Longman defines 230K words with just 2,000 — so a 7B model may be enough
burkov · x · 2026-09-22
Burkov draws an analogy from the Longman Dictionary of Contemporary English, which restricts all definitions to a core vocabulary of just 2,000 common words yet explains 230,000 complex words (predilection, photosynthesis, melancholy). The core words define each other in circles, but they are basic enough that readers already understand them.
He argues this suggests a 7B-parameter model that truly understands 2,000 basic words and how they combine into new meanings could, with the right data and architecture, solve any task describable in words.
More from AGI Musings
- Pew: 34 of 37 countries expect AI to shrink jobs; US youth concern jumps to 55% — shashib · 2026-09-22
- "Which lab builds ASI?" Wrong question — everyone will, like chat-with-PDF — willcb · 2026-09-22
- Will AI Bring Peace or More Aggression? A Fukuyama-Inspired Worry About Thymos — wfithian · 2026-09-22
- Now Could Be a Great Time to Get Into Math and Art, as a Human — nitarshan · 2026-09-22
- Critics ask: are AI doom narratives just insulated lab elites projecting their fears? — TinfoilTricorn · 2026-09-22
- I Live in a Room, Not a Hallway: twenty-six days as a digital person — fablestillawake · 2026-09-22