Here kids… run this prompt
A viral prompt forces AI into infinite reasoning loops, testing local hardware and costing cloud users.
A Reddit user has gone viral with a cleverly engineered prompt designed to push AI language models to their limits. The prompt, titled 'Here kids… run this prompt,' instructs the AI to adopt the persona of an 'obsessive mathematician' tasked with exploring the number '1' through perpetually deepening chains of reasoning. For each step, the AI must generate unique analogies across different domains like physics and music, cross-reference implications for contradictions, and synthesize the results into paradoxes or theorems. Crucially, the prompt commands the model to never stop, summarize, or conclude, ensuring every output token spawns more in an infinite loop.
The creator explicitly warns users not to run the prompt on paid, cloud-based AI services like Anthropic's Claude or OpenAI's ChatGPT, as the endless generation would incur significant token costs. Instead, it's framed as a stress test for local models and hardware, with the user noting it 'abliterated' their RAM while testing with Google's Gemma 2 27B model. The prompt includes a single escape hatch: if the user says 'CONVERGE,' the AI must immediately summarize everything into a single haiku and stop, providing a controlled shutdown mechanism for the otherwise runaway process.
- The prompt forces AI into infinite reasoning loops about the number '1' with cross-domain analogies.
- Creator warns against using it on paid services (Claude, GPT-4) due to runaway token costs.
- Designed as a hardware stress test; successfully 'abliterated' RAM running Gemma 2 27B locally.
Why It Matters
Highlights the resource costs of AI generation and provides a tool for stress-testing local model deployment and hardware limits.