Create intermediate Works with: Any AI

Model Error Analysis Session

Table output

Learn from the mistakes of a model.

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Run an error analysis. Task and model: [TASK_MODEL]. Sample of errors with inputs, predictions and truth: [ERROR_SAMPLE]. Metrics: [METRICS].

Categorise errors into groups (label noise, ambiguous cases, missing features, rare classes, distribution shift) using the sample and count them in a table. Slice performance by segments I can define and identify weak spots. Propose five improvements ranked by expected impact and effort, with the evidence for each, and an experiment to test the top one. Include questions to ask domain experts. Recommend how to document findings and how to avoid overfitting to the error sample.

Add a template table with columns for error id, input summary, predicted, actual, error category, and suspected cause, and a rule for sampling errors without cherry-picking. Include a one-page summary format for sharing findings with the team and a plan for retesting after each change.

Also add three questions to ask the people who labelled the data about cases they found hard.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

• ml