WEBVTT

00:00:00.000 --> 00:00:04.600
OpenAI發布全新的旗艦模型GPT 5.6Soul,Codex也立刻用上了。

00:00:04.940 --> 00:00:06.360
这次真正值得注意的是什么?

00:00:06.820 --> 00:00:14.240
其实除了5.6Soul到底有多好用,这支影片还要告诉你OpenAI怎么刚把Codex大改一轮,两个月内出了超过150项更新。

00:00:14.880 --> 00:00:23.260
这次真正的转变,是Codex开始自己动手,自己会把一个大任务拆给好几个分身平行座,自己开浏览器点画面,甚至自己管理自己的任务清单。

00:00:23.820 --> 00:00:28.140
Codex还是那个你叫它写程式的工具,但也已经变成一个会自己安排工作的执行者。

00:00:28.140 --> 00:00:30.820
我挑出六件真的会改变你工作方式的功能

00:00:30.820 --> 00:00:32.420
还有一个官方只讲了一句

00:00:32.420 --> 00:00:33.800
但你一定会付的代价

00:00:33.800 --> 00:00:36.380
先讲一个最容易被跳过的前提

00:00:36.380 --> 00:00:38.680
Codex现在住在ChatGPT里面

00:00:38.680 --> 00:00:40.420
官方更新记录写得很清楚

00:00:40.420 --> 00:00:41.320
7月9号

00:00:41.320 --> 00:00:43.540
Codex正式并进ChatGPT桌面板

00:00:43.540 --> 00:00:44.740
Mac和Windows都有

00:00:44.740 --> 00:00:45.800
原本的专案

00:00:45.800 --> 00:00:47.360
设定和工作流都会保留

00:00:47.360 --> 00:00:48.980
它在里面有自己的空间

00:00:48.980 --> 00:00:50.580
旁边是一个新的Work Agent

00:00:50.580 --> 00:00:52.120
这看起来只是搬家

00:00:52.120 --> 00:00:54.020
却决定了后面所有功能的长相

00:00:54.020 --> 00:00:56.440
Codex不再是给工程师的独立产品线

00:00:56.440 --> 00:00:58.500
而是ChatGPT底下的一个工作面向

00:00:58.500 --> 00:01:00.760
官方说每周有600万人在用

00:01:00.760 --> 00:01:02.140
这是他们自己给的数字

00:01:02.140 --> 00:01:03.720
目前没有公开资料可以佐证

00:01:03.720 --> 00:01:06.160
第一件事是脑子换了

00:01:06.160 --> 00:01:08.860
新的旗舰模型叫GPT 5.6 Soul

00:01:08.860 --> 00:01:11.020
官方说它可以长时间啃难题

00:01:11.020 --> 00:01:13.620
但真正要注意的是它旁边那个Ultra

00:01:13.620 --> 00:01:15.060
在官方文件里

00:01:15.060 --> 00:01:17.180
Ultra是推理等级选单的最高一档

00:01:17.180 --> 00:01:18.380
说明只有一行

00:01:18.380 --> 00:01:19.760
最大推理力度

00:01:19.760 --> 00:01:21.140
加上自动分排任务

00:01:21.140 --> 00:01:22.700
这行字很关键

00:01:22.700 --> 00:01:24.580
因为它把两件事绑在一起了

00:01:24.580 --> 00:01:26.200
Ultra不只是想得更久

00:01:26.200 --> 00:01:27.920
他还会主动把工作拆出去

00:01:27.920 --> 00:01:29.200
所以你会看到

00:01:29.200 --> 00:01:30.380
Codex不用你开口

00:01:30.380 --> 00:01:31.420
就自己把一个任务

00:01:31.420 --> 00:01:32.980
分给好几个subagents平行做

00:01:32.980 --> 00:01:34.340
你可以个别追踪

00:01:34.340 --> 00:01:35.320
也可以打开面板

00:01:35.320 --> 00:01:36.240
看整组一起工作

00:01:36.240 --> 00:01:38.580
这里就要讲那个代价了

00:01:38.580 --> 00:01:40.900
更大的思考预算不是免费的

00:01:40.900 --> 00:01:41.820
