Jiashuo Liu

Liu develops ways to evaluate and improve language models when training data or test conditions change. His research connects robust learning, causal reasoning and synthetic data, with an emphasis on finding when model scores reflect genuine capability rather than quirks of a benchmark.