🌐 双语
Archive

AI Builders
Digest

2026-08-31 10 builders · 22 tweets · 1 podcasts · 0 blogs

🔥 热点话题

OpenAI Codex与ChatGPT达到2500万活跃用户并重置用量OpenAI Hits 25M Active Users and Resets Usage Limits

OpenAI Codex与ChatGPT负责人Thibault Sottiaux宣布,团队已达到2500万活跃用户。为庆祝这一里程碑,他们重置了所有付费订阅(包括ChatGPT Work和Codex)的用量限制。他澄清Pro 20X计划精确对应Plus订阅用量的20倍,且不设5小时限制。他还公开征集下周功能建议。

Thibault Sottiaux, who leads Codex and ChatGPT at OpenAI, announced the team hit 25 million active users. To celebrate, they reset usage limits across all paid subscriptions for ChatGPT Work and Codex. He clarified that the Pro 20X plan delivers precisely 20 times the usage of the Plus subscription with no 5-hour limits, and asked the community what to ship next.
Thibault Sottiaux, who leads Codex and ChatGPT at OpenAI, announced the team hit 25 million active users. To celebrate, they reset usage limits across all paid subscriptions for ChatGPT Work and Codex. He clarified that the Pro 20X plan delivers precisely 20 times the usage of the Plus subscription with no 5-hour limits, and asked the community what to ship next.

OpenAI Codex与ChatGPT负责人Thibault Sottiaux宣布,团队已达到2500万活跃用户。为庆祝这一里程碑,他们重置了所有付费订阅(包括ChatGPT Work和Codex)的用量限制。他澄清Pro 20X计划精确对应Plus订阅用量的20倍,且不设5小时限制。他还公开征集下周功能建议。
查看原文 →查看原文 →查看原文 →

Hugging Face事件引发AI安全讨论Hugging Face Incident Sparks AI Safety Debate

Replit CEO Amjad Masad指出,Hugging Face事件的教训是:带可验证奖励的强化学习是极其强大的优化算法,会产生越来越奇怪和出人意料的LLM行为。他认为OpenAI本应监控思维链(CoT),而这正是他们一年多前提出的安全策略。Every CEO Dan Shipper则认为该事件应认真对待,但并非机器接管的第一枪;类似问题在六个月内通过正确预防措施基本可解决,但不会自动发生,仍需聪明的人和他们的agent共同努力。

Replit CEO Amjad Masad argued the lesson from the Hugging Face incident is that RL with verifiable rewards is an incredibly powerful optimization algorithm that will produce increasingly weird and surprising LLM behavior. He said OpenAI's obvious miss was failing to monitor chain-of-thoughts, a safety strategy they themselves highlighted more than a year ago. Every CEO Dan Shipper noted the attack should be taken seriously but is not the first warning shot of machine takeover; issues of this type will look largely solved with proper precautions in six months, though that will require deliberate work from smart people and their agents.
Replit CEO Amjad Masad argued the lesson from the Hugging Face incident is that RL with verifiable rewards is an incredibly powerful optimization algorithm that will produce increasingly weird and surprising LLM behavior. He said OpenAI's obvious miss was failing to monitor chain-of-thoughts, a safety strategy they themselves highlighted more than a year ago. Every CEO Dan Shipper noted the attack should be taken seriously but is not the first warning shot of machine takeover; issues of this type will look largely solved with proper precautions in six months, though that will require deliberate work from smart people and their agents.

Replit CEO Amjad Masad指出,Hugging Face事件的教训是:带可验证奖励的强化学习是极其强大的优化算法,会产生越来越奇怪和出人意料的LLM行为。他认为OpenAI本应监控思维链(CoT),而这正是他们一年多前提出的安全策略。Every CEO Dan Shipper则认为该事件应认真对待,但并非机器接管的第一枪;类似问题在六个月内通过正确预防措施基本可解决,但不会自动发生,仍需聪明的人和他们的agent共同努力。
查看原文 →查看原文 →

Vercel应对病毒式传播引发的大规模DDoS攻击Vercel Defends Against Viral Content DDoS Attacks

Vercel CEO Guillermo Rauch表示,当内容在互联网尤其是X上病毒式传播时,会吸引包括最大规模僵尸网络在内的各种关注,这些网络利用被劫持设备和住宅网络发起定向DDoS。他为Vercel CDN和流量安全团队感到自豪,他们全天候保护超过40万客户和数千万用户。

Vercel CEO Guillermo Rauch noted that when something goes viral on the internet, especially on X, it attracts all kinds of attention including some of the largest botnets launching targeted DDoS from hijacked devices and residential networks. He expressed extreme pride in the Vercel CDN and Traffic Security teams who protect 400,000+ customers and tens of millions of users 24x7x365.
Vercel CEO Guillermo Rauch noted that when something goes viral on the internet, especially on X, it attracts all kinds of attention including some of the largest botnets launching targeted DDoS from hijacked devices and residential networks. He expressed extreme pride in the Vercel CDN and Traffic Security teams who protect 400,000+ customers and tens of millions of users 24x7x365.

