"Ambient Programming" Namer Kapasi's Tip for Improving the Quality of Your AI's Answers: Chat with it for 10 Minutes

IT House reported on July 25 that Anthropic well-known AI researcher and the proposer of the concept of "atmosphere programming" Andrei Kapaci believes that when using AI, you don't have to work hard to polish the prompt words: don't think about it, just say whatever comes to mind, just keep saying it. Kapasi said that he sometimes turns on the voice mode and talks to the AI ​​for about 10 minutes continuously. The ideas do not need to be complete, and the content does not need to be organized in advance. The whole process is messy, saying everything, and saying everything exactly as he thinks. "Sometimes, large language models need to get more information to really understand what you want to accomplish." Kapasi said that chatting can allow users and chatbots to divide labor more reasonably. Users only need to speak out all the unformed project ideas in their minds, and then the chatbot organizes them into clear and organized content. He usually reminds the chatbot at the beginning that his thoughts may be confusing, and the speech transcription results may also contain typos. "This allows people and models to understand each other better, and there will be much less corrections to be made later." Capasi's views come as Silicon Valley companies are focusing on improving AI voice functions and trying to get users to switch from keyboard input to direct communication. OpenAI launched GPT-Live earlier this month to provide support for the ChatGPT voice function. It also launched a mini keyboard priced at US$230 (IT House Note: The current exchange rate is approximately 1,560 yuan). It has a voice recording button that can directly hand over the prompts spoken by the user to the intelligent agent for processing. Anthropic also upgraded its voice model this Thursday, and users can choose Opus, Sonnet or Haiku.