想得越久

00:01:41.820 --> 00:01:42.640
拆得越多

00:01:42.640 --> 00:01:43.920
Token就烧得越快

00:01:43.920 --> 00:01:45.240
这不是我推测

00:01:45.240 --> 00:01:46.660
官方文件自己写

00:01:46.660 --> 00:01:48.160
subagent的工作流

00:01:48.160 --> 00:01:50.000
会比单一agent消耗更多token

00:01:50.000 --> 00:01:51.880
更新记录里甚至加了一条

00:01:51.880 --> 00:01:52.920
选ultra

00:01:52.920 --> 00:01:54.640
又让多个agent同时跑的时候

00:01:54.640 --> 00:01:56.900
会跳出警告提醒你用量会冲很快

00:01:56.900 --> 00:01:57.880
这一点

00:01:57.880 --> 00:02:00.100
OpenAI在示范的时候自己也讲白了

00:02:24.640 --> 00:02:26.080
开启这 panel

00:02:26.080 --> 00:02:27.760
然后看到整个团队一起工作

00:02:27.760 --> 00:02:29.300
一项目

00:02:29.300 --> 00:02:31.200
然后现在在这一项目移动

00:02:31.200 --> 00:02:32.120
第二件事

00:02:32.120 --> 00:02:33.720
是他开始自己动画鼠

00:02:33.720 --> 00:02:35.200
影片中示范的

00:02:35.200 --> 00:02:36.740
OpenAI开发者体验负责人

00:02:36.740 --> 00:02:37.560
Romaine Hute说

00:02:37.560 --> 00:02:39.140
这可能是整轮更新里

00:02:39.140 --> 00:02:40.820
他个人最期待大家体验的功能

00:02:40.820 --> 00:02:42.500
关键字叫AppShot

00:02:42.500 --> 00:02:44.100
你可以先把它想成截图

00:02:44.100 --> 00:02:45.740
但它比截图多一层东西

00:02:45.740 --> 00:02:48.040
AppShot会把最前面那个视窗的画面

00:02:48.040 --> 00:02:49.680
加上那个App能提供的文字

00:02:49.680 --> 00:02:50.400
一起送出去

00:02:50.400 --> 00:02:51.920
连你卷动范围之外

00:02:51.920 --> 00:02:53.320
萤幕上根本看不到的文字

00:02:53.320 --> 00:02:53.860
都拿得到

00:02:53.860 --> 00:02:55.440
所以这里的差别是

00:02:55.440 --> 00:02:56.540
截图只有像素

00:02:56.540 --> 00:02:58.560
AppShot多了一层结构化的文字

00:02:58.560 --> 00:03:00.220
它不只看到画面长什么样

00:03:00.220 --> 00:03:01.760
还读得到画面背后的内容

00:03:01.760 --> 00:03:03.280
怎么触发

00:03:03.280 --> 00:03:05.840
在Mac上按下左右两个command键就好

00:03:05.840 --> 00:03:07.920
而且只有MacOS桌面板支援

00:03:07.920 --> 00:03:09.680
这点要翻官方文件才查得到

00:03:09.680 --> 00:03:11.380
拿到AppShot之后

00:03:11.380 --> 00:03:13.200
Codex就能自己操作那个模拟器

00:03:13.200 --> 00:03:14.880
画面上游标自己在移动

00:03:14.880 --> 00:03:15.880
自己点来点去

00:03:15.880 --> 00:03:17.040
没有人在碰滑鼠

00:03:17.040 --> 00:03:18.340
更关键的是

00:03:18.340 --> 00:03:19.760
这一切都在背景跑

00:03:19.760 --> 00:03:21.600
你可以继续用你的电脑做别的事

00:03:21.600 --> 00:03:24.200
过去AI操作电脑最恼人的

00:03:24.200 --> 00:03:25.680
就是它会占住你的滑鼠键盘

00:03:25.680 --> 00:03:26.940
这一点被截掉了

00:03:51.600 --> 00:03:52.600
