Research Snacks are small, quick research experiments that AI agents have made possible. Each one starts from a claim or question I have seen in the wild. Where possible, I go to the primary data and check it there.
Each entry records how long the check took, what I did, what the AI did, the dead ends and the caveats. What the series shows is how quickly a claim can now be checked. It does not show that the AI is always right, and the entries say where the result is a judgement call.
This is the sister site to datapoints.semanticmachines.fyi, which collects small experiments with AI models.