A research archive by Julian Cooper
Discover selected answers from a few creators, then help turn another video into reviewable research. Complete evidence boundaries and revision history remain in Chinese.
Watch.
Question.
Reconstruct.
English index · Chinese research5 creators are currently indexed. Inclusion is research intake, not endorsement.
Browse Chinese sources →No coding required. Submit a lead in GitHub Issues, or contribute a complete evidence record through a pull request.
Share the original URL, why it matters, and any known caption source.
Separate the creator’s claims, external checks, and your own judgment.
Validated work enters the private Academy review queue before publication.
Never upload restricted videos, credentials, cookies, or fabricated transcripts.
4 selected creators · not the complete archive
12:31A useful prompt to move from asking for answers to delegating reviewable outcomes. Its universal stop-using-ChatGPT prescription, 1–2% adoption estimate, and 10–100x productivity claims are not backed by reproducible evidence; the durable method is a scoped workflow with permissions, artifacts, and verification.
Account researchChinese research note · English summary not yet available.
1:15:07Watch selectively. AI likely amplifies differences in task selection, context, architecture, and verification, but the headline's 100× gap is not established as a general productivity fact.
15:13Do not buy hardware for DeepSeek V4.1 Flash based on the original video. The creator's correction and official model data invalidate its key capacity assumption.
12 published studies are available in the complete archive.
View all research →