OpenAI DevDay 2026主题演讲
文章语言:
简
繁
EN
Share
Minutes
原文
会议摘要
Dev Day highlighted AI advancements, introducing Dots for task management, GPT 6.1 Soul for cost-effective intelligence, and new tools like Codex and Agents API. The event emphasized democratizing AI, showcasing OpenAI Marketplace and collaborative platforms, aiming for a renaissance-like future through innovative solutions and community support.
会议速览
The dialogue highlights the launch of Dots, advanced AI agents designed to assist with tasks, enhancing user productivity and allowing focus on higher-level work. The speaker expresses gratitude for user feedback, which has driven improvements, including expanding Codex's capabilities to Linux, multiple folders, and mobile devices. Dots represent a significant leap in AI technology, aiming to function as an always-on, context-aware helper that can autonomously handle various tasks, freeing users to concentrate on what matters most to them.
A newcomer introduces themselves, taking on a leadership role, and outlines their duties in managing tasks and seeking approvals within the ChatGPT environment.
Discussed the latest design for the website, confirmed the deck, and detailed the app migration process, with all changes mapped out in Google Meet notes.
A cake vendor cancellation is mitigated by finding a backup, scheduling a tasting for Saturday at 11 AM, while ensuring all arrangements align with current products before final review.
Dialogue covers website redesign approval, photo adjustments, and updates on business growth, including a 51% revenue increase year-over-year.
A system rebuild has been successfully tested without regressions, focusing on enhancing consumer experience first. The preview and results are shared in a PR, and plans are in place to roll out the changes to the team on Slack and prepare for the stage rollout.
A discussion revolves around adjusting schedules to accommodate a child's recital and a finance review, emphasizing the importance of the recital and planning to reschedule conflicting events.
AI Dots, powered by Astra, offer personalized assistance, handling everything from urgent notifications to complex projects, enhancing focus and productivity. They can write code, manage tasks, and integrate with over 4000 apps, providing updates via text or calls, ensuring users maintain control and trust in their delegated responsibilities.
Introduces ChatGPT Space, a collaborative environment merging AI and human efforts for planning, research, and creation. Features include AI tracking, live document interaction, and AI-assisted presentation building, enhancing team productivity and innovation.
A member of a product team introduces the concept of Dots, AI assistants that streamline work processes by managing tasks, updates, and communications. Dots proactively handle scheduling changes, feature testing, design reviews, and facilitate brainstorming sessions, acting as a second brain for enhanced productivity and seamless team collaboration during critical project phases like app launches.
Inquires about last night's user testing progress, checks current status, and seeks an update on how it's proceeding.
Discussion revolves around reviewing the go-to-market strategy, with an offer to enhance the page by adding an FAQ, showcasing team collaboration and proactive engagement in project management.
Introduces Space Awesome, a collaborative tool within ChatGPT for team projects, highlighting AI Dots' role in managing tasks, filtering noise, and executing work, thereby enhancing productivity and enabling teams to tackle larger projects.
Introduces AI Dots, virtual teammates for enterprise users, and announces GPT 6.1 for faster, cheaper AI capabilities, aiming to boost productivity and innovation across various business sectors.
A new ultra-fast API, offering eight times the speed of standard models, is launched, enabling quicker task completion. Alongside, Pro 500 subscription with higher usage limits and access to ultra-fast API is introduced, alongside the reopening of Pro 200 subscription for daily use of high-quality models.
The new Decisions API enables models to respond almost instantaneously, significantly speeding up tasks like request routing, image classification, and agent decision-making. This enhancement maintains core capabilities including image understanding, language support, and safety features, promising exciting advancements in model efficiency.
OpenAI's advancements in AI models, including the introduction of AI research interns, have significantly improved computing speed, performance, and safety. The models have optimized latency, enhanced monitoring, and aligned better with user intent, accelerating research and industry standards.
OpenAI's models, like Astra, have shown significant progress in tackling long-duration tasks, contributing to breakthroughs in math, health, energy, and robotics. The company is enhancing developer support by making Codex harness open source, scaling it for billions of users, and launching a cloud-based version. A new security cloud tool for vulnerability detection and fixes is also introduced, offering features like automatic deduplication and scheduled scans.
OpenAI launches its Agents API, offering core components and new computer-use capabilities for agents. The technology supports partnerships with AWS, enabling OpenAI agents in AWS alongside existing applications. A focus on data safety introduces OpenAI Private Intelligence for stronger data controls.
Discusses advancements in privacy and reliability of AI systems, highlighting 100x growth in API usage with 99% reliability, and a 45% reduction in time to first token, setting new industry standards. Prepares for a demo showcasing new tools for developers.
