Andrej Karpathy’s trick for getting better answers from AI: Just ramble at it for 10 minutes (www.businessinsider.com)

Andrej Karpathy, a prominent AI researcher at Anthropic, recently shared an unconventional approach to enhancing interactions with large language models (LLMs) on X. Instead of crafting polished prompts, Karpathy advocates for “rambling” for about 10 minutes in a "stream of consciousness" style. He argues that this method enables users to unload their jumbled thoughts, giving the AI ample context to reorganize and produce clearer, more structured responses. By warning the AI about potential typos and messy ideas upfront, users can improve the overall synergy and minimize the need for corrections later. This technique is particularly relevant as AI companies like OpenAI and Anthropic enhance their voice interaction capabilities, making conversational AI more accessible and intuitive. OpenAI's recent launch of GPT-Live and Anthropic's updates indicate a shift towards encouraging users to articulate their thoughts verbally. Karpathy's strategy showcases the evolving landscape of human-AI collaboration, emphasizing that unconventional methods can yield significant improvements in AI comprehension and output quality.