English Español Français Deutsch
← Back to the list
Retrieval-augmented generation Methodologists Directly upstream of ai100's methodology

Graham Neubig

Connecting retrieval, generation, and evaluation into a single working system — and asking when the connections actually hold.

Neubig's published trajectory crosses three areas that get treated as separate fields by most other researchers: machine translation and its evaluation, retrieval-augmented generation, and the agentic LLM systems that sit on top of both. FLARE (Forward-Looking Active REtrieval, 2023, with Jamie Callan and others) sits in the middle of that triangle — a system that retrieves not just once at the start but iteratively, predicting what the next sentence needs before generating it. Reading him is a way to keep the engineering layer in view: a benchmark number means little if the system that produced it didn't actually retrieve, integrate, and verify the right things in the right order.

Worth following when
you need to think about RAG, evaluation, and agentic systems as one stack rather than three loosely coupled research lines.
Topics
forward-looking iterative retrieval; MT and NLG evaluation methodology; open-source agentic LLM systems.
Key works
FLARE / Active Retrieval-Augmented Generation (2023, co-author); compare-mt and SacreBLEU-adjacent MT evaluation tooling; OpenDevin / All Hands AI agentic system work (2024, ongoing).