Seraphina Goldfarb-Tarrant

Planning failures in multilingual systems can be systematically diagnosed and mitigated, leading to significant performance gains in low-resource languages.

LLMs' harmful outputs stem from a surprisingly compact and unified set of weights, suggesting a fundamental, addressable structure underlying even emergent misalignment.

LLM performance isn't just about size, but about how efficiently they compress information during training, offering a new lens for understanding and predicting model capabilities.