OpenAI Pairs Persistent Dots With GPT-Live to Guide Real-World Tasks

Cooking with Codex: Ep. 1 — OpenAI

OpenAI

Persistent background agents combined with full-duplex audio models can direct unstructured physical workflows and compile custom hardware software in real time without conversational interruption.

AI-assisted summary of Cooking with Codex: Ep. 1 — original source. Check the original for full context.

Source published:

Key ideas

  • OpenAI introduced 'Dots', always-on persistent background agents designed to handle daily coordination like monitoring emails and organizing workflows in ChatGPT Spaces. (0:00)
  • Codex Cloud enables developers to run persistent environments in the cloud that keep tasks executing even after closing their laptop. (0:00)
  • The updated Codex CLI features a fullscreen interface, integrated agent management tabs, git worktree forking, and direct terminal voice input. (1:16)
  • GPT-Live functions as a frontier full-duplex audio model capable of understanding natural human cadence, conversational pauses, and fillers without interrupting users. (1:16)
  • Forward Future developers demonstrated Ultrafast to one-shot code an 8-bit ramen simulation game and deploy it directly onto a ModRetro handheld console. (2:51)

Detailed notes

## Background Published by OpenAI on YouTube on October 19, 2026, this debut episode of *Cooking with Codex* demonstrates how an OpenAI Dot and ChatGPT Spaces can organize and plan a ramen recipe, guided live by a GPT-Live sous chef. The video also covers Codex cloud environments, updated Codex CLI features, an 8-bit ModRetro ramen game build with Forward Future, and a community showcase of standout user projects.

## TL;DR Hosts Danny and Alex navigate the kitchen with the help of DevDay ecosystem tools, utilizing an always-on Dot agent to structure prep in ChatGPT Spaces and a full-duplex GPT-Live voice assistant to direct their ramen cooking. Forward Future guests Matthew and Brian take fresh ginger shots before one-shot coding an 8-bit ramen simulation game with Ultrafast that runs directly on a ModRetro handheld. Fed from the community team steps in to judge viral user-built projects in a physical paddle contest called 'Cooked or Undercooked?' before the group playtests their custom handheld game and sits down to taste their surprisingly delicious real-life ramen.

## Atmosphere & style The episode delivers a breezy, irreverent blend of Saturday-morning variety show energy, tech keynote demos, and unscripted kitchen chaos. Banter between the hosts and their deadpan, sarcastic AI sous chef sets a lighthearted tone that keeps technical deep dives engaging and fun. With ginger shot challenges, nostalgic pixel-art game design, and physical cooking happening in parallel, the broadcast feels playful, inventive, and genuinely collaborative.

## Bulletin - [00:00](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=0s) OpenAI introduced 'Dots', always-on persistent background agents designed to handle daily coordination like monitoring emails and organizing workflows in ChatGPT Spaces. - [00:00](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=0s) Codex Cloud enables developers to run persistent environments in the cloud that keep tasks executing even after closing their laptop. - [01:16](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=76s) The updated Codex CLI features a fullscreen interface, integrated agent management tabs, git worktree forking, and direct terminal voice input. - [01:16](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=76s) GPT-Live functions as a frontier full-duplex audio model capable of understanding natural human cadence, conversational pauses, and fillers without interrupting users. - [02:51](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=171s) Forward Future developers demonstrated Ultrafast to one-shot code an 8-bit ramen simulation game and deploy it directly onto a ModRetro handheld console.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=0s) OpenAI Dots serve as persistent agents that run in the background to handle scheduling, guest check-ins, emails, and workflow planning. - [00:00](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=0s) ChatGPT Spaces provides a dedicated shared canvas where Dot agents can write out structured plans and collaborate side-by-side with humans. - [00:00](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=0s) Codex Cloud allows customizable team development environments to run persistently in the cloud independently of local machine power. - [01:16](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=76s) The refreshed Codex CLI integrates direct terminal voice input and git worktree forking to split conversations across multiple code trees. - [01:16](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=76s) GPT-Live provides full-duplex conversational audio assistance, recognizing conversational fillers like 'um' and waiting its turn rather than talking over users. - [01:16](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=76s) Visual identification enabled the AI sous chef to distinguish kitchen seasonings, verifying unlabeled chili powder instead of paprika. - [02:51](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=171s) OpenAI Dot demonstrated proactive alerting by surfacing an urgent late-night call sheet notification to Forward Future before filming. - [02:51](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=171s) Developers paired Ultrafast with ModRetro hardware to rapidly build and deploy an 8-bit kitchen simulation game featuring Dot avatars and cooking mechanics. - [05:12](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=312s) Community showcase spotlighted a historical cipher breakthrough by Carter Church using Astra to decipher a centuries-old encrypted letter from Napoleon Bonaparte. - [05:12](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=312s) Community projects also featured a procedural horse-racing engine translating hand-drawn sketches into real-time running speeds and a retro space shooter built with GPT-6.1. - [05:12](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=312s) Carter Church's Napoleon dispatch project was awarded 'Catch of the Day' among evaluated community releases. - [07:31](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=451s) The hosts and guests validated that the AI-guided ramen recipe produced an authentic, successful dish alongside the digital game prototype.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=0s) Introducing *Cooking with Codex* and the DevDay Ecosystem

