Mechanistic Interpretability
Jacob expresses skepticism about using mechanistic interpretability to uncover philosophical truths about language. He suggests that while understanding how language models operate can provide insights into language acquisition, it doesn't necessarily reveal how reference works in human language. The surprising finding that vast amounts of text data can lead to adult-like linguistic competence raises further questions about the mechanisms behind human language learning.In this clip
From this podcast
The Gradient
Jacob Andreas: Language, Grounding, and World Models
Related Questions