Nick Nisi describes an experimental code-reviewCode review examines proposed source changes to assess their correctness, clarity and suitability before or around integration. system that translates features of code changes into signals for a simulated fruit fly connectome. The model learns from labelled examples and produces an approval or rejection signal; it does not understand or explain JavaScript.
In the audience exercise, the simulated fly makes one mistake across three examples. Nisi reports broader test resultsEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. and shows a live request that does not produce a clear answer, while cautioning that the experiment is not a practical replacement for code review.
Nisi proposes combining team-specific review rules, risk classification, automated reviewers and human escalationHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work.. He suggests providing test results, screenshots and other evidence to help people understand what a change does.
Watch on YouTube
