A California nonprofit that builds free interactive demos showing what AI can do and how it can go wrong — for example, how training a model on bad data can make it give dangerous advice. It also gives briefings to government and civic groups.
Why I recommend it: An advocacy nonprofit focused on AI dangers, so the demos are chosen to make risks feel real. Great for a quick, hands-on sense of why AI safety matters.
A Wharton working paper by Steven D Shaw and Gideon Nave, written 11 January 2026 and posted to SSRN on 2 February 2026. It proposes "Tri-System Theory" — adding a "System 3" (thinking done outside your head by a machine) to Kahneman's fast/slow account — and names "cognitive surrender": taking an AI's answer with barely a glance. Across three preregistered experiments (1,372 people, 9,593 trials) the researchers secretly varied whether the AI was right. People consulted it on more than half of questions; accuracy rose about 25 percentage points when the AI was right and fell about 15 when it was wrong, and confidence went up either way — even after errors. Time pressure, cash incentives and feedback all moved baseline scores but never removed the pattern.
Why I recommend it: Free to read and free to download the full 58-page PDF — no account needed. Two honest flags. It is a working paper: posted by the authors, and the SSRN version has not been through journal peer review, so treat the numbers as a strong first result rather than settled fact. And the copyright line says all rights reserved — read and cite it, do not republish the text. The finding worth carrying around is the one about confidence: people felt surer of themselves after the AI led them wrong. That is exactly why the checks on our AI Basics page are worth doing out loud.
A free explainer of Roko's Basilisk, the AI thought experiment about a hypothetical future superintelligence that would punish people who knew about it but did not help bring it into existence.