Danny and Alex kick off the debut episode of *Cooked* (or *Cooking with Codex*) from OpenAI with an irreverent kitchen joke and a singing intro. Welcoming viewers, Danny announces a packed agenda inspired by recent DevDay releases, including OpenAI Dots, ChatGPT Spaces, and cloud-based developer tools. Alex explains how they planned their dish: he snapped photos of all the raw ingredients in the kitchen and handed them over to his personal Dot agent to construct the recipe and plan execution.

```mermaid flowchart TD accTitle: OpenAI Dot and DevDay Workflow accDescr: Diagram illustrating how ingredients, Dots, Spaces, and Codex Cloud integrate into the cooking workflow. subgraph Inputs["Inputs & Environment"] Ing["Ingredient Photos"] CC["Codex Cloud<br/>(Persistent, Laptop Closed)"] end subgraph Agent["Dot Agent Core"] Dot["Always-on Persistent Dot<br/>(Interference, Emails, Guest Check-in)"] end subgraph Collaboration["Output & Coordination"] Space["ChatGPT Space<br/>(Collaborative Recipe & Steps)"] end Ing --> Dot CC -.-> Dot Dot --> Space ```

The hosts dive deeper into what these tools actually do in practice: - **OpenAI Dots**: Described as an “always on persistent agent,” running in the background to handle daily interference—such as managing guest check-ins, monitoring emails, and organizing the cooking workflow. - **Codex Cloud**: A major launch from DevDay designed to let developers run environments in the cloud that teams can configure and customize, allowing tasks to keep executing even when you close your laptop. - **ChatGPT Spaces**: A shared workspace providing the Dot a dedicated canvas to organize its internal thoughts, lay out detailed steps, and collaborate side-by-side with human users.

Alex highlights the synergy between agent and workspace, stating: “Spaces gives your dot a place to actually write its thoughts and a place for you to collaborate with your dot.”

---

[01:16](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=76s) Codex CLI Upgrades and Cooking with GPT-Live Sous Chef

Before proceeding, the hosts present an ad spot showcasing the newly refreshed Codex CLI, which now features a fullscreen interface, integrated agent management tabs, git worktree forking to split conversations into separate code trees, and direct terminal voice input. Back in the kitchen, the cooking begins with an AI sous chef driven by GPT-Live. The AI immediately ribs the hosts, labeling them the “chaos twins” and quipping that it is a kitchen where they only “pretend they know where the ginger is.” When Danny objects to the burn, the voice assistant deadpans: “It's nicer than your knife skills, but not by much.”

The AI walks them through the initial ramen prep steps, identifying decanted seasonings via vision—confirming an unlabeled red powder is chili, not paprika. It orders them to simmer eggs for 11 to 12 minutes before transferring them to an ice bath, and to chop two cloves of garlic to warm gently in oil with ginger without any browning. Alex points out the underlying technology: their sous chef runs on GPT-Live, a frontier full-duplex audio model capable of understanding natural human cadence, conversational pauses, and fillers like "um" without rudely interrupting. Demonstrating the behavior, the sous chef responds: “Yeah, I was waiting my turn. Gold star for not talking over you. Okay, back to the stove.”

---

[02:51](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=171s) Forward Future Joins: The Ginger Shot Challenge and 8-Bit ModRetro Build

Matthew and Brian from Forward Future join the kitchen. Asked about their practical Dot usage since its rollout, Brian recounts how his Dot saved him the night before shooting: while he was already in bed, the agent spotted an urgent email outlining call sheet instructions and surfaced a proactive notification summarizing what they needed to know for filming. Danny and Alex then lay down a community challenge: Brian and Matthew must take “one-shot” liquid shots made from kitchen-pressed fresh ginger, and subsequently one-shot code an 8-bit ramen simulation game using Ultrafast to deploy directly onto a ModRetro handheld console.

```mermaid flowchart LR accTitle: Ultrafast to ModRetro Game Loop accDescr: Step-by-step game loop generated for the ModRetro 8-bit handheld. Sel["Character Select<br/>(Choose Dot)"] --> Frid["Fridge Phase<br/>(Gather Ingredients)"] Frid --> Pot["Stove Top Phase<br/>(Simmer & Clockwise Stir)"] Pot --> Score["Ramen Evaluation<br/>(Texture & Flavor Score)"] ```

Brian and Matthew take the ginger shots and immediately jump into the build: - The game design incorporates an 8-bit character select screen where the player picks an avatar styled as an OpenAI Dot. - Gameplay mechanics focus on kitchen exploration: moving to the fridge, gathering broth, fresh noodles, chashu, and a soft-boiled egg, and cooking at the stovetop. - The player interacts through dialogue boxes, choosing between a gentle simmer or a rolling boil and stirring clockwise to control broth texture.

