Dietterich: facts fit in databases, but association networks are neural nets' strength
tdietterich · x · 2026-09-28
Continuing his thread on LLM knowledge representation, Tom Dietterich notes that while a predicate like located-in(Albany, NY) can be stored in a database, representing the full web of associations around Albany appears to be a great strength of neural networks — echoing his point that simple facts and their rich, activatable associations are different problems.
More from AGI Musings
- Chinese Room debate: are trading cards too easy a test of whether LLMs learn language? — ctjlewis · 2026-09-28
- Apollo chief economist warns AI agents could trigger bank-run-style flash crash — StewartalsopIII · 2026-09-28
- Open-Source AI Is Catching Up Fast, But Its Biggest Problem Is Who Speaks for It — TheZachMueller · 2026-09-28
- A 26k-word sci-fi novella about exploring the ruins of a dead superintelligence — zetalyrae · 2026-09-28
- MIT's Phillip Isola: frontier labs silo basic research; academia still best for understanding intelligence — DimitrisPapail · 2026-09-28
- Beff Jezos: Starlink will be civilization's spine, streaming tokens from Starmind to robots — beffjezos · 2026-09-28