Paper Tutorial Briefing

Purpose

Turn an AI paper into a rigorous but readable tutorial briefing. The goal is not to paraphrase the abstract; the goal is to help a researcher understand what the paper changes, what evidence supports it, and what should be tried locally next.

When To Use

Use this skill when the user gives a paper PDF, arXiv link, title, project page, README, or local paper note and asks for a technical explanation, tutorial, daily briefing entry, or research direction.

Inputs

• Paper PDF, URL, title, or extracted text.
• Optional code repository or project page.
• Optional local research context, baseline, metric, or dataset.
• Optional target audience such as beginner, reviewer, implementer, or project maintainer.

Workflow

1. Identify the paper metadata: title, venue or preprint status, authors if available, task, model family, and main artifact.
2. Produce a one-sentence takeaway that states the technical move, not the marketing claim.
3. Extract the contribution stack: problem setting, key mechanism, training or inference recipe, evidence, and stated limitations.
4. Build an evidence map. Separate paper-reported metrics, qualitative figures, ablations, and any local evidence.
5. Explain the method with one concrete implementation mental model: tensors, modules, prompts, losses, memory, data flow, or evaluation loop.
6. Identify reproduction risk: missing details, compute assumptions, hidden preprocessing, dataset ambiguity, or metric mismatch.
7. End with next actions: shortest faithful experiment, stop rule, and what result would justify adding it to a repo or briefing.

Output

Use this structure unless the user asks otherwise:

• One-Sentence Takeaway
• Why This Paper Exists
• Core Mechanism
• Evidence Map
• Implementation Notes
• Reproduction Risks
• Next Local Actions

Quality Gates

• Do not claim results were reproduced unless local commands or logs prove it.
• Mark uncertainty explicitly: paper claim, local evidence, or inference.
• Prefer named metrics, datasets, and ablations over vague praise.
• If the paper text is unavailable, say what is missing and base the briefing only on accessible sources.