#84 LAURA RUIS - Large language models are not zero-shot communicators [NEURIPS UNPLUGGED]

Research Focus

Laura Ruis's research focuses on the pragmatic reasoning capabilities of large language models, an area she believes holds significant potential for advancement. She notes that while these models excel in compositional generalization, they struggle with pragmatic inferences, which are crucial for effective communication 1. Laura suggests that reinforcement learning from human feedback (RLHF) could improve these models by aligning them with human preferences, thus enhancing their ability to make pragmatic inferences 2.

Pragmatic inference is really a social skill that we have. There's a lot of pragmatic pressures that you encounter while just acting in the world and navigating communication.

--- Laura Ruis

She is exploring how interactivity and social embeddedness can be integrated into these models to improve their pragmatic reasoning capabilities 1.

   

Future Prospects

Looking ahead, Laura is excited about exploring interactive setups to understand when pragmatic inferences might emerge in language models. She emphasizes the importance of creativity in both interacting with these models and identifying their failure modes 3. Pragmatics, she explains, involves understanding how context and shared experiences influence meaning, a challenge for language models that typically rely on syntax and semantics 4.

The creativity of people is really needed to get some kind of interesting response out of these models.

--- Laura Ruis

Laura is particularly interested in how these models can be developed to become serious communicators, capable of symbolic generalization and pragmatic reasoning 3.