🌐 双语
Archive

AI Builders
Digest

2026-11-14 9 builders · 21 tweets · 1 podcasts · 0 blogs

🔥 热点话题

Stand with Jensen 运动迅速发酵Stand with Jensen campaign gains momentum

Replit CEO Amjad Masad 呼吁大家阅读并签署 Stand with Jensen 公开信,同时帮助提升了 StandWithJensen 页面的速率限制,使其能更好地扩展。Y Combinator 总裁 Garry Tan 也签署了该信,强调 Jensen 应获得正当程序而非当前的处理方式。Guillermo Rauch 亦 co-sign 相关内容。这一话题在 AI 圈引发广泛关注与转发。
Replit CEO Amjad Masad urged people to read and sign the Stand with Jensen letter and worked with the page owner to raise rate limits so the Replit-hosted site can scale. Y Combinator President & CEO Garry Tan signed as well, stating Jensen deserves due process. Vercel CEO Guillermo Rauch co-signed related posts. The hashtag and letter became a focal point across the builder community.
查看原文 →查看原文 →查看原文 →查看原文 →

FAA 全国机场地面停飞事件FAA nationwide airport ground stop

Nikunj Kothari 指出,这是现代史上仅第三次全国性地面停飞(前两次为 9/11 和 2023 年 NOTAM 系统故障),原因尚不明确。Matt Turck 则以“VC 们很快就会对此发表意见”调侃这一突发状况。事件发生在周五的 13 号,引发科技圈关注。
Nikunj Kothari noted this is only the third nationwide ground stop in modern history (after 9/11 and the 2023 NOTAM outage), with reasons still unclear. Matt Turck quipped that VCs would soon opine on the situation. The event on Friday the 13th drew attention across tech and AI circles.
查看原文 →查看原文 →

💰 创业成功案例

Trajectory:让持续学习变得人人可用Trajectory: Making continual learning accessible to every company

The Takeaway:真正的持续学习会把生产环境中的 traces 闭环成模型改进,让每家公司都能像拥有自己的研究实验室一样训练模型,而无需数百人团队和上亿算力。

Ronak(Trajectory 创始人,前 Windsurf)与来自 DeepMind、Apple Vision Pro 的联合创始人一起,正在把训练、评估和部署做成“一键完成”的产品体验。他把当前状态比作 AAA 纸质地图时代:智能体很强,但更新路线必须整批重训;目标是达到 Google Maps 那种边用边学的实时更新。Harvey 的案例显示,用 Trajectory 训练出的模型成本降低 6 倍,准确率从开源基线约 4% 提升到 18%。核心原语是生产 traces、可回放环境,以及把 gRPO 改造成 group size 1 的算法和用少量标注训练 meta grader 的奖励机制。Ronak 强调:“我们把 training、eval 和 deployment 放在同一个地方,让每个人都能轻松完成,然后把模型再传回生产、收集新 traces——这会是每个公司很快都会采用的范式。”最终愿景是从今天只有 1% 的企业真正用上 AI,变成 99% 都能把领域经验固化进自己的模型。
The Takeaway: True continual learning closes the loop from production traces to model improvement so every company can train models like it has its own research lab, without hundreds of researchers or hundreds of millions in compute.

Ronak (Trajectory founder, formerly Windsurf) and co-founders from DeepMind and Apple Vision Pro are packaging training, evaluation and deployment into a one-click experience. He compares today’s agents to the AAA paper-map era—powerful but requiring full batch retraining for any course correction—versus the Google Maps future of continuous online updates. On Harvey, a Trajectory-trained model became 6× cheaper while lifting accuracy from roughly 4% on an open-source baseline to 18%. The primitives are production traces, replayable environments, a group-size-1 variant of gRPO that removes the need for identical rollouts, and meta-graders trained on a small dense set of labels. Ronak notes: “By putting training, eval, and deployment in one place and making that dead simple… it suddenly is this magical paradigm that I think every single company is going to adopt very soon.” The long-term goal is to move enterprise AI adoption from the current 1% to 99% by letting domain experience continuously improve the models that run the work.
查看原文 →

Replit 与 Visa 合作推进 Agentic CommerceReplit partners with Visa on agentic commerce

Replit CEO Amjad Masad 表示,与 Visa 在 agentic commerce 上的合作进展顺利,更多功能和用例即将上线。同时他直接参与支持了 StandWithJensen 页面的扩展,体现了 Replit 在快速部署和实时扩容上的能力。
Replit CEO Amjad Masad highlighted the ongoing partnership with Visa on agentic commerce and said more features and use cases are coming. He also personally helped raise rate limits on the StandWithJensen page hosted on Replit so it could handle the surge in traffic.
查看原文 →查看原文 →

