AI training data race puts human-written books in focus

Yonhap's feature highlights a strategic question about training inputs: what makes a collection valuable once text is abundant? Its account emphasizes human-written books published before widespread generative AI use. The relevant signal is interest in the provenance and character of training material, rather than a newly announced model or technical result. Readers should not mistake the publication of this feature for evidence that a fresh procurement deal occurred inside the monitoring window. ([yna.co.kr](https://www.yna.co.kr/view/AKR20261016020500017?utm_source=openai))

My assessment is that defensible data access could become an important competitive distinction. A useful collection must be evaluated for learning value, duplication and legal usability, not simply counted by volume. For an AGI forecast, however, acquiring books is an input rather than an outcome: it does not demonstrate better reasoning, transfer or reliable autonomy. The stronger evidence would be controlled results connecting a documented dataset to measurable improvements. Until those exist, this story belongs in the discussion of research resources and governance. It should not be treated as proof that scaling a particular corpus will overcome current capability limits.