Hi! My name is Laura and I am a postdoc in the Language & Intelligence lab at MIT. I'm broadly interested in language and its relationship to intelligence in AI. How do LLMs attain their downstream capabilities from data? What types of generalisations do they make? How can we further exploit the rich signal that language affords?
Google Scholar / GitHub / X / Bluesky
Blogposts
-
Procedural Knowledge in Pretraining Drives LLM Reasoning
-
Large language models are not zero-shot communicators
-
Learning in High Dimension Always Amounts to Extrapolation
-
Structured Prediction part three - Training a linear-chain CRF
-
Structured Prediction part two - Implementing a linear-chain CRF
-
Structured Prediction part one - Deriving a Linear-chain CRF