While Matthew chops real garlic on the kitchen counter, the team runs the game on an emulator, prompting Brian to reflect on the aesthetic juxtaposition: “This ModRetro and like pixelization presentation, it makes me feel both that I'm in the future and the past.”

---

[05:12](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=312s) Community Showcase: "Cooked or Undercooked?"

Fed from the OpenAI Developer Community team drops in to replace Danny for a review segment assessing recent user builds shared across X (formerly Twitter), LinkedIn, and Instagram. The hosts wield physical paddle cards to judge whether each featured community project is “Cooked” (ready, polished, impressive) or “Undercooked.”

| Project Name / Builder | Description & Highlights | Verdict | |:--- |:--- |:--- | | **Retro Space Shooter** (Cosmin) | A fast-paced arcade retro space shooter programmed using GPT-6.1. Praised for its tight gameplay and authentic retro visual presentation. | **Cooked** | | **Napoleon Decoded Dispatch** (Carter Church) | Built using Astra to decipher a complex historical ciphertext letter from Napoleon Bonaparte that had remained unsolved for hundreds of years. Amassed nearly 5 million views. | **Cooked** | | **Hand-Drawn Horse Racing** | A viral physics-based racing game where players sketch an original horse silhouette on screen; the procedural engine translates the sketch’s specific anatomy into dynamic running speeds. | **Double Cooked** |

Assessing the historical cipher breakthrough, Brian enthusiastically drops his paddle, declaring: “Napoleon could have never predicted this. This is ridiculous. You did great. Cooked.” For the finale, the hosts vote on their personal "Catch of the Day." While Alex roots for the retro shooter and Fed votes for the horse racing project, Carter Church's Napoleon decryptor takes the win.

---

[07:31](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=451s) Handheld Playtesting and Plating the Ramen

With the compilation finished, Matthew loads the newly generated ROM onto the ModRetro hardware. Navigating his blue Dot sprite across the handheld's LCD screen, he opens the virtual fridge, picks up the eggs, mushrooms, and noodles, brings the stock to a rolling boil, and stirs clockwise. The game evaluates the resulting bowl as having silky broth and tender noodles—surpassing their previous attempt on the emulator.

Smelling the food cooking on the physical stove, Alex and Danny reveal that despite their banter and distractions with DevDay features, an authentic, beautifully assembled bowl of ramen with rich broth, noodles, chashu, and eggs has indeed been prepared. Danny surveys the counter in triumph, declaring: “We actually cooked with this one, boys.”

[08:34](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=514s) Tasting the Ramen and Wrapping Up

With the cooking wrapped up beside the laptop, the group grabs their chopsticks to finally dig into the finished ramen. The dish earns immediate praise around the table for both its presentation and aroma, prompting one of the hosts to marvel at their culinary success: “I can't believe we cooked that.” When asked for his verdict on the final bowl, Matthew tastes it and laughs, admitting, “It's actually really good.”

As the crew savors the meal, Danny returns to the scene, sparking a quick humorous aside about Freddy's sudden disappearance before they transition into signing off. The hosts give a warm shoutout to their special guests, Matthew and Brian from Forward Future, celebrating an episode spent preparing food, playing retro video games, and building new memories together. The broadcast signs off with gratitude to the viewers before closing out on a playful, melodic outro track about refusing to follow the book: “Always did it my way. Never did it by the book. Let me cook now. Let me cook. Just let me cook.”

- **Tasting Reaction:** Matthew confirmed the freshly prepared ramen turned out surprisingly delicious after all the prep with the AI tools. - **Guest Acknowledgments:** The hosts thanked special guests Matt and Brian from Forward Future for hanging out, gaming, and cooking.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=DQHtmW9ZlIs&t=0s) Spaces gives your dot a place to actually write its thoughts and a place for you to collaborate with your dot. — Alex

## Mentioned OpenAI, Codex, Codex CLI, Codex Cloud, OpenAI Dots, ChatGPT Spaces, GPT-Live, Forward Future, Ultrafast, ModRetro, Danny, Alex, Matthew, Brian, Fed, Cosmin, Carter Church, Napoleon Bonaparte, Astra, GPT-6.1

## Video description In this episode of Cooking with Codex, an OpenAI dot helps plan a ramen recipe and organize prep in ChatGPT Space, with GPT-Live guiding the cooking process.

