Measuring cup logic puzzle exposes LLM reasoning gaps

LLMs remain trapped in a world of abstract logic, often failing to simulate the most basic physical-world constraints.

• –Models over-index on mathematical "water jug" riddle patterns from their training data instead of applying physical-world grounding.
• –The failure demonstrates that even advanced reasoning models rely on heuristic pattern-matching over true spatial simulation.
• –This "vibe check" benchmark highlights the brittleness of AI logic when faced with simple, non-abstracted real-world tools.