Pedro Domingos: Transformers just read, write and combine symbols in embedding space
pmddomingos · x · 2026-10-18
Computer scientist Pedro Domingos, author of The Master Algorithm, offers a one-line reduction of what transformers actually do: they read, write and combine symbols in embedding space — nothing else. The statement distills his long-running argument that LLMs are fundamentally symbol manipulators, continuing the debate over whether transformers are genuine symbolic systems.
More from AGI Musings
- Nathan Young: AI extinction is very unlikely — but if it happens, it happens for everyone — NathanpmYoung · 2026-10-18
- Longevity influencer: today's workers are 'the last generation to have a career' — rand_longevity · 2026-10-18
- Startup lecture dissects 7 durable moats, opens with Google's chatbot blunder — richardreis · 2026-10-18
- Professor shares 'shocking' ChatGPT graph on academia vs industry gap — asusarla · 2026-10-18
- Stuart Russell vs Schmidhuber debate: could AI destroy the world? — SchmidhuberAI · 2026-10-18
- No Robots, No Apocalypse: Why AI's Best and Worst Outcomes May Be Years Away — hrrm · 2026-10-18