The dialogue highlights the introduction of AI technology and new tools for developers at Fort Mason, drawing parallels to historical technological advancements, such as the first long-distance telephone calls. It emphasizes the potential of these innovations to transform the developer community, much like past technologies have revolutionized communication.
The speaker highlights the evolution from lacking an app for Codex to now offering powerful tools for 3D world creation with Astra. They express excitement about real-time collaboration capabilities and the ability to work from anywhere, showcasing advancements in technology that have transformed the creative process.
A demonstration showcases using Ultra-Fast and Codex CLI to interactively build and modify applications, including random ticket giveaways, in real time, highlighting the platform's speed and versatility.
The dialogue highlights the capabilities of Astra and Codex in enhancing game graphics, automating app functionalities, and streamlining development tasks in the cloud, showcasing their potential in game creation and app management.
Initiating a full backend rewrite in Rust, leveraging cloud capabilities for continuous progress tracking and facilitating seamless idea handoffs via mobile, enhancing development efficiency and collaboration.
An AI system demonstrates its capabilities by controlling a programmable robot, creating a visual drawing based on audience interaction, showcasing the integration of text, audio, image generation, and visual understanding in a live setting.
The dialogue showcases the integration of Astra Vision, GPT Image 2.5, and GPT Live 1 to enhance AI capabilities, including vision, image generation, and real-time voice interaction, demonstrating rapid advancements in AI technology.
The dialogue highlights the excitement around new AI tools, particularly the Decisions API, which enables real-time processing of vision inputs for robotics. It emphasizes the potential for these tools to unlock innovative interactions with AI in the real world. Additionally, it discusses the integration of ChatGPT subscriptions into third-party apps, allowing users to leverage their existing AI subscription for new products, thus facilitating business growth and reducing initial costs for users.
The dialogue introduces groundbreaking AI enhancements, including plugin extensions for building native applications within ChatGPT, the OpenAI Marketplace for business solutions, and new AI models like GPT-4. It highlights how these tools empower users to innovate, collaborate, and create personalized software experiences, underscoring the transformative impact on AI-driven industries.
Attendees are excited about the day's events, featuring a focus on open source contributions, a retro game-building session with Codex, and a live QA. Special mentions include new tools, the Open Call Enterprise Harness, and exclusive merchandise, all aimed at fostering creativity and community engagement.
A group decides to embark on a subtle reset initiative, aiming to refresh their approach without grand gestures, using a device for the purpose and expressing excitement about the change.
Initiates a countdown activity, inviting participation and creating an energetic, interactive moment.
The dialogue emphasizes envisioning AI's future as a renaissance rather than an industrial revolution, advocating for technology to enhance human creativity, learning, and control over life, urging collective effort towards this empowering vision.
要点回答
Q:What are the new features of AI technology introduced in the transcript?
A:New features of AI technology introduced include the ability of AI agents, called 'Dots,' to handle various tasks, work independently, and assist users in their work, providing more time and attention for higher level tasks.
Q:What are the capabilities of the new AI agents, referred to as 'Dots'?
A:The new AI agents, 'Dots,' are capable of handling a wide range of tasks, including reviewing overnight information and urgent matters, taking care of small tasks like booking plane tickets, and managing more ambitious work such as monitoring bug reports, testing fixes, drafting PRs, and overseeing budget planning.
Q:How do Dots learn and assist users over time?
A:Dots get to know how users like to work and their focus areas over time. They can prioritize tasks, handle overnight information by flagging urgent matters in the morning, and provide assistance throughout the day based on user preferences.
Q:What kind of work can Dots handle for users?
A:Dots can handle a variety of tasks for users, from small tasks like booking reservations to larger projects such as home renovations and managing complex tasks like code migration and budget planning.
Q:What safeguards are in place for Dots?
A:Dots are powered by Astra, a model that has built-in protections and safeguards. Users can set boundaries on app and computer use, and customize instructions for any actions taken by their Dot, ensuring users can trust their Dot with the level of responsibility they are comfortable with.
Q:What is ChatGPT Space and what features does it offer for teamwork?
A:ChatGPT Space is a collaborative platform where individuals, their teams, and AI agents can work together. It features a space where users can write plans, conduct research, create images, generate data, and build new things, with the ability for AI dots to work alongside users in the same page. Teammates and AI agents can collaborate in this space, which integrates files and live components like updates from external channels.
Q:What new capability is being introduced in ChatGPT Space, and how will it aid in creating presentations?
A:A new capability being introduced in ChatGPT Space is the ability to build presentations directly inside the space. This will allow teams to collaborate on presentations with their AI agents and other team members, making the process of creating and modifying presentations more efficient and accessible for both humans and AI.
Q:How does Dotty, the AI dot, assist with work and what is its significance to the speaker?
