Hot Take: LLM can 'jump' | Yong Zheng-Xin
yongzx.github.io · 1,671 words · saved by 1 readers
An interconnected-knowledge thesis: what it means for scientific breakthroughs, potential safety risks, and continual learning.
I want to lodge my takes on the how LLMs can advance science through major breakthroughs, and what the assumptions behind my takes imply for safety risks and continual learning. Let me start with my disagreement with the ICML’26 position paper that “LLM can’t jump” by Tom Zahavy. Preliminaries In “LLMs can’t jump”, Tom used the General Relativity case study, where Einstein was inspired by his elevator thought experiment while trying to construct a mathematical theory of gravity. Mercury’s anomalous orbit later became an important test of the theory. Specifically, this required abductive…
saved by
related reading
- llms-cant-jump.pdftomzahavy.com
- LLM Daydreaming · Gwern.netgwern.net
- Taking LLMs Seriously (As Language Models) — LessWronglesswrong.com
- You can’t imitation-learn how to continual-learn — LessWronglesswrong.com
- Your Evals Will Break and You Won't See It Coming - Lun Wangwanglun1996.github.io
- Thoughts on AI in academiatheinfinitesimal.substack.com
- The Need for Verification in AI-Driven Scientific Discoveryalphaxiv.org
- Context Widowsartificialbureaucracy.substack.com
- I am worried about near-term non-LLM AI developments — LessWronglesswrong.com
- Machina Mirabilis - Michael Hlamichaelhla.com
- We May be Surprised Again: Why I take LLMs seriously.inference.vc
- Alex L. Zhangalexzhang13.github.io