Firehose for 28 October 2026
New repositories covered agent tooling and image-model evaluation, with a plugin for documenting and analysing home noise pollution and a periodic check of whether image models can produce stylised tech diagrams with correct text and arrows. On Hugging Face, the new work was evaluation material for that graphics benchmark: its test prompts, every generated image, and the raw human ratings.
GitHub Repositories
2 repos created
- Noise Pollution Analysis GitHub · Agent plugin Agent skills to document home noise pollution and analyse recordings at the frequency level: hums, tones, low-frequency drone, cycling.
- Tech Graphics Benchmark GitHub · Evaluation Lightweight periodic check: can image models make colourful, stylised tech diagrams with exact text and correct arrows? Text-to-image and image-to-image.
Datasets
3 datasets created
- Tech Graphics Benchmark Hugging Face · Evaluation dataset Test prompts for Tech-Graphics-Benchmark: can an image model make a characterful banner or a restyled diagram for a tech repo or blog post without getting the text or the arrows wrong?
- Tech Graphics Benchmark Ratings Hugging Face · Evaluation dataset Raw human evaluation data for Tech-Graphics-Benchmark: one JSON file per annotated generation, written by the annotation Space.
- Tech Graphics Benchmark Runs Hugging Face · Evaluation dataset Every image generated for Tech-Graphics-Benchmark, one folder per round.