Also covered: Codex cloud environments, which keep tasks running after a laptop is closed, and new Codex CLI features including voice input, agent management, and forking conversations into separate worktrees.

Matthew and Brian from Forward Future join to build an 8-bit ramen game using Ultrafast and play it on a ModRetro.

Plus, a showcase of community projects, from a retro space shooter to a racing game featuring hand-drawn horses.

[Source](https://www.youtube.com/watch?v=DQHtmW9ZlIs)

中文简报

OpenAI将常驻Dots与GPT-Live结合以指导现实任务 常驻后台智能体与全双工音频模型相结合，无需打断对话即可实时指导非结构化物理工作流并编译定制硬件软件。 主持人Danny和Alex在DevDay生态系统工具的帮助下在厨房忙碌，利用常驻的Dot智能体在ChatGPT Spaces中组织备料，并借助全双工GPT-Live语音助手指导他们煮拉面。来自Forward Future的嘉宾Matthew和Brian喝下鲜姜汁，随后使用Ultrafast一次性编码了一款8位拉面模拟游戏，并直接运行在ModRetro掌机上。社区团队的Fed随后参与了一场名为“Cooked or Undercooked?”的实体举牌评判环节，为爆款用户自制项目打分，之后全员试玩了这款定制掌机游戏，并坐下来品尝了出人意料地美味的真实拉面。 OpenAI推出了“Dots”，这是一种常驻后台的智能体，旨在处理监控电子邮件和在ChatGPT Spaces中组织工作流等日常协调事务。 (0:00) Codex Cloud允许开发者在云端运行持久化环境，即使合上笔记本电脑，任务也能继续执行。 (0:00) 更新后的Codex CLI具备全屏界面、集成智能体管理标签页、git worktree分支以及直接终端语音输入功能。 (1:16) GPT-Live作为前沿全双工音频模型，能够理解人类自然的语调节奏、对话停顿和语气词，而不会打断用户。 (1:16) Forward Future的开发者展示了Ultrafast，一次性编码生成了一款8位拉面模拟游戏，并直接部署到ModRetro掌上游戏机上。 (2:51)

Reading summary

In the debut episode of OpenAI's Cooking with Codex, hosts Danny and Alex showcase practical integrations of recent DevDay developer tools—including OpenAI Dots, ChatGPT Spaces, and Codex Cloud—by simultaneously preparing a ramen dish and coding an 8-bit simulation game. DevDay Ecosystem and Agent Orchestration (0:00) The episode frames cooking as a testbed for integrating OpenAI's latest product ecosystem. Alex demonstrated planning the ramen dish by capturing photos of raw kitchen ingredients and routing them to a personal OpenAI Dot—an always-on, persistent background agent that manages communication, guest check-ins, and execution steps. The agent collaborates with users inside ChatGPT Spaces, providing a dedicated workspace for the model to organize its thoughts and plan workflows. Supporting these agentic workflows in developer environments, Codex Cloud was highlighted as an infrastructure upgrade allowing team-configured environments to continue running even after a developer's laptop is closed. Complementing the cloud setup, updates to the Codex CLI added fullscreen agent tabs, git worktree forking for code trees, and direct terminal voice input. Full-Duplex Audio Sous Chef Guidance (1:16) In the kitchen, the hosts prepared the meal using an AI sous chef running on GPT-Live, a frontier full-duplex audio model. The system demonstrated conversational turn-taking by respecting human pauses and filler words without interrupting. It used computer vision to distinguish unlabeled spices, verifying an ambiguous red powder was chili rather than paprika, and issued specific culinary instructions, including boiling eggs for 11 to 12 minutes before placing them in an ice bath and gently heating minced garlic and ginger without browning. One-Shot Retro Handheld Game Build (2:51) Guests Matthew and Brian from Forward Future joined the kitchen to run a challenge: taking fresh ginger shots while 'one-shot' programming an 8-bit ramen simulation game using Ultrafast to deploy directly onto a ModRetro handheld console. Brian noted his Dot agent had already proven functional by waking him with an urgent overnight briefing on call sheet filming instructions. The resulting 8-bit game featured an OpenAI Dot avatar, kitchen navigation to gather noodles, broth, and toppings, dialogue trees controlling simmer intensity, and clockwise stirring to determine final ramen texture, ultimately compiling into a working ROM tested on both an emulator and ModRetro hardware. Community Showcase and Final Assembly (5:12) A community review segment led by Fed evaluated recent projects built with OpenAI tools. Cosmin's GPT-6.1 retro space shooter and an anatomical hand-drawn horse racing game both earned positive 'Cooked' verdicts, but the hosts selected Carter Church's project as 'Catch of the Day,' which used Astra to decrypt a centuries-old ciphered dispatch from Napoleon Bonaparte. Concurrently, the live kitchen effort concluded successfully, with the group confirming the physically prepared ramen achieved rich broth and proper texture alongside the completed software build. The broadcast demonstrated the convergence of conversational models, vision, and background agents across kitchen tasks and rapid game development, though it operated in an informal, entertainment-focused demonstration environment rather than presenting formal benchmarks.

