Foreigners Provide Datasets, While Domestic Folks Blow Hot Air: A Sharp Review of Nüwa, Musk, Jobs Skill

A Chinese-language critique says some domestic AI “open source” projects overstate their capabilities in READMEs while withholding the raw data needed to verify them.

The piece argues that prompts, screenshots, and documentation are not enough; real openness requires datasets, annotations, scripts, and reproducible pipelines.

It contrasts this with projects like EleutherAI, The Pile, and LAION, which publish datasets, code, and citations through platforms such as Zenodo and Hugging Face.

The core dispute is whether AI is truly open source if only code or prompts are shared but the underlying training data and methodology remain hidden.