Compare audio and visuals before summarizing video

Published by OpenAI Cookbook / Juston Forte

This credits the original publisher. Better Loop membership or a shared assessment is not implied.

The public work

OpenAI’s notebook processes a real DevDay recap into frames and audio, then prints summaries and answers from visual input, audio-derived text, and both together. The differing outputs make missing context visible.

What to notice

Compare what each input channel supports, then verify the combined account against the source instead of assuming more context guarantees accuracy.

Keep the context

This archived notebook demonstrates outputs, not an independently scored accuracy comparison. Historical API descriptions and some model-produced names are unreliable; the source video and claims still need checking.

AI use: Documented in the source.

The displayed workflow uses gpt-4o-mini and gpt-4o-audio-preview through the OpenAI API. Its prose also refers to Whisper, but the displayed transcription call uses the audio-preview model.

A useful public example is not an assessment of a reader, a publisher or a Better Loop member.

Authored practice suggestion

Try the idea.
Check your own work.

Use material you are allowed to work with. This suggestion is preparation; it does not record a completed task or an improvement.

A prompt to adapt privately

Choose a short public talk with a transcript and visible slides. Draft a summary from the transcript, inspect the slides for missing or conflicting points, then revise. For each added claim, record whether it came from speech, visuals or both, and flag unresolved discrepancies.

A check to adapt

Each substantive claim is supported by the talk or slides, and unresolved audio-visual disagreements remain explicit.

Open a related practice brief

Private practice needs no account. Your configured model provider may process the evidence you select.

See the full skill-to-sharing flow

Another way into the work.

Mapping grocery proximity in Chicago

The guide projects Chicago community areas and grocery locations into a metric coordinate system, buffers stores by one kilometre, and displays intersection and difference maps. It defines coverage geometrically.

Published byGeoPandas developers

Curated public exampleReport

Research & strategy

Accounting for food-system emissions

Hannah Ritchie’s analysis visualizes food emissions using Poore and Nemecek’s published meta-analysis. It separates livestock, crops, land use and supply-chain contributions, explaining category boundaries and the distinction between food-system and global totals.

Published byHannah Ritchie / Our World in Data

Curated public exampleNotebook

Everyday work

Control summary detail without confusing length and quality

OpenAI summarizes a Wikipedia article about artificial intelligence using different chunk counts and a detail parameter. The notebook prints summaries and their lengths, including a recursive variant that supplies earlier summaries as context.

Published byOpenAI Cookbook