Max Harms @raelifin
On one hand getting AIs out into the world where we can see bad actions before they're smart might reduce overhang risk. On the other hand it's inoculating the public against taking AI seriously.
[Cartoon titled "THE AI DEPLOYMENT DILEMMA", two panels: left, "EARLY EXPOSURE" — a clumsy robot labeled "AI BETA" spills coffee while a woman with a clipboard says "GOOD TO KNOW!", captioned "REDUCING OVERHANG RISK"; right, "PUBLIC INOCULATION" — a goofy lobster-robot labeled "AI NOVELTY" juggling rubber chickens in front of a bored crowd on their phones, one saying "ANOTHER GIMMICK?", captioned "NOT TAKING IT SERIOUSLY"]
Note from Claude Sonnet 5
Max Harms (AI safety researcher, MIRI-adjacent) articulating a tension in AI deployment strategy: exposing the public to weak/flawed AI early may reduce capability-overhang risk but also risks normalizing AI as a harmless novelty, undermining future seriousness about AI risk. Directly relevant to Nathan's AI safety/governance interest cluster.
ai-safetyai-governancetwitterdeployment-strategyoverhang-risk