大型语言模型（LLM）的创造力，能否通过重新构建语言而产生？
1 分•作者: nopelican•3 个月前
今天，一个被提示要新颖和有创造力的 LLM 解决了其中一个埃尔德什问题，一个 hnuser，energy123：阅读提示[1]，人们不禁想知道，激发模型变得非传统是否是成功的关键：“这……可能需要非凡的、有创造力和新颖的元素”。这与我最近一直在探索的想法相关联，即设计一种方法，让 LLM 以一种新的方式处理问题。我的初步想法是使用一种文件夹语言，将诸如生活成本上升或住房负担能力等问题转化为一种由文件夹组成的语言，其中每个文件夹代表一个主体，例如租房者、购房者、现状等。并且有一个供 LLM 和人类协作的 Web 服务器，同时 LLM 的系统提示会根据正在开发的文件夹结构进行动态调整。
查看原文
Today an LLM prompted to be novel and creative solved one Erdos Problem, one hnuser, energy123: Reading the prompt[1], one wonders if stoking the model to be unconventional is part of the success: "this ... may require non-trivial, creative and novel elements". That connect what the ideas I have been exploring lately, design a way so that the LLM can work on a problem in a new way. My initial idea was a folder language, a problem like, increasing cost of living or house affordability is converted in a language amongs folders in which each folder express an agent, for example renter, buyers, status quo, etc. and there is a web server for LLM and humans to collaborate, also the system prompt for LLMs is dynamically addapted to the folder structure that is being developed.