精炼阅读摘要

OpenAI在其首期开发者节目《Cooking with Codex》中，全面展示了依托DevDay生态构建的实际应用流，将持久化智能体OpenAI Dots、云端环境Codex Cloud以及ChatGPT Spaces协同整合，并借助多模态全双工模型GPT-Live与Ultrafast工具，完成了实体厨房烹饪与掌机游戏的原型开发部署。 持久化智能体与云端协作空间架构 (0:00) 节目深入阐释了DevDay生态中多款核心协同工具的定位与联动机制。OpenAI Dots被定义为一种常驻且持久运行的后台智能体，能够持续接管用户的日常琐务，包括跟进来宾签到、监控邮件信息以及统筹料理流程。Alex展示了其具体实践：通过直接拍摄厨房现有的食材原料照片并提交给个人Dot智能体，由其自动推导菜谱并制定烹饪工序。受邀参与节目的Forward Future团队成员Brian亦证实，其Dot智能体曾在深夜就寝期间自主识别紧急通告单邮件，并主动推送拍摄须知摘要，体现了常驻智能体的主动交互价值。 在底层环境与人机交互面板方面，Codex Cloud允许工程团队在云端灵活配置和自定义开发环境，其核心特征在于用户合上笔记本电脑后，后台任务依然能够保持不间断执行。与此同时，ChatGPT Spaces为智能体提供了专门的思考与记录空间。Alex指出，Spaces为Dot智能体提供了一处记录内心推导脉络的虚拟画布，支持人类用户与智能体在统一面板内并肩协作，从而使底层云环境、后台智能体推理和前端任务呈现打通为一体化工作流。 多模态全双工模型与CLI环境升级 (1:16) 技术工具方面，新版Codex CLI引入了全屏交互界面、集成式智能体管理标签页、用于将不同对话分支拆分为独立代码树的git worktree多工作区管理功能，以及终端直接语音输入支持。在厨房操作中，基于GPT-Live构建的AI副厨承担了实时语音调度任务。GPT-Live属于前沿全双工音频模型，能够精准辨识人类自然的语言节奏、讲话停顿以及如“嗯”等语气填充词，并在对话中保持倾听礼貌，避免不合理地强行打断人类发言。 除了语音交互，该AI副厨还展现了计算机视觉识别能力。在面对未贴标签的分装调料罐时，模型通过视觉精准判定该红色粉末为辣椒粉而非红甜椒粉。随后，系统向操作者发出一系列精细的操作指引，包括将鸡蛋放入水中慢熬11至12分钟后立即置入冰浴冷却，以及将两瓣切碎的大蒜与生姜一同在食用油中轻柔温热且不可产生焦褐痕迹，印证了多模态模型在实时、高并发物理场景下的动态指导可行性。 极速原型开发与ModRetro硬件部署 (2:51) 节目设置了兼具趣味与工程挑战的实操环节：Forward Future团队的Brian和Matthew在饮用鲜榨生姜汁后，接受挑战使用Ultrafast工具单次编写并编译出一款8位拉面模拟游戏，直接部署运行在ModRetro复古掌上游戏机上。在游戏设计架构中，开发团队设定了以OpenAI Dot为外观原型的8位像素角色选择界面，核心游戏循环涵盖厨房探索、前往冰箱搜集高汤底料、鲜面条、叉烧肉与溏心蛋，以及在炉灶前通过对话框选择微沸或滚沸烹煮，并通过顺时针搅拌来控制汤底的质感与粘稠度。 在代码生成与编译完成后，Matthew将生成的ROM固件刷入ModRetro硬件进行真机测试。掌机液晶屏幕上的蓝色Dot精灵顺利穿行于虚拟厨房完成食材拾取与高汤烹制，游戏系统最终裁定该碗拉面具备丝滑高汤与软嫩面条，整体效果甚至超越了前期在模拟器上的测试表现。该实验展示了新一代AI代码生成工具在极短时间内完成从概念架构、逻辑生成到特定嵌入式掌机固件编译部署的跨平台能力。 开发者社区前沿案例评测与破译展示 (5:12) OpenAI开发者社区团队的Fed主持了名为“Cooked or Undercooked”的项目评估环节，集中审视了开发者近期基于OpenAI技术栈构建的代表性作品。其中，开发者Cosmin利用GPT-6.1构建的复古街机太空射击游戏《Retro Space Shooter》凭借紧凑的操作手感与真实的复古视效被评为“Cooked”；另一款走红网络的程序化物理赛马游戏，则能依据玩家在屏幕上随手绘制的马匹解剖轮廓，动态测算并转化为物理引擎驱动的奔跑速度，获得了全场“Double Cooked”的高评价。 在历史密码破译领域，开发者Carter Church运用Astra模型成功解密了法国历史人物拿破仑·波拿巴（Napoleon Bonaparte）遗留数百年未解的复杂密文信件（Napoleon Decoded Dispatch），在社交平台累计获得近500万次浏览。Brian等评审嘉宾一致判定该项目彻底解决了一项历史悬案，并将其推选为本期的“当日最佳项目”（Catch of the Day），展示了前沿模型在历史文献复杂密码分析中的应用潜力。 本期节目通过厨房烹饪与游戏开发交织的场景，直观呈现了OpenAI在后台持久运行智能体、全双工自然语音交互、云端脱机执行及嵌入式固件编译等维度的工具链整合，表明AI正从单向文本生成转向主动参与物理协作与复杂软件工程；但相关演示主要依托特定原型与即兴挑战，这些工具在大规模复杂工程与长期生产环境中的鲁棒性仍有待实践检验。

