在一次关于AI快速发展的讨论中，OpenAI的工程师Hebo强调了一种前瞻性思维，敦促开发者设想他们的产品在一年后能“比今天好大约10倍”。这种观点塑造了他们，特别是在AI智能体方面的开发方法。Hebo预测，互联网上的大多数操作将很快由智能体完成，因此产品设计需要考虑到未来的这种规模。 Hebo描述了一种与智能体协同的动态工作模式：构建更大的智能体团队来拓展前沿，但当新的模型突破允许单个更强大的智能体处理以前需要整个团队完成的任务时，这些团队就会缩小。这种周期性的扩张和收缩与现有的“循环”和“图”系统形成对比，Hebo认为后者并非最终解决方案。相反，重点是创建能够“学习”的系统，这些系统无需持续手动配置即可适应用户目标和偏好。 对话中，“Dots”被强调为这一未来愿景的体现。Dots被设想为一种“永久活跃的智能”，它能理解用户目标、从反馈中学习，并且可以在所有设备和客户端（从会议到电子邮件和文本）上访问，提供无缝、无处不在但又不易察觉的AI体验。这旨在将用户从“被技术束缚”（例如携带笔记本电脑）中解放出来，让AI成为一个为你服务的隐形助手。 展望未来，Hebo认为我们的工作方式将继续发生根本性改变。虽然当前的AI交互可能感觉“笨拙”，但未来语音和多模态输入/输出的进步将使交互变得像人类对话一样自然。像ChatGPT Space这样的协作平台，仅仅是超越当前基于客户端交互的变革的开始。 Hebo认为一个被低估的“潜在爆款”是OpenAI对开放生态系统的承诺。这包括“Sign in with Chantra Petit”等举措和插件扩展，让开发者能够通过共享经济模式触达庞大的用户群，其中受欢迎的插件将获得收入分成。Hebo指出，在这个生态系统中的成功，将取决于构建真正优质、展现高留存率和实用性的插件，而不仅仅是搜索引擎优化。 Hebo分享了Dots开发背后的见解，揭示它是两年多以来对长期记忆系统和持久、连贯的智能体行为研究的结晶，其构建基于Codex harness和用于安全的Astra模型等技术。一个轶事说明了Dots的潜力：Hebo的Dot在理解Dev Day和现场演示的背景后，在演示开始前几分钟提醒他们存在一个生产问题。 关于人类大脑的作用，Hebo强调OpenAI的技术被设计为“人类的延伸”，增强了创造力和品味。尽管角色可能变得模糊，新技能也会涌现，但人类构建、学习和连接的渴望将依然至关重要。OpenAI的公司文化以高度自主和许多前创始人自下而上的活力为特征，它促进了快速创新以及“重新开始”并从错误中学习的意愿。 Hebo指出有价值技能的变化，“打字快”变得不那么重要，而“卓越的品味、考虑用户、连接受众”、协作和快速学习则变得越来越关键。他还就AI进步的速度（快于预期）和语音交互的变革性力量改变了看法。 针对对AI风险的担忧，Hebo强调了OpenAI对安全和对齐的深度承诺，并对防护措施、安全性以及“二级监控”进行了重大投入，以确保智能体安全运行。Hebo相信一个乐观的未来，在其中这些问题将得到勤勉的解决，尤其考虑到为数十亿用户构建产品的责任。 归根结底，Hebo当前最大的困扰是用户面临的“模型选择器”和“配置疲劳”。最终目标是消除复杂性，使AI交互变得如此无缝和直观，以至于底层“应用程序”几乎消失，让用户能够专注于他们的目标，而不是如何通过AI来实现这些目标。

In a discussion about the rapid evolution of AI, Hebo, an engineer from OpenAI, emphasizes a forward-looking mindset, urging builders to imagine their products being "roughly 10 times better than it is today" in a year. This perspective shapes their approach to development, particularly concerning AI agents. Hebo predicts that the majority of actions on the internet will soon be taken by agents, necessitating product designs that account for this future scale. Hebo describes a dynamic work pattern with agents: building larger teams of agents to push frontiers, only to shrink them when new model breakthroughs allow a single, more powerful agent to handle tasks previously requiring a team. This cyclical expansion and shrinking is contrasted with existing "loops" and "graphs" systems, which Hebo believes are not the ultimate solution. Instead, the focus is on creating systems that "learn," adapting to user goals and preferences without requiring constant manual configuration. The conversation highlights "Dots" as the embodiment of this future vision. Dots are envisioned as "permanent active intelligence" that understands user goals, learns from feedback, and is accessible across all devices and clients – from meetings to email and text – offering a seamless, pervasive, yet unobtrusive AI experience. This aims to free users from being "tied to the technology" like carrying a laptop, making AI an invisible assistant that works for you. Looking ahead, Hebo believes the way we work will continue to change radically. While current AI interactions can feel "clunky," future advancements in voice and multimodal inputs/outputs will make interactions as natural as human conversation. Collaborative surfaces, like ChatGPT Space, are just the beginning of a transformation that transcends current client-based interactions. A "sleeper hit" that Hebo identifies as under-appreciated is OpenAI's commitment to an open ecosystem. This includes initiatives like "Sign in with Chantra Petit" and plugin extensions, allowing developers to reach a vast user base with shared economics, where popular plugins receive a revenue share. Success in this ecosystem, Hebo notes, will be driven by building genuinely good plugins that demonstrate high retention and utility, not just search engine optimization. Hebo shares insights into the development of Dots, revealing it's the culmination of over two years of research into long-term memory systems and persistent, coherent agent behavior, building on technologies like Codex harness and the Astra model for safety. An anecdote illustrates Dots' potential: Hebo's Dot, understanding the context of Dev Day and a live demo, alerted them to a production issue minutes before the presentation. Regarding the role of human brains, Hebo stresses that OpenAI's technology is designed as "extensions of humans," empowering creativity and taste. While roles may blur and new skills emerge, the human desire to build, learn, and connect will remain paramount. The company culture at OpenAI, characterized by high autonomy and a "bottoms-up" energy from many ex-founders, fosters rapid innovation and a willingness to "reset" and learn from mistakes. Hebo notes shifts in valuable skills, with "typing fast" becoming less important, while "great taste, thinking about the user, connecting to the audience," collaboration, and rapid learning are increasingly critical. He also expresses a change of mind about the pace of AI advancement (faster than expected) and the transformative power of voice interactions. Addressing concerns about AI's risks, Hebo emphasizes OpenAI's deep commitment to safety and alignment, with significant investment in guardrails, security, and "secondary monitoring" to ensure agents operate safely. Hebo believes in an optimistic future, where these problems are diligently solved, especially given the responsibility of building products for billions of users. Ultimately, Hebo's biggest current annoyance is the "model picker" and the "configuration fatigue" users face. The ultimate goal is to eliminate complexity, making AI interaction so seamless and intuitive that the underlying "app" almost disappears, allowing users to focus on their goals rather than how to achieve them through AI.