectures in Clojure - Some ideas around “small AI” that I wish I understood.
• Design thinking was supposed to fix the world. Where did it go wrong? - TL;DR: technosolutionism dropped from the sky that often isn’t grounded in the boring but real reasons that things are as they are.
• Re-frame Database Best Practices + Code Examples
• Autocommands - Time spent learning vim is never wasted, in my experience.
• OpenFeature - Like OpenTelemetry, but for feature flagging, aka dark launching.
• Visual design rules you can safely follow every time
• “Read Jane Jacobs by Jane Jacobs” - An extremely niche joke in whose tiny blast radius I’m quite happy to be.
• Using the Error Model - Error handling is one of those things that seems like it ought to be easy, but unpacks into deep philosophy that could almost be used as some kind of workplace personality test.
• Unbundling Tools for Thought - About “tools for thought” as largely pointless busywork.
• On Chindogu - “Chindogu is a Japanese word meaning ‘weird tool.’”