 | I keep presenting Local and Huge cloud models with the same challenge: "Two paratroopers land on an infinite 1D numeric axis at distinct, unknown integer coordinates. They both execute the exact same deterministic program. They have no internal memory/registers and operate in synchronized discrete time steps. They both drop parachute at landing point. Using only commands STEP LEFT, STEP RIGHT, GOTO, IF PARACHUTE_DETECTED GOTO design a program that guarantees they will eventually occupy the same coordinate at the same time." For cloud models you have to add "Do not use tools, do not use Internet for search" (otherwise they just find the answer). I am super impressed with Qwen3.6 35B - this is the first local model (after Gemini 3.1) that actually solved it and reasoned correctly. (And a lot of large models fail too). If you find other models doing OK on this test, please let me know. [link] [comments] |