🛠️ 开发者工具与技巧

Vercel Workflow SDK 5 与 durable-harnessVercel Workflow SDK 5 and durable-harness

Vercel CEO Guillermo Rauch 宣布 Workflow SDK 5 和 Vercel Workflows 是 agent 基础设施的重大进步:首步延迟和步骤间延迟大幅降低,加密与安全进一步加强,系统可靠性在大规模场景下得到验证,并抽出可复用的 durability log,催生了 durable-harness。他同时表示 Vercel Sandbox “不可避免”。
Vercel CEO Guillermo Rauch described Workflow SDK 5 and Vercel Workflows as a monumental step for agents: dramatically lower time-to-first-step and inter-step latency, stronger encryption and security, proven reliability at massive scale, and a reusable durability log that gave birth to durable-harness. He also declared “Vercel Sandbox is inevitable.”
查看原文 →查看原文 →

Garry Tan 的 harness 评测与模型成本对比Garry Tan’s open-source harness eval and model cost comparison

Y Combinator 总裁 Garry Tan 分享了自己的开源 harness 评测仓库。在 Claude Code 的 52 个编程任务上,Fable 5.2 解题数量与 Opus 5.5 接近(82.7% vs 84.6%),但每个成功任务的成本却是后者的 2.80 倍。因此他建议继续把 Opus 5.5 作为 Claude Code 的默认编码模型,并指出真正拉开差距需要更难的任务。
Y Combinator President & CEO Garry Tan released his open-source harness eval repo. On 52 coding tasks in Claude Code, Fable 5.2 solved almost as many problems as Opus 5.5 (82.7% vs 84.6%) yet cost 2.80× more per solved task. He therefore recommends keeping Opus 5.5 as the coding default and notes that harder tasks will be needed to differentiate the models further.
查看原文 →查看原文 →

Peter Yang 的 /fuck-cancer AI skillPeter Yang’s free /fuck-cancer skill

Peter Yang 发布了一个可免费使用的 AI skill:把患者全部病历上传到文件夹,然后让 ChatGPT 或 Claude 安装该 skill,即可生成一页摘要、检索临床试验等。他鼓励有需要的人 fork 并适配其他长期疾病,用 AI 全力对抗恶性疾病。
Peter Yang released a free AI skill that lets anyone upload a patient’s full medical records, point ChatGPT or Claude at the skill, and instantly generate a one-page brief, research clinical trials, and more. He invites people to fork it for other long-term conditions and use the full power of AI against serious disease.
查看原文 →

反对盲目刷榜,真正用起来才知道Stop benchmark-maxxing; real workflows matter

Nikunj Kothari 批评在 Llama 4 之后本该停止的“刷榜”又卷土重来。他强调真正重要的只有极少数基准,其余多为营销姿态;唯一可靠的方法是把模型放进自己的真实工作流里去感受。
Nikunj Kothari observed that benchmark-maxxing has returned after Llama 4. He argues only a handful of benchmarks truly matter and the rest is mostly posturing; the only reliable way to know which model works is to run it on your actual workflows.
查看原文 →

🌍 其他动态

Dan Shipper 谈软件付费与 AI 核心公式Dan Shipper on why teams still pay for software and the most important formula in AI

Every CEO Dan Shipper 分享了为什么即使 Fable 5.2 能 vibe-code 出任何东西,团队仍愿意为软件付费的文章,并指出“AI 里最重要的公式”。同时他强调某些正在变得更常见的问题必须尽快修复,整个 AI 社区都应帮忙。
Every CEO Dan Shipper shared a piece on why teams still pay for software even when Fable 5.2 can vibe-code almost anything, and highlighted what he calls the most important formula in AI. He also flagged emerging issues that need fixing and urged the whole AI community to help.
查看原文 →查看原文 →查看原文 →

SPC 投资组合入选 AI 100SPC portfolio companies in the AI 100

SPC 普通合伙人 Aditya Agarwal 为自己投资组合中的多家公司入选 AI 100 感到自豪,包括 Baseten、Render、Profound、Replit、Gamma、Cog 和 Density,并感叹当前处于“极其奇怪的时间线”。
SPC General Partner Aditya Agarwal expressed pride that portfolio companies Baseten, Render, Profound, Replit, Gamma, Cog and Density made the AI 100 list, while also remarking that “we are in an extremely weird timeline right now.”
查看原文 →查看原文 →