Which custom instruction in ChatGPT could you not live without?
热度趋势
百分比基于当前可用热度信号，而非评论数或独立用户人数。
OpenAI 相关模型动态已经出现，适合跟踪能力变化、生态影响和后续可用性。
一位开发者社区用户讨论了他们在ChatGPT中自定义指令的演变。最初，他们使用大量指令来确保回复紧凑、信息密集，避免冗余并整合表情符号。然而，他们现在觉得许多指令对于当前模型来说已不再必要。他们正在考虑是否指示AI提供更简洁的初始回复，尽管他们发现这难以一概而论，并且可能更倾向于在特定情况下使用新的记忆系统。他们询问其他用户目前有哪些不可或缺的自定义指令。
I used to have relatively extensive custom instructions and frequently defined that it should think through things compactly, increase information density, avoid unnecessary filler sentences, integrate emojis before the punctuation of sentences, avoid hyphens and more.
However, with the current models, I increasingly feel that many of these instructions are no longer needed, so I recently removed everything for tests. Lately, I've only been thinking about whether it makes sense to communicate in some form that initial responses shouldn't always consist of huge text paragraphs, meaning the answers can be a bit more compact and to the point.
On the other hand, I find this hard to generalize and might prefer working with the new memory system for specific cases. Which custom instructions do you currently have implemented, i.e., which user-defined instructions can you not do without?