KAN author Ziming Liu: the "language of AI science" hasn't been invented, so LLMs can't learn it
ZimingLiu11 · x · 2026-10-24
Ziming Liu (KAN author, MIT) offers an analogy for why LLMs struggle at AI research: an LLM trained by Aristotle would know ancient Greek but not modern math or physics — those languages didn't exist yet. Likewise, the "language of Science of AI" hasn't been created by humans, so LLMs have never seen it.
His logic: if training intuition were like math or coding, more compute should yield better performance — but it doesn't, hence the language is yet to be invented. Accompanied by a three-panel AI-generated comic of Aristotle training a model that can only speak ancient Greek.
More from AGI Musings
- "People Still Manually Fill Forms and Code Like It's 1990," AI User Laments — djcows · 2026-10-24
- Terence Tao: AI turns beautiful mathematics into 'toxic waste' — TheObserverUK · 2026-10-24
- AI didn't democratize coding if companies decide what you're allowed to build — ThiefClashRoyale · 2026-10-24
- Runway CEO: generative AI quality, resolution and control double roughly every year — c_valenzuelab · 2026-10-24
- dhh: China's Programmers' Day shows why its engineering culture wins at AI — prasanna_says · 2026-10-24
- Founder: The Best AI Use Is Augmenting Your Best People, Not Full Agent Autonomy — jacob_posel · 2026-10-24