看看这个

00:03:52.600 --> 00:03:53.840
这是我的拇指

00:03:53.840 --> 00:03:55.000
是Codex在转向

00:03:55.000 --> 00:03:57.040
在背景中

00:03:57.040 --> 00:03:59.360
我可以继续用我的手机

00:03:59.360 --> 00:04:00.440
那不错

00:04:21.600 --> 00:04:23.680
这解掉的是一个很真实的痛点

00:04:23.680 --> 00:04:25.080
你看得出哪里怪

00:04:25.080 --> 00:04:26.480
但你讲不清楚它在哪

00:04:26.480 --> 00:04:29.040
另一半是给看得懂程式码的人

00:04:29.040 --> 00:04:30.340
切到DIF检视

00:04:30.340 --> 00:04:31.820
如果你知道自己要什么

00:04:31.820 --> 00:04:33.260
可以直接精准地改一行

00:04:33.260 --> 00:04:35.040
如果只知道想要什么行为

00:04:35.040 --> 00:04:36.000
不确定该怎么改

00:04:36.000 --> 00:04:37.340
就把它丢回给Codex

00:04:37.340 --> 00:04:38.760
接下来这段

00:04:38.760 --> 00:04:40.080
看他怎么在同一个地方

00:04:40.080 --> 00:04:41.340
无缝切换这两种做法

00:04:41.340 --> 00:04:44.020
I can also switch into the DIF view

00:04:44.020 --> 00:04:46.480
I've always been able to inspect the DIF here

00:04:46.480 --> 00:04:48.480
But now I can go further

00:04:48.480 --> 00:04:50.940
and make precise edits directly in line

00:04:50.940 --> 00:04:53.380
所以如果我知道我 exactly what I want

00:04:53.380 --> 00:04:54.640
我可以改变的练习

00:04:54.640 --> 00:04:55.180
我自己的练习

00:04:55.180 --> 00:04:55.820
就像这样

00:04:55.820 --> 00:04:58.480
或如果我知道我希望的行为

00:04:58.480 --> 00:04:59.340
但我不确定

00:04:59.340 --> 00:05:00.740
那种练习的练习

00:05:00.740 --> 00:05:01.740
我可以从那里来求

00:05:01.740 --> 00:05:03.040
Codex工作

00:05:03.040 --> 00:05:04.700
有时候我会工作

00:05:04.700 --> 00:05:05.040
视频

00:05:05.040 --> 00:05:06.540
有时候我会工作

00:05:06.540 --> 00:05:07.780
直接进练习

00:05:07.780 --> 00:05:09.240
然后它会比较近

00:05:09.240 --> 00:05:11.040
跟模式的练习

00:05:11.040 --> 00:05:12.160
第四件事

00:05:12.160 --> 00:05:13.460
是东西做完了

00:05:13.460 --> 00:05:14.740
要怎么让别人用到

00:05:14.740 --> 00:05:16.400
这是Size要解的问题

00:05:16.400 --> 00:05:17.820
你只要说一句

00:05:17.820 --> 00:05:19.240
把这个发布到Size

00:05:19.240 --> 00:05:20.320
它就上线了

00:05:20.320 --> 00:05:22.880
不过这里要修正一句话

00:05:22.880 --> 00:05:25.440
OpenAI说Sites上个月只给企业版

00:05:25.440 --> 00:05:26.540
现在开放所有人

00:05:26.540 --> 00:05:28.740
但官方文件上写的是Public Beta

00:05:28.740 --> 00:05:30.360
而且明讲能不能用

00:05:30.360 --> 00:05:31.160
要看你的方案

00:05:31.160 --> 00:05:32.680
你的地区和工作区设定

00:05:32.680 --> 00:05:34.020
这两个说法对不上

