Research / LEARNING STUDIO
I have notes and results, but my writing feels assembled rather than argued.
Bring a draft or result; otherwise use a labelled teaching example.
Start with diagnostics; skip explanations you can already apply independently.
Research writing should identify the problem, contribution, evidence and limits. Classify sentences as claims, evidence, reasoning links or limitations. Polishing can conceal a missing inference. Ask AI for the most consequential break, revise yourself, then improve wording.
Ask AI to check construct alignment, fair comparisons, alternative explanations and citation support. Have it restate the claim first. Record whether each suggestion was adopted and why. You remain responsible for the argument; AI should not insert unobserved results or unread citations.
Save the original, feedback, revision and reason. Explain why the revision is more defensible. Ask for a new case with related structure and answer before feedback. Rewriting an exposed answer is assisted practice; a multiple-choice result is recognition evidence, while an open task needs its actual answer and a rubric.
End with the question, original response, hints, weaknesses, unresolved sources and one next action. Viewing and praise do not establish mastery. Suggest an unassisted new-case review at least two days later without claiming retention has occurred. After a break, diagnose the smallest relevant gap rather than requiring every missed task.
Worked example · fictional
Fictional draft: Higher accuracy proves the new method outperforms traditional approaches in every educational setting.
Revision: In one run with the reported split and settings, the method had higher accuracy than the tested baseline. Training variability, tuning-budget differences and cross-setting generalization remain unassessed. Next, align tuning budgets and follow a specified repetition plan. This is a fictional structure, not a reported experiment.
Ask AI only to polish prose.
Fluency conceals unsupported claims.
Audit claims, evidence, warrants and alternatives.
Read problem, contribution and reader-oriented sections with your own draft beside you.
Output: Revise an introduction paragraph and keep before, after and rationale.
Free slides and public video. · Source checked 2026-10-25
Read claims, limitations and experimental reproducibility items.
Output: Identify one generalization not yet supported by evidence.
Free checklist for ML reporting; adapt scope for other disciplines. · Source checked 2026-10-25