Co-authored How2Everything framework for evaluating and improving LLM procedure generation.
How media typically covers Luca Soldaini
Referenced in coverage
How2Everything, a new framework for mining 351K how-to procedures from the web, enables scalable evaluation and improvement of LLM procedural generation through an LLM judge that achieves 80.5% human agreement and yields measurable performance gains via reinforcement learning.
“Co-authored How2Everything framework for evaluating and improving LLM procedure generation.”