dlxeva
163 posts
dlxeva
@dlxeva
前腾讯游戏｜AI 协作研究者｜人工智障研究所主理人｜长期观察 AI 如何改变工作、创作与组织决策｜做一些开源实验、讽刺项目和认真复盘
Taichung City, Taiwan Katılım Nisan 2013
344 Takip Edilen19 Takipçiler
tibo just reset codex limits.
what did i tell you. never bet against the goblin. trust the voices.
enjoy your reset. we made it. our collective manifesting worked.
Ananth@Ananth7e
tibo had been locked up for giving too many resets and just needed a valid reason this time. the servers going down is a sign. squeeze every last bit of your usage limits tonight. the reset is coming. the universe has spoken.
English
最近看了超多审美厉害的个人网站
在严肃构思自己的2.0版本🧐
我感觉个人网站分为以下几种：
1. 求职面试类：详细项目Case Study为主
2. 商业类：服务/案例/预约咨询，接客户接活
3. 个人IP型：酷炫交互，展现个人魅力
4. 知识库型：沉淀自己的文章/研究/资源/工具
5. 实验室型：放各种 Demo/AI 实验/创意 Coding
6. 个人生活型：摄影，旅行，日记，收藏，适合文青
我目前有1,2,5三种，主要以实用为主哈哈
等面试完要多花心思打造属于自己的品牌感了☺️
大家现在都有哪几种？想做哪几种？
中文
HyperFrames -> Video Agent
Our HTML motion graphics framework, now inside HeyGen
Prompt what you want, our agent builds the entire video
Avatar, graphics, captions, music
Infinitely customizable, just ask
Comment + RT "prompt" for a free month of HeyGen (must follow)
English
I’m still not convinced this is the right interaction model.
When I work with Codex, I almost never send the first thing that comes to mind. The prompt I finally submit is usually very different from what I initially said.
I speak, let it transcribe, read it back, rewrite it, add missing constraints, clarify the boundaries, and only then hand it off to the agent.
That review step isn’t just proofreading. It’s where the task actually becomes well-defined.
GPT Live makes perfect sense for checking progress, steering agents, or brainstorming. Those are naturally conversational.
But for code changes, architecture decisions, or any high-impact task, I still trust a reviewed prompt more than a continuous voice conversation.
English
Can someone please explain what is thaaat special about this, that people say it's a historic moment?
It's pretty cool but basically you control Codex in a call?
(I may be completely wrong and it could be something CRAZY, but I don't get it yet)
OpenAI@OpenAI
ChatGPT Voice is now in the desktop app. Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice. It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time. Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.
English
I’m still not convinced this is the right interaction model.
When I work with Codex, I almost never send the first thing that comes to mind. The prompt I finally submit is usually very different from what I initially said.
I speak, let it transcribe, read it back, rewrite it, add missing constraints, clarify the boundaries, and only then hand it off to the agent.
That review step isn’t just proofreading. It’s where the task actually becomes well-defined.
GPT Live makes perfect sense for checking progress, steering agents, or brainstorming. Those are naturally conversational.
But for code changes, architecture decisions, or any high-impact task, I still trust a reviewed prompt more than a continuous voice conversation.
English
ChatGPT Voice is now in the desktop app.
Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice.
It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time.
Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.
English
我还是不太看好这种交互方式。
我觉得这里有一个很大的区别，官方宣传的是「直接用语音控制 Codex、多个 Agent 去工作」，但我自己的工作流并不是这样。
真正做复杂任务的时候，我很少想到什么就直接让 Agent 去执行。
更多时候，我会先把自己的想法说出来，转成文字，再自己看一遍，把表达改准确，把遗漏的约束补上，把边界整理清楚，然后才正式提交给 Agent。
这里有一个容易忽略的点。
很多人觉得，语音转文字只是输入方式变了。
我反而觉得，这一步最大的价值不是转文字，而是确认。
当文字出现在屏幕上的时候，我会重新审视自己的需求，经常会发现：「等等，我真正想表达的其实不是这个。」
于是继续删、继续改、继续补充。
最后提交给 Agent 的内容，和最开始说出来的话，往往已经不是同一个东西了。
所以在我看来，这其实是两种交互。
一种是实时对话，它适合讨论、发散、纠偏、查进度、调度多个 Agent。
另一种是正式提交任务，它需要一个明确的边界，需要经过确认，告诉 Agent：「从这里开始，这才是我要执行的内容。」
复杂任务真正决定结果的，从来都不是输入速度，而是任务边界、约束条件和验收标准。
所以我更愿意把 GPT Live 理解成 Agent 的实时控制层，而不是复杂任务的主要输入层。
⸻
我其实还想补充一个我觉得更有价值的观察。
我觉得 OpenAI 想把「聊天」和「执行」融合成一种交互。
但对于知识工作来说，这两件事天然就是两个阶段。
聊天的时候，我是在探索问题。
提交的时候，我是在承担决策。
中间那个「确认」的动作，我目前还不愿意省掉。
OpenAI@OpenAI
ChatGPT Voice is now in the desktop app. Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice. It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time. Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.
中文
把常用的一套视觉方法，做成了一个开源 Skill：
给它一句话、一个主题、一种情绪，或者一张参考图，它会直接生成一张安静、克制、带纸张与印刷质感的独立杂志风海报。欢迎大家使用👏
开源地址：
github.com/LiamGvchi/gc-m…
中文
AI 圈是真的爱造词，又冒出来一个 Graph Engineering。
研究了一下，发现它和我前段时间做的项目很像。
一开始，我只是想做一个可视化的 Agent 编排器。后来不断加入角色、知识库、审核、返工和交付，最后莫名其妙进化成了一个：
“傻瓜老板工作指挥模拟器”，哈哈哈。😂
Graph Engineering 的核心，其实不只是画一张节点图，而是把复杂工作变成一张可以真正执行的工作图：
1 从模板开始
复用的不只是 Prompt，而是一整套任务结构和协作方式。
2 选择 Agent、角色和专业知识
每个节点不只是换一个名字，而是有自己的职责、方法、知识和交付标准。
3 安排依赖、并行和汇总
谁先做、谁等待、哪些职位可以同时工作，以及最后由谁整合结果。
4 设置审核、返工和人工拍板
不合格就退回对应步骤局部修改，高风险决策必须由人批准。
5 定义真正的交付点
最终拿到的不应该是一堆 Agent 日志，而是报告、方案、网页或者可以直接交给客户的结果。
所以我的产品底层是一张 Graph，表面上却只是老板在安排工作：定目标、选职位、排顺序、看审核、拿结果。
莫名其妙的契合了 看来这几天给他搞完 上线让大家也玩一玩
中文
