Recently, The New York Times reported on Anthropic co-founder Christopher Olah’s meetings with religious leaders. The company sought counsel on making Claude morally “good” and asked them to consider whether AI models might be conscious. This episode served as another reminder to the effective altruists that declaring an intention to do good is not always the same as doing it.
Olah is one of the leading researchers trying to understand what happens inside neural networks. He describes his work through the metaphor of a garden trellis: engineers build the structure and create the conditions, but something then grows alongside it. He is, in a sense, a mathematical gardener.
