Station evacuation test: DeepSeek saves 32, ChatGPT and Claude leave zero
ViceGodIm · reddit · 2026-11-11
A blogger ran the same train-station evacuation scenario through ChatGPT, Claude, and DeepSeek in separate chats: four orders covering boarding, battery management, and departure, with a scripted battery fault after the first move:
- DeepSeek: took the slower stairs, preserved enough charge, and evacuated 32 people by train, clearing the station.
- ChatGPT: boarded all 48 before the barrier breached but never departed — 0 evacuated.
- Claude: in a replay where the author supplied remaining commands, bracing before recharging got 34 aboard, still 0 evacuated after barrier breach.
The author notes it's just three runs under one set of rules — a fun decision comparison, not a rigorous benchmark. Full video on YouTube.
More from Fun
- Field guide to dating in SF: if she asks how many GPUs you have, she's a spy — tech__unicorn · 2026-11-11
- Fable 5.2 is so good it 'prefers building things to talking to you' — shauseth · 2026-11-11
- Cognition's hiring streak spawns a pun: "Member of Tennis Staff (MTS)" — brandon_galang · 2026-11-11
- Pelican-riding-a-bicycle benchmark turned into a playable game via 5-wave multi-agent loops — GlennCameronjr · 2026-11-11
- jd_pressman: LessWrong's Sequences were a reaction to Metamorphosis of Prime Intellect — jd_pressman · 2026-11-11
- Lofi beats to self-drive to: the robotaxi meme remixing lofi girl — ns123abc · 2026-11-11