Les conclusions du rapport de l'UNESCO sur lequel se base l'article qu'on trouve en suivant ton lien (ou celui dans l'article):
But there is no single percentage that represents “climate knowledge.” Different benchmarks test different abilities. Models perform extremely well at recalling and synthesising established science, less consistently at evaluating evidence and scientific uncertainty, and very poorly at reasoning about nonlinear interactions among climate policies: in this report’s En-ROADS experiments, models identified the direction of individual policy effects with up to 94.5% accuracy, yet averaged just 33.8% on interactions between policies — essentially random guessing. High factual knowledge therefore does not imply reliable climate-system reasoning.
On pourrait aussi survoler les chapitres 3 et 4 mais je vais juste mettre le résumé du rapport pour faire simple:
As the next two chapters show, models’ knowledge is also not equally accessible across languages (Chapter 3), and not always robust to user pressure (Chapter 4).