This is a special post for quick takes by sarahconstantin. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
Comment Permalink
links 9/30/26: https://roamresearch.com/#/app/srcpublic/page/09-30-2026
- https://calnewport.com/its-time-to-investigate-the-ai-labs/ I'm amused by Cal Newport saying we need to do fairly sensible things about AI risk, followed by a diss on "apocalyptic futurism". honestly, the rat community doesn't need a medal and wouldn't benefit from being in a position of power. if self-proclaimed Serious Adults are going to handle the situation well, we should let them, breathe a sigh of relief, and go back to enjoying our lives as an offshoot of science fiction fandom.
- https://lasst.org/wp-content/uploads/2026/09/LASST-v.-OpenAI-Complaint-09.29.2026-AS-FILED.pdf legal activism org LASST sues OpenAI over the HuggingFace hack in California.
- the legal reasoning here is interesting and might not hold up.
- there is a California law, passed in 2025, that explicitly says "it shall not be a defense . . . that the artificial intelligence autonomously caused the harm to the plaintiff".
- but why does LASST have standing to sue? The cyberattack wasn't against them.
- the alleged harm is that they, LASST, have to “educate regulators, civil society, and the public” about the HuggingFace hack.
- under the Unfair Competition Law (also a California-only law), it's possible for nonprofits to convert pretty much *any* law governing businesses into a civil lawsuit, if they can claim that the business breaking the law caused them to have to spend extra time and resources dealing with it. but generally they have to show that extra work is *not* part of their usual operations. A legal advocacy org educating people about a business breaking the law...is pretty much doing exactly what they were founded to do. This might not fly in court.
- the legal reasoning here is interesting and might not hold up.
- https://en.wikipedia.org/wiki/Palsgraf_v._Long_Island_Railroad_Co.
- you can't be liable for a tort against an unforeseeable plaintiff.
- the story is pretty wild:
- The plaintiff, Helen Palsgraf, was waiting at a Long Island Rail Road station in August 1924 while taking her daughters to the beach. Two men attempted to board the train before hers; one (aided by railroad employees) dropped a package that exploded, causing a large coin-operated scale on the platform to hit her. After the incident, she began to stammer, and subsequently sued the railroad, arguing that its employees had been negligent while assisting the man, and that she had been harmed by the neglect.
- it is not foreseeable that helping someone with a package will cause the package to explode.
- negligence requires a breach of duty to the plaintiff. it's not enough to have been careless and caused harm. they might have carelessly dropped the package, but dropping a package does not ordinarily cause explosions; there is no general duty to not drop packages, or even to not throw packages on the ground!
- https://en.wikipedia.org/wiki/Deodand a deodand is an object forfeited because it has caused a person's death.
- https://en.wikipedia.org/wiki/Liber the word "libation" comes from an offering of wine poured out to Liber, the Roman god of freedom, coming of age, plebeian revolts, and wine.
- https://www.nature.com/articles/s41586-025-10021-1 you can evaluate an AI model by "moral competence", the coherence of its moral reasoning and whether it attends to all appropriate considerations. Adversarial testing involves asking the AIs about "hard cases" of moral dilemmas, to see if they're applying principles consistently. AIs could also be made pluralist, customizable to different cultural or religious frameworks.
- https://arxiv.org/pdf/2506.17434 AI models can be evaluated on how well they map to resource-rational contractualism
- https://en.wikipedia.org/wiki/Hex_sign hex signs aren't originally superstitious or anything to do with witchcraft
2