False memories, false friends, and false justice — Sept 28 Weekly Roundup
What I read, watched, and listened to this week.
Reading
Hacker plants false memories in ChatGPT to steal user data in perpetuity (Ars Technica)
Prompt injection may be an unsolvable problem, and this illustrates the danger that poses to the future of LLMs.
AI tool cuts unexpected deaths in hospital by 26%, Canadian study finds (CBC)
When people talk about the promise of AI, this is the sort of thing they (hopefully) mean. AGI is a fantasy, but there’s real human upside to this technology (and innumerable downsides, of course).
Listening
Blue Velvet with Jamie Loftus (Blank Check Podcast)
2 hours and 40 minutes of podcast talking about David Lynch’s 1986 masterpiece Blue Velvet? Don’t mind if I do!