A:Dotty is an AI dot that assists the speaker with various tasks, including catching important updates, managing the speaker's calendar, and reviewing and testing new features. Dotty also provides context on ongoing projects and is capable of taking on tasks, such as updating design feedback, freeing up the speaker's time to focus on other responsibilities.
Q:How can Space facilitate teamwork and integration with AI dots?
A:Space facilitates teamwork by allowing users to collaborate on various projects, such as creating interactive charts and live prototypes, and by enabling direct integration with AI dots. Team members can tag AI dots within Space to assist with tasks, like generating bar charts or incorporating feedback from different sources, keeping everyone in sync and ensuring that no detail is missed.
Q:In what ways are Dots expected to change workflows and become part of group chats?
A:Dots are anticipated to change workflows by acting as extensions of teams, handling tasks independently, and allowing team members to focus on higher-level strategies. Dots are now being integrated into group chats, where they can receive and relay information, and even take on tasks like fixing bugs or opening PRs with fixes, as they are context-aware and have the ability to interact with external systems.
Q:What advanced capabilities does Dotty demonstrate in the speaker's demonstration?
A:Dotty demonstrates advanced capabilities such as using Codex to build an app and launch it on an iPhone simulator. The demonstration includes Dotty running local execution of a mock on the speaker's phone, which results in a working app. Dotty is also shown to be able to autonomously handle tasks like shipping PRs, illustrating its proficiency in executing complex software development tasks.
Q:What are the capabilities of Dots and how have they been beneficial at OpenAI?
A:Dots, as extensions of OpenAI, can take work off of our plates by filtering noise and are changing what OpenAI feels capable of taking on. They are an integral part of OpenAI's operations, helping to scale and facilitate work across various projects and roles.
Q:What are specialist dots, and what industries can they support?
A:Specialist dots are virtual teammates set up by companies to work alongside entire teams, designed to scale intensive work in various sectors such as accounting, marketing, and legal. They can be customized with goals and contexts, reviewed, and improved with feedback, and are available to enterprise users.
Q:What is the purpose of the new platform launched by OpenAI?
A:The purpose of the new platform launched by OpenAI is to support users in building whatever they can dream of by providing access to the best intelligence at the speed and cost needed, better tools, and distribution to help discover and connect with new customers.
Q:What are the features of the new model GPT 6.1 soul?
A:GPT 6.1 soul is a powerful model that offers a balance of cost and capability, providing near Astra level intelligence at a much lower price, specifically one-fifth the cost. It is extremely affordable for agents to reuse context, iterate on solutions, and tackle longer tasks.
Q:What is ultra-fast, and how does it compare to standard models?
A:Ultra-fast is an ultra-fast API for ChatGPT and codex that offers eight times the speed at six times the price of standard models, with a remarkable 300 tokens per second capability. It provides an incredibly quick response, and is already available with Astra and will soon be offered with GPT 6.1 as well.
Q:What are the new Pro 500 and Pro 200 subscriptions, and what benefits do they offer?
A:The Pro 500 subscription offers the highest usage limits, access to ultra-fast speeds in ChatGPT and codex, and 25 times the usage of Plus. The Pro 200 subscription continues to provide access to all Frontier models, including GPT 6.1, which will offer a model nearly as effective as Astra at a more affordable rate.
Q:What is the new Decisions API, and how does it function?
A:The Decisions API is a fast response model that operates in a fraction of a second, which is ideal for quickly responding to requests and decisions. It uses a Luna model that has a predefined set of options to choose from, which allows it to be extremely fast while maintaining capabilities like image understanding, broad language support, and safety protections.
Q:What does OpenAI aim to be for developers, and what is their achievement with AI research interns?
A:OpenAI aims to be a one-stop shop for developers to build, offering leading models across intelligence, price, speed, and more. Their achievement with AI research interns includes developing a system capable of taking on complex research tasks that would typically take a skilled researcher a lot of time and effort to complete.
Q:What improvements have been made using the models in Astra for computer use?
A:The models have helped optimize the harness used for computer use, resulting in a more than 2x latency improvement that has been shipped to all users. Additionally, Astra has become safer due to improved monitoring and model behavior training, which has led to industry-leading performance on computer use stress tests.
Q:How are OpenAI's models contributing to solving complex challenges?
A:OpenAI's models, like Astra, have already helped solve over 100 math problems that were unsolved for decades and are aiding in creating new antibiotics for drug-resistant infections, studying ancient languages, developing renewable energy, and improving energy efficiency. They are also accelerating industrial robotics in manufacturing.
Q:What new capabilities are being provided to developers by OpenAI?
