Перейти к содержимому

#002 iPhone Duo 来了,Astra 离 AGI 还有多远?

Next Token | 词元之外

0:00 / 0:00

#002 iPhone Duo 来了,Astra 离 AGI 还有多远?

122 просмотра · 2 недели назад
Next Token | 词元之外
18 подписчиков
122 просмотра · 2 недели назад
本期节目与更新 / Episode page: https://nexttoken.tv/weekly/002/?utm_... iPhone Duo 发布,折叠屏会带来怎样的新交互?另一边,Astra 已经能操作电脑、调用工具完成任务——会用电脑,就算 AGI 了吗? 节目从 GPT-6 Astra 的 3D 建模体验和 Next Token 自己的播客制作实践聊起:能操作电脑、完成任务,与通用智能之间还差什么?几位参与者给出了不同判断。 DeepSeek Flash 与 Harness 带来另一条线索:当速度、成本和能力逐渐够用,瓶颈可能转移到测试、交付和人的判断。我们讨论云端 Agent、常驻电脑与 Agent API,以及小团队应该自己搭框架,还是专注数据、接口与具体场景。 后半场转向 iPhone Duo、折叠屏交互、界面一致性、手表与个人上下文、音乐和图像编辑、语音输入,以及模型是否真的“变笨”。 本期录制于 2026 年 9 月 12 日,地点为上海 AGI Bar。体验、观点与预测属于参与者个人判断,不代表产品能力的独立验证;功能与可用性以官方信息为准。健康相关讨论仅为个人体验,不构成医疗建议。 本期主理人:向阳乔木(本期当值)、歸藏、橘子、杨攀。 What does the iPhone Duo launch mean for interaction design? And if Astra can use a computer, does that make it AGI? Recorded at AGI Bar in Shanghai on September 12, 2026, this roundtable starts with GPT-6 Astra and 3D modeling, then turns to our own experiments using agents to produce and publish a podcast. We ask what separates completing real tasks from general intelligence, and where we disagree. DeepSeek Flash and Harness raise another question: once models are fast and capable enough, do testing, delivery, and human judgment become the bottlenecks? We discuss persistent cloud computers, Agent APIs, and whether small teams should build their own harness or focus on data, interfaces, and specific use cases. The second half explores iPhone Duo, foldable interfaces, design consistency, watches and personal context, music and image editing, voice input, and whether models really get worse. These are participants’ experiences, opinions, and predictions, not independent verification of product capabilities. Consult official sources for features and availability. Health-related discussion is personal experience, not medical advice. Co-hosts: Xiangyang Qiaomu (this episode’s host), Guizang, Orange, and Yang Pan. Audio language: Mandarin Chinese. Chinese and English subtitles available. Chapters / 时间轴 00:00 节目片头 / Intro 00:49 上海开场:本周模型、Agent 与硬件 / Opening in Shanghai 01:44 GPT-6 Astra:从 3D 建模聊起 / GPT-6 Astra and 3D creation 04:45 Blender 与代码:给模型合适的工具 / Blender, code, and the right tools 09:17 我们如何用 Agent 制作和发布播客 / Using agents to produce and publish a podcast 17:21 Astra 会用电脑,就算 AGI 了吗? / Does computer use make Astra AGI? 20:31 写作体验:不同模型各有所长 / Different models, different writing experiences 21:54 DeepSeek Flash 与 Harness 的使用体验 / Trying DeepSeek Flash and Harness 23:06 速度变快之后,人的判断更重要了吗 / When speed shifts the bottleneck 24:41 Harness 更新太快,插件和工作流如何跟上 / Keeping workflows working through rapid updates 27:16 能力够用之后,测试与交付成为瓶颈 / From model capability to testing and delivery 29:29 模型路由与稳定性:用户到底在购买什么 / Model routing and stability 34:15 Grok Bot / Muse:为什么 Agent 需要云电脑 / Why agents need cloud computers 38:58 Agent API:还需要自己维护一套 Harness 吗 / Agent APIs and maintaining a harness 44:51 Agent 服务会成为新的云基础设施吗 / Agents as a cloud service 48:29 垂直应用的价值:数据、接口与上下文 / Vertical applications: data, interfaces, and context 50:20 平台内 AI 助手:入口简单,体验才完整 / AI assistants inside existing platforms 55:02 iPhone Duo 发布:折叠屏与交互的新问题 / iPhone Duo launch: foldables and interaction design 59:26 界面标准与长期一致性为什么重要 / Interface standards and consistency 01:06:40 手表、录音与个人上下文 / Watches, recording, and personal context 01:09:48 健康数据与日常记录:有用与隐私的边界 / Health context, usefulness, and privacy 01:16:02 音乐与图像编辑:可控修改的进步与局限 / Controllable music and image editing 01:18:57 设计 Agent 为什么也需要执行环境 / Why design agents need execution environments 01:22:16 临时环境与常驻电脑的区别 / Temporary sessions versus persistent computers 01:23:06 语音输入:准确率之外的使用摩擦 / Voice input beyond transcription accuracy 01:26:56 模型真的变笨了吗:体验、路由与评测 / Do models get worse? Experience, routing, and evaluation 01:33:31 Mistral、Kimi 与模型产品的不同选择 / Different choices in model products 01:39:39 从代码到实物:3D 打印带来的创作空间 / From code to physical objects with 3D printing 01:41:51 上海 AGI Bar 致谢与收尾 / Closing and thanks to Shanghai AGI Bar 01:42:14 片尾 / Outro 相关资料 / Related links DeepSeek Harness: https://github.com/deepseek-ai/deepse... Godot: https://godotengine.org/ AGI Bar: https://agi.bar/ 关于节目 / About the show Next Token|词元之外是一档关注 AI 技术、产品、创业与创作的视频播客。 Next Token|词元之外 is a video podcast about AI technology, products, startups, and creative work. Beyond the Next Token. 节目主页 / Show website: https://nexttoken.tv 你更期待 iPhone Duo 改变交互,还是 Astra 改变工作?会用电脑是否足以成为 AGI 的判断标准?欢迎留言。 Which matters more: iPhone Duo changing how we interact, or Astra changing how we work? Is computer use enough to call something AGI? Tell us what you think. 制作与致谢 / Credits 感谢上海 AGI Bar 提供录制场地,感谢赛博禅心与南乔邀请。 Venue partner: AGI Bar, Shanghai (https://agi.bar/). Thank you to 赛博禅心 and 南乔 for inviting us. “Electric Dreams” by Scott Buckley — released under CC BY 4.0. www.scottbuckley.com.au “Origami” by Scott Buckley — released under CC BY 4.0. www.scottbuckley.com.au Edited and mixed for Next Token Weekly.