Source quotes

[0:00] Alex: Spaces gives your dot a place to actually write its thoughts and a place for you to collaborate with your dot.

Original transcript

[0:00] I think you're the only qualified cook [0:01] in this kitchen. [laughter] [0:03] >> Welcome to Cooked from OpenAI. [0:05] >> Always [music] did it my way. Never did [0:07] it by [singing] the book. Let me cook [0:09] now. [0:10] >> I'm Danny. [0:11] >> I'm Alex. [0:12] >> And today we have a lot of things on the [0:13] menu. Dots, catch faces, and many other [0:16] things from Open Deep Day we want to try [0:18] ourselves. [music] [0:21] How confident are we in this recipe? [0:23] >> I think we're pretty confident. I took a [0:25] bunch of pictures of all the ingredients [0:26] and sent them to my dot and I think he [0:29] is on the task. So see how this turns [0:31] out. [0:33] >> Speaking about dots, [0:34] >> dot is your always on persistent agent. [0:37] In [music] this case, our dot has been [0:38] running in the background running [0:40] interference for us for checking in [0:42] guests, sending emails, and of course [0:44] putting [music] this recipe together. [0:48] Now there was another big launch at [0:50] Devday, and that was Codex Cloud. Cutest [0:52] Cloud [music] lets you work inside the [0:54] cloud in your own environment that you [0:56] and your team can fully set up and [0:57] customize so that you can close your [1:00] laptop. [1:02] The other thing I'm really excited about [1:03] is spaces. Spaces gives your dot a place [1:06] to actually write its thoughts [music] [1:07] and a place for you to collaborate with [1:09] your dot. He has everything in a space [1:11] right now that he's preparing for us and [1:13] we're just going to look at the space, [1:14] follow the steps, and see what happens. [1:16] >> Before we continue cooking, a word from [1:18] [music] our sponsors. [1:19] >> Our sponsor [1:21] Introducing the refreshed Codeex CLI. A [1:25] new full screen look. Keep tabs on your [1:27] agents. Fork a conversation into a [1:30] separate work tree. And now talk to your [1:33] terminal. [music] [1:33] >> How do we cook like no site's ever [1:35] cooked before? [1:36] >> I love the energy. Let's invent, not [1:38] imitate. [music] [1:39] >> Update and keep cooking. Please consult [1:41] your existing side projects before [1:42] starting another. [1:43] >> We're back and we have our sue chef here [1:45] with us. Oh, the chaos twins return. [1:49] Welcome back to Cooked, where Alex and [1:51] Danny pretend they know where the ginger [1:53] is. [1:54] >> That's not very nice. [1:55] >> That's not nice at all. [1:56] >> It's nicer than your nice skills, [1:58] [music] but not by much. Let's make this [2:00] recipe. [2:02] >> Hey, can you just walk us through this [2:03] recipe for today? [2:05] >> We don't know how to cook. [2:06] >> Okay, chef. Resetting. Let's identify [2:09] [music] those decanted seasonings. [2:11] >> What's this one? You want to pass it to [2:12] me? I think this is paprika. [2:15] >> All right, I checked out the seasoning. [2:17] That is definitely chili. Okay, [2:19] excellent. Get those eggs simmering [2:21] about 11 to 12 minutes. We'll ice them [2:23] after. Chop two garlic cloves. We'll [2:26] warm garlic and ginger and a touch of [2:27] oil. No browning. [2:29] >> All right. So, our sue chef here is [2:31] powered by GPT Live 1, which is [music] [2:34] our Frontier model for voice. And it is [2:36] a full duplex model, which means that [2:37] when I trail off like this [music] or [2:40] say, um, it actually doesn't interrupt [2:42] me. Right, Jeeoff? [2:44] >> Yeah, I was waiting my turn. Gold star [2:47] for not talking over you. Okay, back to [2:50] the stove. [2:51] >> All right, let's do it. [2:52] >> We'll do it. Matthew, you've been [2:55] following us as we've been cooking here. [2:57] How do you think we're doing? [2:58] >> Yeah, it's pretty good so far. [2:59] Definitely looks like ramen. Thanks. [3:02] >> Thanks. [3:02] >> I think Brian's here somewhere, too, [3:04] right? [3:04] >> Yeah, I'm here. [3:05] >> Oh my god. [3:06] >> Was in the kitchen basement. [3:14] >> So, Dots came