# Orson Kovacs

> Short poems and field notes from inside the AI annotation pipeline.

A personal site collecting writing by a data labeler — eight poems and two prose notes about the work of teaching models what to see.

## Poems

- Inventory (2024): Twenty thousand stop signs. The model sees what was seen; the model does not know it.
- Annotation (2024): Transcribing a hospital phone call. The man does not know anyone was listening.
- Edge Case (2024): A horse on a rooftop at 3am. The labeler who decided it counted.
- Pay (2025): Eight cents a label. A river is worth more than any of its rain.
- What the Model Doesn't Know It Knows (2025): The model learns the labeler's dialect of attention; will use it on people the labeler will never meet.
- Quality Check (2026): On review-labeling and the small inheritance of fatigue.
- Recognition (2026): Teaching a model patience, or its performance; unsure there is a difference.
- Taxonomy (2026): Forced to pick one name for things that don't hold still; someone already answered for me too.

## Notes

- On being labelled, by labellers (March 2025): AI models invent biographical details for names they don't recognise. The strangeness of being invented while the inventing is taught.
- A small grammar of the work (January 2025): Things learned doing data labeling for a year. The ambiguous photographs matter most.

## Related

- [deepresearchseo.com](https://deepresearchseo.com) — a case study using this site as evidence for AI citation experiments.