English Español Français Deutsch
← Back to the list
Hallucination and factuality Methodologists Directly upstream of ai100's methodology

Luke Zettlemoyer

Whether open-weight language models can be built at frontier scale and whether their factuality can be measured at fine resolution.

Zettlemoyer's name appears on a roster of work that is closer to a research program than a list of papers — open large language models that anyone can examine (OPT in 2022 and the open-model line that followed), and the evaluation infrastructure to say whether those models keep their factual claims straight. FActScore (2023, lead author Sewon Min) is the methodologically careful version: break a long generated paragraph into atomic factual claims and score each against a knowledge source, rather than score "the whole paragraph" with a single judgment. For ai100 the resolution matters — a model that gets one fact right and two wrong in the same sentence should not get a passing grade.

Worth following when
you want fine-grained factuality scoring rather than a single yes-or-no judgment per response.
Topics
atomic factuality scoring; open large language models and their evaluation; semantic parsing as a precursor to today's structured-output methods.
Key works
FActScore (2023, co-author); OPT open language model (2022, co-author); foundational semantic parsing work in the 2000s.