RLHFJobs's published data
1 dataset, put together from the sources' own pages and refreshed on a set cycle; last read 2026-10-22.
| Dataset | Headline figure | Sources | Read |
|---|---|---|---|
| RLHF work by platform: the tasks, tracks and entry rules | 6 of 12 platforms | 12 of read | 2026-10-22 |
The method
The work starts with a question and the URLs of the pages that should answer it. We read each one and record the figure exactly as the page puts it, with the link and the date of the read.
Nothing counts until a reviewer has accepted the sentence, and every rejection is noted with its reason. One figure per source feeds the median, or one for every breakdown a source prices. Silent sources are recorded as such, and later reads reveal what has changed.
RLHF work by platform: the tasks, tracks and entry rules
6 of 12 platforms whose own pages we read describe ranking, comparing or rating AI model responses, the core of RLHF work, read 22 October 2026.
Ranks, compares or rates model responses, by platform: not stated 6 of 12, describes ranking, comparing or rating responses 6 of 12 (12 platforms, read 22 October 2026).
Spreadsheet (CSV) (90 rows) · JSON · Markdown