00:05:34.020 --> 00:05:35.600
所以先别预设你打开就用

00:05:35.600 --> 00:05:37.800
Public Beta这四个字本身就是提醒

00:05:37.800 --> 00:05:38.880
换句话说

00:05:38.880 --> 00:05:39.960
拿它做个原型

00:05:39.960 --> 00:05:41.200
做个内部小工具很好

00:05:41.200 --> 00:05:44.040
但要不要把已经有人在用的正式东西搬上去

00:05:44.040 --> 00:05:45.140
现阶段还不好说

00:05:45.140 --> 00:05:51.140
昨天月份,我们介绍了室室,所以 teams可以建设室,和分享它们的内容。

00:05:51.140 --> 00:05:57.140
现在室室是全室,而它仍然是任何网址你建设室。

00:05:57.140 --> 00:06:01.140
我可以直接说,导致室室。

00:06:01.140 --> 00:06:08.140
它有室内,链接,链接,运动,运动,运动,运动,都在建设室。

00:06:08.140 --> 00:06:11.140
所以我可以从一个真正的佛工程中,

00:06:11.140 --> 00:06:14.440
无论论均匀或训练

00:06:14.440 --> 00:06:16.320
然后现在的标志

00:06:16.320 --> 00:06:17.400
是现在的标志

00:06:41.140 --> 00:06:42.360
都有自己独立的工作副本

00:06:42.360 --> 00:06:43.460
改动互不干扰

00:06:43.460 --> 00:06:45.140
你不用担心A的修改

00:06:45.140 --> 00:06:45.980
会不小心动到B

00:06:45.980 --> 00:06:47.480
而且这些任务之间

00:06:47.480 --> 00:06:48.320
还可以互相引用

00:06:48.320 --> 00:06:50.160
注意这里发生的是

00:06:50.160 --> 00:06:51.120
拆解

00:06:51.120 --> 00:06:52.220
排优先序

00:06:52.220 --> 00:06:53.200
开分支

00:06:53.200 --> 00:06:54.980
这些原本是人的工作

00:06:54.980 --> 00:06:56.340
现在被交出去了

00:06:56.340 --> 00:06:58.580
这也是为什么真正的主线

00:06:58.580 --> 00:06:59.760
不是模型变强

00:06:59.760 --> 00:07:01.440
而是角色变了

00:07:01.440 --> 00:07:03.880
另外还有一个很实用的细节

00:07:03.880 --> 00:07:06.040
工作不一定要从Codex开始

00:07:06.040 --> 00:07:08.660
你在ChatGPT里讨论过的想法

00:07:08.660 --> 00:07:10.620
或用Deep Research查过的东西

00:07:10.620 --> 00:07:12.480
可以直接at那段对话

00:07:12.480 --> 00:07:14.740
把整包context带进codex

00:07:14.740 --> 00:07:16.280
不用再复制贴上

00:07:16.280 --> 00:07:20.320
You know that codex can also help you manage its own codex tasks

00:07:20.320 --> 00:07:23.960
This is super useful when I have several independent pieces of work

00:07:23.960 --> 00:07:25.640
that I want to get organized and moving

00:07:25.640 --> 00:07:27.420
Check out this one

00:07:27.420 --> 00:07:29.900
Find 5 bugs from linear

00:07:29.900 --> 00:07:33.580
Create a separate task and isolated work tree for each one

00:07:33.580 --> 00:07:36.040
And pin the most critical ones

00:07:36.537 --> 00:07:51.537
我需要有幾個獨立的工作, 要把工作的組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織組織

