CONCEPT
The Teapot Test
Gopnik's laboratory finding that when children and LLMs are asked how to draw a circle without a compass, children suggest teapots (round objects in the real world) and LLMs suggest rulers (statistically close to compasses in training text) — the sharpest empirical demonstration of the difference between discovery and imitation.
The teapot test is a class of experimental tasks developed in Gopnik's Berkeley laboratory to distinguish genuine innovation from sophisticated imitation. When asked to solve problems requiring novel solutions — not the application of known methods to familiar problems, but the generation of new methods for problems the training data did not contain — children and large language models produce systematically different answers. Children draw on their engagement with the physical, causal structure of the real world; LLMs draw on the statistical regularities of text. The children innovate. The machines imitate. The finding crystallizes in a single experiment what the entire cultural technology thesis argues at the theoretical level.
In The You On AI Field Guide
Asked how to draw a circle without a compass, the language models suggested rulers — because in the statistical landscape of training data, rulers are close