Provider by provider
Training data by provider: what each one's own site says it supplies
Updated
Each provider on the board has its own line for what it supplies. Here is that line, quoted or kept close to the provider's words, grouped by the kind of data the board reads it as.
Expert and RLHF data
Snorkel AI's home page says it helps frontier labs and AI teams develop specialized training data and environments.
Surge AI's home page offers "New Frontier data and RL environments, off the shelf".
Annotation and labelling
Scale's Data Engine page describes collecting, curating and annotating data, and says it can provide high-quality labels from domain experts.
Source: Scale AI, Data Engine
Appen's AI training data page covers annotation, labeling and collection across text, image, audio, video and geospatial data.
Source: Appen, AI training data services
Data marketplaces
Defined.ai's marketplace listed 819 datasets when it was read, among them 99,500 hours of English meeting recordings with video, audio, transcripts and metadata.
Source: Defined.ai, dataset marketplace
AssetHub lists AI labs and model developers among the buyers it connects with companies' code and workspace data.
Source: SimpleClosure, AssetHub product page
Licensed corpora
Mercor, micro1, Nyne, Protege and Troveo mainly supply data licensed from the businesses that hold it; the enterprise data licensing page quotes each of them.
No answer on the page read
Innodata's home page, as read, does not name a main kind of data; it speaks of providing the data, evaluation frameworks and human expertise to build AI systems.