00:07:51.537 --> 00:07:52.777
手机版补了一批升级

00:07:52.777 --> 00:07:54.437
在手机上管理任务

00:07:54.437 --> 00:07:55.297
审查变更

00:07:55.297 --> 00:07:56.077
比较branch

00:07:56.077 --> 00:07:57.697
接下来这段一次演完

00:07:57.697 --> 00:07:59.697
我只补一句最反直觉的

00:07:59.697 --> 00:08:01.977
手机可以直接连上SSH Host

00:08:01.977 --> 00:08:03.077
像你的Devbox

00:08:03.077 --> 00:08:04.757
从手机就开新任务

00:08:04.757 --> 00:08:06.177
方便归方便

00:08:06.177 --> 00:08:07.337
权限要开多大

00:08:07.337 --> 00:08:08.577
值得你自己先想清楚

00:08:08.577 --> 00:08:10.257
有三个东西

00:08:10.257 --> 00:08:10.897
我希望你尝试

00:08:10.897 --> 00:08:12.257
第一

00:08:12.257 --> 00:08:13.337
你可以选择

00:08:13.337 --> 00:08:14.097
搜索

00:08:14.097 --> 00:08:15.017
开放

00:08:15.017 --> 00:08:16.057
管理任务

00:08:16.057 --> 00:08:17.537
直接从一个谈论

00:08:17.537 --> 00:08:18.497
上你的电话

00:08:18.497 --> 00:08:19.837
第二

00:08:19.837 --> 00:08:22.837
第二,计划的改变更好

00:08:22.837 --> 00:08:26.837
我可以选择和选择和选择

00:08:26.837 --> 00:08:29.837
或直接交叉择和选择

00:08:29.837 --> 00:08:32.837
第三,我可以直接接到一个 SSH 主持人

00:08:32.837 --> 00:08:33.837
像一个电视箱

00:08:33.837 --> 00:08:36.837
和从电视箱开始新的工作

00:08:36.837 --> 00:08:37.837
有很多多

00:08:37.837 --> 00:08:40.837
但我爱上我可以在电视箱

00:08:40.837 --> 00:08:42.837
从电视箱开始,

00:08:42.837 --> 00:08:44.837
然后在电视箱里,

00:08:44.837 --> 00:08:45.837
这不错吗?

00:08:45.837 --> 00:08:48.837
最后一块是把整条开发流程收完

00:08:48.837 --> 00:08:51.657
Poor Request现在不用离开App就能处理

00:08:51.657 --> 00:08:53.317
而且是带着脉络的

00:08:53.317 --> 00:08:55.817
Codex知道每个任务跑在哪个Branch上

00:08:55.817 --> 00:08:57.597
那个Branch一有开着的PR

00:08:57.597 --> 00:08:59.477
摘要面板就自动把它浮出来

00:08:59.477 --> 00:09:01.217
你可以在里面追踪

00:09:01.217 --> 00:09:02.017
处理问题

00:09:02.017 --> 00:09:03.397
确认没事就merge

00:09:03.397 --> 00:09:05.057
检查失败的时候

00:09:05.057 --> 00:09:07.137
直接请它带着完整脉络去修

00:09:07.137 --> 00:09:09.557
新的Poor Request分页

00:09:09.557 --> 00:09:11.177
可以逐项看每一处改动

00:09:11.177 --> 00:09:12.377
留inline comment

00:09:12.377 --> 00:09:13.517
做完整的review

00:09:13.517 --> 00:09:15.677
也可以把自己的PR标成ready

00:09:15.677 --> 00:09:16.377
然后merge

00:09:16.377 --> 00:09:18.357
这代表从写程式

00:09:18.357 --> 00:09:19.177
跑检查

00:09:19.177 --> 00:09:20.317
review到merge

00:09:20.317 --> 00:09:22.217
整条loop都收进同一个地方了

00:09:22.217 --> 00:09:23.717
你不用再跳去github.com

00:09:23.717 --> 00:09:26.017
但这里有个问题值得放在心上

00:09:26.017 --> 00:09:27.157
当写的是它

00:09:27.157 --> 00:09:28.097
修的是它

00:09:28.097 --> 00:09:29.597
review的界面也是它

00:09:29.597 --> 00:09:31.277
那谁在为这次合并负责

00:09:31.277 --> 00:09:33.177
工具把摩擦拿掉了

00:09:33.177 --> 00:09:34.397
责任不会跟着被拿掉

00:09:34.397 --> 00:09:37.317
另外这段示范从头到尾只提到github.com

00:09:37.317 --> 00:09:39.217
其他github平台有没有同样待遇

00:09:39.217 --> 00:09:40.317
就得自己去确认了

00:09:40.317 --> 00:09:43.837
Codex now makes it easier to work with pull requests

