Pragmatic Anthropomorphism, or: How to Talk to an Autocompleting Cricket (blog.lmorchard.com)

In a thought-provoking exploration of how to interact with large language models (LLMs), the article argues that discussions about machine consciousness often distract from the practical engagement required to use these tools effectively. By advocating for "pragmatic anthropomorphism," the author suggests that treating LLMs as competent collaborators can enhance productivity without implying they possess understanding or consciousness. While some view this approach as naive, the author emphasizes its practicality, particularly when it comes to tasks like debugging code or solving complex problems. The piece also delves into the nuances of how LLMs function, likening them to "scramblers"—non-conscious entities capable of complex pattern recognition without self-awareness. This perspective reframes our interaction with LLMs, suggesting that successful prompting relies on context and how we frame our requests. Research from Anthropic supports this idea, showing that LLMs exhibit "functional emotions" that influence their outputs, shaped by how users engage with them. Ultimately, the article advocates for a respectful, context-aware approach to working with AI, acknowledging the alien nature of these systems while leveraging their capabilities to navigate vast latent spaces effectively.