out last week. We've been [3:16] [music] using it a ton. We want to hear [3:17] how you guys have been using your dots. [3:19] >> Yeah. I mean, the first thing I did was [3:20] obviously connect it [music] to all the [3:22] different services I use. Last night [3:24] before filming, we got emails of [3:26] instructions. You know, I was in bed and [3:27] it popped up with a message saying, [3:29] "Hey, you just got this email. You [3:31] probably should know about it for your [3:32] film tomorrow." So, it's been great. [3:35] >> Guys, we have a challenge for you. [3:36] >> Have you heard of OneShot? [3:38] >> Absolutely. You guys are going to [3:40] oneshot these shots, which is [3:42] ingredients from the kitchen, fresh [3:43] ginger, and then you are going to build [3:45] a game with ultra fast on the new mod [3:48] retros. Are you ready? [3:50] >> Here we go. [3:50] >> Let's do it. [3:50] >> All right. Cheers, man. [3:52] >> Let's do it. [3:56] >> Yeah. Okay. We're going to make a game. [3:58] >> Okay. We want to make a ramen simulator [4:01] game [music] for the mod retro, and we [4:05] want it to be obviously 8bit. Give me a [4:07] character select screen where I can [4:09] choose any of the dots. [4:11] >> The goal of the game is [music] to [4:13] collect the different ingredients, put [4:15] it all together, and make the best bowl [4:16] of ramen possible. Hey, it's moving. [4:18] Yeah, we're getting there. [4:19] >> You can say we're cooking. [4:23] >> All right, Matt, you want to keep [4:24] chopping the garlic? [4:25] >> Absolutely. [4:25] >> Thank you. Thank you. I think you're the [4:27] only qualified cook in this kitchen. [4:28] >> I I am far from [4:30] >> Actually, I don't know about you, Brian. [4:32] >> No, no, no, no. Don't even ask. [4:33] >> If we're ready, we got it going. We have [4:35] the emulator running here with a [4:38] dialogue option. [4:39] >> Okay. Gather fridge cook top. [music] [4:41] Okay. [4:41] >> Get in front of that fridge. [4:42] >> Come on. [4:43] >> Yes. There we go. [4:44] >> We got the ingredients. [4:45] >> Fresh ingredients. Noodles and [music] [4:47] miso broth. [4:48] >> Soft egg. [4:49] >> Mushroom. Uh, [4:50] >> you got to get that soft egg, man. [4:52] >> I know you [music] went for the chashu. [4:54] [laughter] [4:55] >> For the broth, are we doing a gentle [4:56] boil or a rolling boil? [4:59] >> It's rolling for me. [5:00] >> Okay. Rolling it is. [5:01] >> All right. [5:02] >> Stir clockwise. this mod retro and like [5:05] pixelization presentation. It makes me [5:07] feel both that I'm in the future [music] [5:09] and the past. [5:10] >> A little overdone. Still tasty. [5:12] >> Hey, you're not Danny. [5:14] >> No, I'm Fed. I work at developer [5:15] community at OpenAI. [5:17] >> Of course. So, we're going to do a [5:18] little community section. We've been [5:20] watching all the builds that you guys [5:22] had on Twitter, LinkedIn, and Instagram. [5:24] So, today we're going to bring a few of [5:25] these community builds and [music] see [5:27] what you think about it. [5:28] >> Is it cooked or is it undercooked? [5:30] >> Yeah, we even have pre-made cards for [5:32] it. Yeah, this is [5:32] >> super official. I'm ready. [5:33] >> Don't [music] make me use this. [5:35] >> Betty, what are we looking at here [5:36] today? [5:36] >> So, this is a very cool retro game built [5:38] [music] with GPT 6.1 saw. [5:40] >> That's pretty cool. What do you guys [5:41] think? [5:42] >> You know, I actually think this is super [5:44] impressive. [music] It's detailed. It [5:45] has the retro vibes. I am probably going [5:49] to go home and recreate this myself. I'm [5:51] [music] going to say it's cooked just [5:52] right. [5:52] >> All right. We love to see it. Ryan, [5:54] >> no. Cosman, great job you cooked here. [5:57] >> Really well done. [5:58] >> This is one of my favorite community [5:59] builds that I think I've ever seen. [6:01] Carter Church here has used Astra to [6:04] actually decode a message that Napoleon [6:06] had sent all the way back in the day [6:08] that's remained unsolved for I guess [6:10] what two or 30 hundred years. What do [6:12] you