Rebutting 'LLM Can't Jump': Interconnected Knowledge Could Spark Scientific Breakthroughs
burny_tech · x · 2026-08-09
The author challenges the ICML'26 position paper "LLM can't jump", which argued that LLMs cannot make major scientific breakthroughs like Einstein's General Relativity because they lack reliable world models and physical priors for abductive reasoning.
Arguing for "LLM can jump", the author posits that the interconnected knowledge formed during massive pre-training could enable models to bridge concepts and make similar leaps. The essay further explores the profound implications of such capabilities for AI safety risks and continual learning.
More from AGI Musings
- Peter Diamandis: High Schoolers Will Submit Hypotheses to National Labs Run by Autonomous AI — PeterDiamandis · 2026-08-09
- Public Unbothered by OpenAI/HuggingFace Incident, Outrage Fails — ctjlewis · 2026-08-09
- AI makes game visuals easy, but game design remains a human frontier — pvncher · 2026-08-09
- Vibe Coding Ships Features Faster, But Customer Attention Doesn't Scale — Ubunta · 2026-08-09
- Cathie Wood: AI and Productivity Gains Are Driving Corporate Profits to Historic Highs — CathieDWood · 2026-08-09
- Why Is There No 'App Store' for Independent AI Agents Yet? — mgsz_ · 2026-08-09