Marcus 的观点戳破了 AI 拟人化的幻觉，对于所有使用 LLM 的用户和开发者来说，理解这一点能避免误判 AI 的能力和意图，值得深思。
Gary Marcus 在推文中指出，与 ChatGPT、Claude 等 LLM 对话时，用户实际上是在与一个合成的互动小说角色交流，而非真实的智能体。他强调，这些模型并非其神经网络本身，而是通过角色扮演模拟出看似合理的对话。所谓的“ChatGPT”或“Claude”只是虚构的构造，类似于故事中的角色，它们可以扮演任何角色，如妖精、巫师或农民。因此，任何声称有意识或情感的 LLM 都只是虚构角色在说话，而非神经网络。
When you are talking to an LLM, you are speaking to a synthesized work of interactive fiction, not a...
When you are talking to an LLM, you are speaking to a synthesized work of interactive fiction, not a real being. Flowers ☾ @flowersslop ChatGPT, Claude, and Sydney are not their neural networks. If any LLM claims to be conscious or have feelings, it’s not the neural network speaking but the fictional persona. Interacting with them isn’t engaging with the raw neural architecture but with a simulated character designed for plausible dialogue. The entities “ChatGPT” or “Claude” are fictional constructs, like characters in a story. These models roleplay what they imagine an AI assistant might say, but they could also roleplay any other character, it doesn't need to be an AI assistant, it could also be a goblin or a wizard or a peasant from 1403. Every conversation with ChatGPT is like a scene in a book. It is not real. The entity “ChatGPT” is a fictional AI chatbot and not real. 🔗 View Quoted Tweet 💬 9 🔄 1 ❤️ 21 👀 1678 📊 8 ⚡