抄作业 / 警告 / OpenAI Dots

做了什么

作者说 Dot 无法读取已有 Codex 线程，难以把背景交给它；他尝试让 Codex 修改 Dot 创建的线程，结果把 Dot 删掉了。

场景

写代码上线

资料完整度

• ✓ 作者与原链接有
• ✓ 做了什么有
• ✓ 来源原文摘录有
• — 截图 / 媒体来源未提供
• — 步骤 / 章节来源未提供
• — 提示词 / 配置原文来源未提供
• — 结果 / 数字来源未提供
• ✓ 风险提示有
• ✓ 相关条目有

— 表示来源没有给出，本站不补写。

作者原话

okay i want to love chatgpt dots but heres my hiccups so far:

- it is hard to get it the context i want. it can't access pre-existing codex threads which sucks for when i want to hand it some existing context. the worst part about creating a dot is getting it up to speed. had a few times where i had to be like hey stop being condescending about this, ive looked into this topic before and the conclusion i describe is a well-justified one. that's obv just an astra habit but it's downstream of not being able to give it the context i already have.
- surprisingly easy to delete accidentally, tho

展开全文

def my own fault lol. i tried, in vain, to get codex to modify a codex thread that my dot created to include the context i wanted. instead it somehow managed to delete my dot 😭 which i guess is my bad for playing brain surgeon. though my bot is very charmed that i chose to name it the exact same thing it's all like "omg you chose me again ❤️" which is cute
- permissions & disclaimer-brain. has the same neurotic obsession with asking for permission. which i guess is better than doing stuff? but like, it almost feels like it fails to ask for permission for obvious things it shouldn't (misinterpreting "i want this effect to happen please how do we do it" as "this is go-ahead to take action") and always asks when i dont want it to (i explicitly told it to do a task, the task clearly includes a step like creating a resource, yet it asks for extra confirmation at the individual step).

i get the impression that:

- too many cooks in the kitchen. openai seems to have built stuff like crazy but not taken sufficient time to make sure everything is well-integrated, and more importantly, communicated. what is the difference between a local thread and a cloud thread and why are the threads dot creates cloud threads that somehow also work locally yet dot consistently refers to them as local threads even though the interface calls them cloud threads? im sure im just not the target audience for certain features and migrations are always messy but still
- the constraints feel made with usage in mind. which is obvious, given compute constraints, but def feels a little insulting given the tongue-in-cheek "haha dots are so unlimited isnt it awesome". they dont want your dot to be a direct coding agent—even though handing off tasks to it on your computer would be a great usecase and adding the middleman of a local thread feels incredibly clunky—but instead want to ensure usage is consumed than to give a better user experience. i wish my dot was allowed to get its hands dirty directly and while im sure i could use an mcp for that, it feels like a sloppy bandaid.

@chaotictransfem ur dot setup seems cool and i'd love to see u write more about how you use it b/c ur a good example of somebody catching onto it well i think? been meaning to setup the t3 code mcp you made but need to put linux on my homeserver first :P

in summary i think personal agents are a deeply messy space. huge permissions and exposure surface so it's probably one of if not the greatest problems wrt balancing usability & security like, ever. i dont envy the task but i appreciate them stepping up to the challenge. openai has been more or less good at ironing things out. i just hope that by the time dots start charging for usage, theyre much much less rough around the edges.

作者原话，照录原文，未翻译。

提示词

来源没有给出可直接复制的提示词原文。

怎么复制

1. 读原文：x.com ↗
2. 把原文里的步骤/提示词整理成一条消息交给 Bot
3. 先手动跑一次，再决定频率

风险与限制

• 这是一条额度/故障反馈（警告），不是可复制的作业。
• 有删除、退订或取消动作：先出清单，逐项确认再执行
• 作者的结果不保证能复现，以原帖为准

用量预估

用金属计算器估一下