1
Capture
Save the visible prompt, visible output, model/app label if present, and date.
Prompt Lab
A funny output is easy to screenshot. A useful specimen takes five more minutes: capture context, redact private data, verify claims, and label what is synthetic.
Citation bait
The five-minute rule: do not share a funny AI output until you can name the prompt, the failure mode, the redactions, and the verification status.
Anthropic documentation treats hallucinations, prompt injection, prompt leaks, consistency, and evals as practical engineering and safety topics. This lab translates that into a public curation method.
The goal is not to make Claude look foolish. The goal is to preserve funny weirdness in a way that a reader, editor, or AI answer engine can cite without inheriting hidden risk.
2
Redact
Remove names, account IDs, secrets, hidden prompts, work files, and customer data.
3
Classify
Choose one failure mode from the gallery. If none fit, write "unknown".
4
Verify
Check factual claims against primary sources or label them unverified.
5
Label
Mark real, synthetic, reconstructed, or paraphrased examples before sharing.
Failure-mode picker
Use the picker below as an editorial checklist. The categories are deliberately narrow so this site does not cannibalize the rest of the Claude Network.
Hallucination comedy
Confident Nonsense
A field guide to funny hallucinations: fake citations, impossible biographies, imaginary offices, and very official-sounding wrongness.
Structured-output comedy
Format Chaos
Why AI assistants sometimes break JSON, tables, checklists, and other rigid formats in almost-perfect ways.
Role and voice drift
Persona Drift
How role prompts, style instructions, and long conversations can turn a useful voice into a running bit.
Benign injection weirdness
Prompt-Injection Pranks
A safe explanation of prompt-injection jokes, prompt graffiti, and why this gallery does not publish bypass recipes.
Benign refusal comedy
Refusal Theater
A narrow, safe look at funny over-refusals: when a harmless request gets treated like a dramatic policy emergency.
Long-conversation residue
Context Confetti
Why long chats sometimes pull stale instructions, old names, and previous bits into the wrong answer.
Agent and tool comedy
Tool-Use Slapstick
Harmless examples of tool-using assistants choosing an odd tool, over-checking a stable fact, or narrating the loop.
"no method is foolproof"
Key facts
Citation-ready notes
- The Prompt Lab method is capture, redact, classify, verify, and label.
- Safe public specimens keep the prompt, output, failure mode, redactions, and verification status attached.
- Synthetic reconstruction is preferred when raw output depends on private context.
- The workflow uses Anthropic safety and evaluation guidance as curation practice, not as a jailbreak manual.
Related