Language Model Reasoning: A Comparative Analysis Case Study by Google DeepMind

​There has been substantial debate about the capabilities of language models — for example whether they can be said to 'reason', or simply repeat familiar patterns.

In this talk, our guest, Andrew Lampinen from Google DeepMind, will suggest that approaches from cognitive science can provide useful tools for approaching these questions.

Specifically, he will focus on comparative methods (comparing capabilities across different systems) and rational analysis (analyzing behaviors as a rational adaptation to an environment).

Andrew will use these themes to discuss his recent work evaluating the way that both language models and humans entangle content in their responses to logical reasoning problems, and then discuss the broader implications for understanding and improving language model performance.

Speaker's bio:
Andrew Lampinen's research bridges cognitive science and AI, often with a focus on how the complex behaviors and representations of models, agents, or humans emerge from their learning experiences or data.
His work covers topics ranging from interpretability, to explanations as a learning signal, to embodied intelligence. He is currently a Staff Research Scientist at Google DeepMind. Before that, he completed his PhD in cognitive psychology at Stanford University, and his BA in mathematics and physics at UC Berkeley. 

Join the BuzzRobot Slack channel to stay in touch with the community