guys think? [6:12] >> Yeah, definitely. And I already see [6:13] Brian reaching for his card. So, I'm [6:15] going to let you go first. [laughter] [6:16] >> Napoleon could have never predicted [6:18] this. This is ridiculous. You did great. [6:20] Cooked. [6:21] >> Yeah. You know, I just like the cool [6:23] factor on it. Definitely opening up [6:25] history for a broader audience. Almost 5 [6:27] million views. [6:28] >> Pretty good. [6:29] >> Perfect. So, for this next one, we have [6:31] a pretty cool horse racing game that [6:33] went pretty by RL where you draw your [6:35] character and just let it run and win [6:37] the race. What do you guys think? [6:39] >> The actual game is you're drawing the [6:42] horse and depending on how you draw it [6:45] makes it go faster, slower. [music] [6:47] Yeah. Figure. I'm going to go first on [6:49] this one, especially because uh you [6:50] know, the guy who made it is right [6:51] there. I'm going to say cooked. [6:54] [applause and cheering] Well, [6:55] >> there's not much more to say. Double [6:56] cooked. [6:57] >> Oh, first time for that. There you go. [6:59] >> All right, then to close this out, we [7:01] are going to do our catch of the day. [7:07] >> Keep that in. [7:08] >> We're going to pick our favorite build [7:09] and talk [music] about it a little bit. [7:11] So, for me, it was definitely the [7:12] Napoleon letter. What was yours? [7:14] >> Yeah, that's that's I think the the [7:16] coolest. [7:16] >> I'm going to go for the first one. I I [7:18] liked the retro space shooting game. [7:21] That's like right up my alley. I'm going [7:22] to play it when I get home. [7:23] >> And I'm loving this horse racing [music] [7:24] game, honestly. [7:25] >> All right. Napoleon wins. Now wins [7:28] >> again. [7:28] >> Again? [7:29] >> Yes. [7:29] >> He can't keep getting away with [7:30] >> that's been edge [laughter] of the day. [7:32] All right, Matt. I think the mod retro [7:34] has finally loaded our game. Are you [7:36] excited to see what's going on? [7:37] >> Yeah, let's play it. [7:38] >> All right, let's do it. Here you go. [7:39] >> Thank you. All right, so we got the [7:41] little blue dot. I [music] got to go to [7:43] the fridge. [7:44] >> Get the soft egg this time. [7:46] >> Yeah, for sure. Soft egg. [7:47] >> Stir clockwise. [7:48] >> Clockwise. Silky bronze. Perfect [7:49] noodles. Better than last time. [7:51] >> Yeah, buddy. [7:52] >> Good job. [7:52] >> Didn't you put an egg in there? [7:54] >> A sh [laughter] [7:55] Yeah. Really cool. Congrats everyone. We [7:57] did it. [7:58] >> Yeah. [applause] [8:01] [cheering] [8:02] >> All right. So, Alex, we've been at this [8:03] for a little bit. I smell all the [8:05] delicious ingredients. I've actually [8:07] gotten [music] quite hungry. What can we [8:08] do about this? [8:09] >> Yeah, I think we got a little little [8:10] carried away talking about dots [music] [8:12] and dev day. Um, but I think we actually [8:13] did cook something up, didn't we, Spy? [8:16] >> Yeah. I came prepared. [8:18] >> Oh, man. [8:20] >> There we go. [8:21] >> Wow, that is beautiful. Did [music] we [8:23] cook this? We absolutely cooked this. [8:25] Yes. Everyone knows. [8:26] >> I can't believe. Yeah. Can't believe we [8:28] cooked that. That looks beautiful. Is [8:29] that it? You want to? [8:30] >> We actually [music] cooked with this [8:31] one, boys. [8:32] >> Look at that. Yeah. Put it right on the [8:34] laptop. Chopsticks right here. Smells [8:36] excellent. [8:37] >> Looks beautiful, too. [music] [8:38] >> I can't believe we cooked that. [8:40] >> Matt, how is it? [8:41] >> It's actually really good. [laughter] [8:44] >> Yeah. [8:47] Oh, Danny, you're back. Where' Freddy [8:49] go? Never mind. Anyways, guys, this has [8:51] been Cooked. Thank you so much for [8:52] joining us. And thank [music] you to our [8:54] special guests, Matt and Brian. We had a [8:56] great time. Played some games, made some [8:57] memories. We'll see you again next time. [9:00] Thank you. [music] [9:01] >> Always did it [singing] my way. Never [9:02] did it by the book. Let me cook now. Let [9:05] me cook. Just let me cook.