How 4chan Gamers Accidentally Invented AI Reasoning

In 2020, 4chan users playing AI Dungeon found that asking GPT-3 to explain step by step often improved results, especially on math tasks.

That early workaround effectively uncovered what is now called chain-of-thought prompting, before it became a mainstream AI term.

The article says later companies like Google, OpenAI, and Anthropic repackaged similar behavior as evidence of advanced model reasoning.

The episode suggests a key LLM technique was first identified by nonexperts, not invented by recent model makers.