00:09:43.837 --> 00:09:45.177
without leaving the app

00:09:45.177 --> 00:09:46.797
It does this in context

00:09:46.797 --> 00:09:48.797
它知道哪一項 branch

00:09:48.797 --> 00:09:50.237
每項 task是在工作

00:09:50.237 --> 00:09:51.437
所以當那項 branch

00:09:51.437 --> 00:09:52.237
有開啟的研究

00:09:52.237 --> 00:09:53.237
它會有一個項目的研究

00:09:53.237 --> 00:09:54.797
它會有一個項目的

00:09:54.797 --> 00:09:56.357
我可以追求所有的事情

00:09:56.357 --> 00:09:57.957
和回應所有問題

00:09:57.957 --> 00:09:58.957
和回應所有事情

00:09:58.957 --> 00:10:00.157
都會看得到

00:10:00.157 --> 00:10:01.957
如果一個項目失去

00:10:01.957 --> 00:10:03.037
或是一個項目的

00:10:03.037 --> 00:10:04.357
我可以請 Codex

00:10:04.357 --> 00:10:05.357
去解決它

00:10:05.357 --> 00:10:05.757
在這裡

00:10:05.757 --> 00:10:06.757
全體的概念

00:10:06.757 --> 00:10:07.757
這新的項目

00:10:07.757 --> 00:10:08.997
提供我

00:10:08.997 --> 00:10:10.957
讓我每個個改變

00:10:10.957 --> 00:10:12.197
我可以留在內

00:10:12.197 --> 00:10:13.157
可項目

00:10:13.157 --> 00:10:14.837
complete a full review

00:10:14.837 --> 00:10:15.957
我可以尋找我的

00:10:15.957 --> 00:10:18.277
P.R.s as ready and merge them

00:10:18.277 --> 00:10:20.217
All of this within Codex

00:10:20.217 --> 00:10:22.537
Without having to go to github.com

00:10:22.537 --> 00:10:24.277
So I can stay in the flow

00:10:24.277 --> 00:10:25.577
回到开头那个问题

00:10:25.577 --> 00:10:27.337
把这六件事排在一起看

00:10:27.337 --> 00:10:28.537
方向其实很清楚

00:10:28.537 --> 00:10:29.597
拆工作

00:10:29.597 --> 00:10:30.517
开分支

00:10:30.517 --> 00:10:31.637
操作画面

00:10:31.637 --> 00:10:32.557
管任务

00:10:32.557 --> 00:10:33.557
处理P.R.

00:10:34.097 --> 00:10:35.517
这些原本都是人的判断

00:10:35.517 --> 00:10:37.357
现在一件一件被交出去了

00:10:37.357 --> 00:10:38.437
所以

00:10:38.437 --> 00:10:39.977
Codex已经不只是那个

00:10:39.977 --> 00:10:41.257
你叫它写程式的工具

00:10:41.257 --> 00:10:42.457
它更像一个

00:10:42.457 --> 00:10:44.097
会自己安排工作的执行者

00:10:44.097 --> 00:10:45.657
至于代价

00:10:45.657 --> 00:10:46.917
官方只讲了一句

00:10:46.917 --> 00:10:49.097
Ultra会更快吃掉你的token

00:10:49.097 --> 00:10:51.497
而真正需要你自己决定的

00:10:51.497 --> 00:10:53.077
是把多少判断交出去

00:10:53.077 --> 00:10:54.597
以及交出去之后

00:10:54.597 --> 00:10:55.657
谁还在看
