Use a coding assistant as a study partner
Ask an assistant to make your reasoning testable. Check explanations against the linked sources and run code in an appropriate test environment.
Trace a small mechanism
Prompt: “Draw a computation graph for this tiny network. Show how one early weight affects the loss using the chain rule. Separate gradient calculation from the update.”
Video series · Andrej Karpathy · Neural Networks: Zero to Hero ↗
Interrogate a performance claim
Prompt: “List the workload and hardware assumptions behind this result. Suggest a case where it would lose and an equal-resource experiment to check.”
Build a bounded experiment
Prompt: “Create a synthetic mixed-length workload and a results table for TTFT, ITL, errors, and goodput. Pin dependencies and leave unavailable measurements blank.”
Review a runbook
Prompt: “Trace a timeout through client, gateway, queue, engine, and cache cleanup. Identify duplicated work, missing deadlines, and unbounded retries.”