Review: Three leading AI models call Projekt Y 'rubbish'

Ingeniøren, a Danish tech outlet, asked three leading language models to review Projekt Y, Denmark's new AI-driven news medium. The piece ran on 28 October under the headline "Makværk, der lugter langt væk" (roughly, "rubbish that smells from afar"), and the English title reports that all three models called the output rubbish. Projekt Y is led by Michael Dyrby, a former news chief at TV2 and editor-in-chief at BT.

The visible text gives only a fragment of the findings. The models' detailed verdicts are behind a paywall and are not part of what was available here, so the specific criticisms each model made cannot be reported.

One important correction is visible. After publication, Y contacted Ingeniøren and said the article originally reviewed was one of the few not generated by AI, but written by hand. In response, Ingeniøren had the models review three other articles, plus excerpts from Y's liveblog about the Israel conflict. The review was thus redone on a different sample.

Reader comments, four in total, reflect this. One commenter said the piece lost its footing once the study was changed following Y's complaint, though they still called it a fun idea. Another said the episode raised a separate question: how journalists could publish text with poor grammar and misunderstood proverbs, which they felt the models' review underlined. A third asked how the models would rate Ingeniøren and Version2 by comparison.

The story matters as an early test of how a new Danish AI-based news outlet holds up, and of using language models as critics of journalism, with its limits shown by the need to swap the reviewed material.