Judge whether visual iteration actually helped

Published by Robert Glaser

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

The public work

Glaser published an experiment in which models generated an SVG, inspected a rasterized rendering, and iterated. His before-and-after results were mixed: several models added detail without fixing the underlying composition.

What to notice

Keep each rendered version and judge against explicit visual criteria; additional detail and extra iterations are not automatically improvements.

Keep the context

This is an informal creative experiment with subjective commentary, not a controlled or calibrated model ranking. Some iterations made the image more complicated without making it better; no human-skill claim follows.

AI use: Documented in the source.

The article names Claude Opus 4.1, Sonnet 4.5, Haiku 4.5, GPT-5.1, GPT-5, GPT-5-Codex and Gemini 2.5 Pro. It identifies Chrome DevTools MCP as a rendering tool, but not one common host.

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 simple illustration brief and define three visible success criteria. Draft an SVG, inspect its rendered image, and request one revision addressing a specific failed criterion. Compare both renderings side by side and keep the first if the revision is worse.

A check to adapt

The selected rendering satisfies more stated criteria without introducing a new defect; retain the comparison and any failed revision.

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.

Ground a short caption in the pictured product

OpenAI works through furniture images from an Amazon product dataset, generating descriptions and then concise captions. Printed examples show how the product title identifies the intended item when an image contains several objects.

Published byOpenAI Cookbook

Curated public exampleLesson

Operations & education

Teaching plots with Gapminder data

Software Carpentry’s lesson reads Gapminder country data and displays labeled GDP time-series and scatter plots. It shows table transposition, legends and file export, alongside learner exercises and accessibility guidance.

Published bySoftware Carpentry / The Carpentries

Curated public exampleWorkflow

Writing & design

Treat generated continuations as creative options

Sloan built an Atom editor plugin offering inline completions from a recurrent neural network trained on old magazine text. His public account describes the working interaction and concludes that its creative effects did not yet justify the effort.

Published byRobin Sloan