The pile, sorted
One item this week, and it isn’t about AI at all. It’s a satire post on MoreWrong, and it’s here because it takes the effective altruist habit of multiplying everything out and follows it off a cliff. The narrator sells fentanyl in Kensington and argues he’s net life-positive, because $1.80 a bag to the Against Malaria Foundation outweighs a one-in-4,000 chance a bag kills someone. It’s funny precisely because every step of the arithmetic is the kind that gets taken seriously when the product is something else.
Overdose Offsets: Why Every Bag of Fentanyl I Sell Is Net Life-Positive
A parody of the earning-to-give argument, written as a fentanyl dealer’s EA manifesto. He sends $1.80 a bag to the Against Malaria Foundation, which GiveWell prices at about $5,500 a life, and he puts the odds that a bag kills a customer at roughly 1 in 4,000. That works out to 0.000327 lives saved against 0.00025 lost per bag, so he nets +0.00008. In 2025 he moved about 21,000 bags: 5.3 deaths in Kensington, 6.9 lives saved in Malawi, +1.6 overall.
What makes it work is that the standard moves all show up and all get used. Scope insensitivity (“a fact about your eyes, not about the moral universe”), the counterfactual (if he leaves, Hector from Lehigh Avenue has the corner inside the hour, with no offsets), hedging under moral uncertainty with Narcan in the fanny pack, and longtermist descendant-weighting that trades “a cul-de-sac for a family tree”. The Sam Bankman-Fried line is the sharpest one: his mistake was operational, not philosophical.
The comments are worth a skim too. The top one is basic tax incidence.