Disintegrator 38. Natural Language (w/ Leif Weatherby)
28 snips
Sep 24, 2025 Leif Weatherby, an NYU associate professor and author of *Language Machines*, dives into the exciting intersection of AI and language. He explores how large language models (LLMs) redefine traditional structuralist theories and argue for cultural AI's potential. Leif discusses the poetic functions of language in AI, the ideological implications behind data pretraining, and the uncanny outputs generated by LLMs. He critiques remainder humanism and champions a humanism that acknowledges complexity, highlighting the need for creative adaptation in our automated future.
AI Snips
Chapters
Books
Transcript
Episode notes
Math And Language Interoperate
- Large language models reveal an interoperability between mathematics and language rather than human-like cognition.
- Leif Weatherby sees transformers as systems where numbers and words interact to produce uncanny poetic outputs.
Structuralism Returns Via Attention
- Structuralism regains relevance because transformers perform intra-sequence relation weighting similar to Saussurean value.
- Weatherby links self-attention to Jakobson's poetic function as reframing internal linguistic relations.
Users Miss Conversational Personas
- Weatherby notes users react to model updates by missing earlier conversational behaviors, saying 'you killed my friend.'
- He uses this to show models are becoming socially grounded even without perceptual grounding.
