By skyforbes 
BREVITY BREAKS AI
Limiting output length reveals AI isn’t intelligent — it’s chatty!
While looking for tasks to test how brittle AI’s reasoning becomes under the slightest pressure, I came across this deceptively simple math puzzle:
“When I was 9, my partner would have been 1/3 my age. Now I am 26 years old. How old is my partner?”
The answer of course is 20. Most LLMs get this right nowadays — if you let them think it through. But try telling a chatbot to answer in just one word.
That tiny constraint breaks everything (unless you’re using a model that secretly burns tokens in the background, like GPT-5 Thinking). Suddenly hallucinations appear, and we see the “intelligence” of AI is an illusion:
In fact, this “single-word simpleton effect” often happens when you ask chatbots something that requires more reasoning than a straightforward trivia question. “Multi-hop question answering” tasks require combining two or more reasoning steps or pieces of information. A direct question might be “What is the capital of France?”. It’s easy for AI to just pattern-match the answer. Whereas a MHQA task would be…