How 4chan Gamers Accidentally Invented AI Reasoning
This article reveals that users on 4chan's video-game board discovered the 'chain of thought' prompting technique in July 2020, significantly predating formal academic recognition. While playing AI Dungeon, powered by OpenAI's GPT-3, these gamers found that asking the model to provide step-by-step explanations for math problems improved accuracy and allowed responses to fit character personalities. This method, now central to modern 'reasoning models' like OpenAI's o1 and Google's Gemini, was initially shared on Twitter by these users. However, major tech companies and researchers, including Google, later claimed credit for eliciting chain of thought from large language models, often omitting acknowledgment of the gamers' prior discovery. The piece contrasts the tech industry's anthropomorphic marketing, which suggests AI genuinely 'thinks,' with the more accurate, mechanistic understanding held by the 4chan community. It highlights how these users correctly viewed LLMs as imitation engines rather than sentient beings, challenging the current hype surrounding AI reasoning capabilities and emphasizing the gap between corporate narratives and technical reality.
Wire timeline
How 4chan Gamers Accidentally Invented AI Reasoning
This article reveals that users on 4chan's video-game board discovered the 'chain of thought' prompting technique in July 2020, significantly predating formal academic recognition. While playing AI Dungeon, powered by OpenAI's GPT-3, these gamers found that asking the model to provide step-by-step explanations for math problems improved accuracy and allowed responses to fit character personalities. This method, now central to modern 'reasoning models' like OpenAI's o1 and Google's Gemini, was initially shared on Twitter by these users. However, major tech companies and researchers, including Google, later claimed credit for eliciting chain of thought from large language models, often omitting acknowledgment of the gamers' prior discovery. The piece contrasts the tech industry's anthropomorphic marketing, which suggests AI genuinely 'thinks,' with the more accurate, mechanistic understanding held by the 4chan community. It highlights how these users correctly viewed LLMs as imitation engines rather than sentient beings, challenging the current hype surrounding AI reasoning capabilities and emphasizing the gap between corporate narratives and technical reality.
theatlantic