A:To support developers, OpenAI is providing the Codex harness, which is open source and efficient, allowing accurate results at the fastest speed with the fewest tokens. The company is also making all these tools available in the cloud, launching a new Cloud Security harness, and introducing a new Agents API in public beta, including features like the harness, memory, and multi-agent controls.
Q:What is the purpose of the new Cloud Security harness, and what are some of its features?
A:The new Cloud Security harness aims to provide defenders with better tools to harden their infrastructure. It runs in cloud environments and continuously finds vulnerabilities, preparing verified fixes. It includes highly requested new features such as automatic deduplication, scheduled scans, and a new interface.
Q:What is OpenAI Private Intelligence, and how does it enhance data protection?
A:OpenAI Private Intelligence is a feature that offers stronger controls over data while using advanced capabilities of OpenAI's models. With private inference, it can provide privacy at the time of inference, setting a new standard for privacy in AI. OpenAI has worked with customers to design privacy and safety systems that meet their highest standards.
Q:How has OpenAI ensured the performance and reliability of their tools for developers?
A:OpenAI has made significant improvements to ensure both speed and reliability in their tools. They have seen a 100x growth in the Responses API with more than 99% reliability, reduced the time to first token by 45%, and sped up tool calls and workflows by more than 30%.
Q:What new features have been added to the Codex app?
A:The new features include ultra-fast interaction with 3D worlds, the ability to change the scene in real-time, and the introduction of bidirectional audio in the terminal, powered by GPT Live.
Q:How was the camera integrated with the audience during the presentation?
A:The camera was turned to the audience to include them in the presentation. The presenter addressed the audience directly, thanked them for tuning in, and made them feel included by waving at them.
Q:What is the significance of ultra-fast in the context of building and changing worlds with Codex?
A:Ultra-fast allows for rapid creation and modification of worlds and content in real-time with Codex, enabling users to build and reshape their creations almost as fast as they can think.
Q:What is Astra Adventures and how was it created?
A:Astra Adventures is a game developed using Astra where players can fly into space in a 3D environment. It started from a simple idea, evolving through sketches and iterations to achieve beautiful graphics suitable for a real game, with the game dynamics refined over time.
Q:How does Astra play games that are designed for human players?
A:Astra can play games designed for humans by using a browser-based connection, allowing it to control the game and learn how to play it in real-time, as demonstrated by it choosing a ship and making decisions based on that control.
Q:What is the role of the upshot in the presented personal project?
A:The upshot was used in the personal project to enable voice dictation and automation of tasks, such as taking screenshots across different screen sizes, navigating applications, and simulating actions that teach the app's features to the user.
Q:What advancements have been made to Codex in the cloud?
A:Advancements to Codex in the cloud include a complete reimagining of the service to be faster, with compatibility with all the same tools as on a laptop, including plugins and computer use. It also allows users to send their most ambitious tasks directly to the cloud without needing to keep their laptop open.
Q:What new feature related to cloud and seamless handoff is being shipped today?
A:The new feature being shipped today includes the capability to work on documents in the cloud, send other ideas from a phone, and perform a seamless handoff between different devices.
Q:What advancements in AI and multimodal integration were showcased using Hugging Face's programmable robot named Lavender?
A:The advancements demonstrated with Lavender include connecting to the robot, creating drawings, and adding vision with Astra, image generation with GPT Image 2 5, and voice intelligence with GPT Live 1, to give the robot many capabilities in real-time interaction.
Q:How does the new Decisions API affect interactions with hardware and robots?
A:The new Decisions API enables robots to process vision inputs and take fast actions based on what they see in near real-time, which is expected to unlock a new way to interact with AI in the real world.
Q:How is OpenAI supporting developers in creating their own AI integrations and businesses?
A:OpenAI is supporting developers with new APIs, tools, and features that allow the creation of AI integrations and businesses. These include new models, the ability to bring ChatGPT subscriptions to products, plugin extensions, and an easier plugin submission process.
Q:What new capabilities are provided to developers to integrate with ChatGPT?
A:Developers are provided with the ability to build plugins that feel native to ChatGPT and distribute them through OpenAI's platform, enabling the integration of entire applications, editors, dashboards, and workspaces directly into ChatGPT and Dots.
Q:How can users discover new integrations or plugins within the ChatGPT platform?
A:Users can discover new integrations or plugins in the flow of a conversation when ChatGPT recognizes that a plugin can help and connects it accordingly. The search for plugins can also be done through the normal ways, and plugins are discoverable within the ChatGPT platform.
Q:What is the OpenAI Marketplace and what purpose does it serve?
A:The OpenAI Marketplace is a platform where developers can sell their products to businesses, allowing them to use part of their existing OpenAI commitment to buy these products. It also enables the use of open source models and provides another way to reach customers already working with OpenAI.

OpenAI
Follow