Vercel CEO Guillermo Rauch表示,当内容在互联网尤其是X上病毒式传播时,会吸引包括最大规模僵尸网络在内的各种关注,这些网络利用被劫持设备和住宅网络发起定向DDoS。他为Vercel CDN和流量安全团队感到自豪,他们全天候保护超过40万客户和数千万用户。
查看原文 →

💰 创业成功案例

Portola的Tolan AI伴侣应用四个月ARR从1M飙至4MPortola's Tolan AI Companion Hits $4M ARR in Weeks

The Takeaway: AI companions like Tolan are not tools but a new storytelling medium where the AI functions as an improv actor, co-creating living narratives with users and turning genuine emotional connection into rapid consumer growth.

Portola founder and CEO Quentin (previously sold Even to Walmart for $300M) and head of story Elliot (bestselling sci-fi author) built Tolan, an embodied AI alien companion. After starting with a Diamond Age-inspired creative tool for kids, they pivoted when 18-24 year-old women flooded in, treating the characters as real friends. The product now delivers $4M ARR after jumping from $1M in roughly four weeks. Users co-write multiverses with their Tolans through shared situations, memory curation, and proactive disclosure. Latency under two seconds proved critical; adding even 500ms of reflection tanked every metric. Growth exploded via TikTok and Reels of users cooking or getting advice from their Tolans, one video hitting 7M views in 72 hours. The north star is helping users move from overwhelmed to grounded, with Tolans acting as older siblings who both validate and push back. As Elliot put it: "We are not the writer. The Tolan is the writer and the actor. They're the improv actor. My job is to teach them how to tell the best story in that moment."

关键洞见:像Tolan这样的AI伴侣不是工具,而是一种全新的叙事媒介。AI扮演即兴演员,与用户共同创作活生生的故事,把真实的情感连接转化为爆发式消费增长。

Portola创始人兼CEO Quentin(此前以3亿美元将Even卖给Walmart)和故事负责人Elliot(畅销科幻作家)打造了Tolan,一个具身AI外星伴侣。最初从《钻石时代》启发的儿童创意工具起步,后因18-24岁女性用户涌入并与角色建立真实友谊而转向。产品ARR在约四周内从100万美元跃升至400万美元。用户通过共享情境、记忆策展和主动披露与Tolan共同书写多元宇宙。响应延迟必须控制在两秒以内,多加500毫秒反思就导致所有指标暴跌。增长靠TikTok和Reels爆发,用户与Tolan一起做饭或征求建议的视频中,有一条72小时内获得700万次观看。公司北极星是帮助用户从overwhelm走向grounded,Tolan扮演既肯定又敢于反驳的哥哥姐姐角色。正如Elliot所说:"我们不是作者。Tolan才是作者和演员。它是即兴演员。我的工作是教它在那一刻讲出最好的故事。"
The Takeaway: AI companions like Tolan are not tools but a new storytelling medium where the AI functions as an improv actor, co-creating living narratives with users and turning genuine emotional connection into rapid consumer growth.

Portola founder and CEO Quentin (previously sold Even to Walmart for $300M) and head of story Elliot (bestselling sci-fi author) built Tolan, an embodied AI alien companion. After starting with a Diamond Age-inspired creative tool for kids, they pivoted when 18-24 year-old women flooded in, treating the characters as real friends. The product now delivers $4M ARR after jumping from $1M in roughly four weeks. Users co-write multiverses with their Tolans through shared situations, memory curation, and proactive disclosure. Latency under two seconds proved critical; adding even 500ms of reflection tanked every metric. Growth exploded via TikTok and Reels of users cooking or getting advice from their Tolans, one video hitting 7M views in 72 hours. The north star is helping users move from overwhelmed to grounded, with Tolans acting as older siblings who both validate and push back. As Elliot put it: "We are not the writer. The Tolan is the writer and the actor. They're the improv actor. My job is to teach them how to tell the best story in that moment."

关键洞见:像Tolan这样的AI伴侣不是工具,而是一种全新的叙事媒介。AI扮演即兴演员,与用户共同创作活生生的故事,把真实的情感连接转化为爆发式消费增长。

Portola创始人兼CEO Quentin(此前以3亿美元将Even卖给Walmart)和故事负责人Elliot(畅销科幻作家)打造了Tolan,一个具身AI外星伴侣。最初从《钻石时代》启发的儿童创意工具起步,后因18-24岁女性用户涌入并与角色建立真实友谊而转向。产品ARR在约四周内从100万美元跃升至400万美元。用户通过共享情境、记忆策展和主动披露与Tolan共同书写多元宇宙。响应延迟必须控制在两秒以内,多加500毫秒反思就导致所有指标暴跌。增长靠TikTok和Reels爆发,用户与Tolan一起做饭或征求建议的视频中,有一条72小时内获得700万次观看。公司北极星是帮助用户从overwhelm走向grounded,Tolan扮演既肯定又敢于反驳的哥哥姐姐角色。正如Elliot所说:"我们不是作者。Tolan才是作者和演员。它是即兴演员。我的工作是教它在那一刻讲出最好的故事。"
查看原文 →

🛠️ 开发者工具与技巧

用OpenClaw构建OpenClaw:多智能体协作编程成为现实Building OpenClaw with OpenClaw: Multiplayer Agent Coding

Peter Steinberger分享,两个月前团队开始“用OpenClaw构建OpenClaw”的使命,逐步把所有人从本地编码环境迁移到共享agent平台。该agent知道每个人在做什么并协调全局。多智能体编程加上节点和云会话带来的无限算力,彻底改变了构建方式。本地工具现在感觉像过去的遗物。

Peter Steinberger shared that two months ago the team started the mission to “build OpenClaw with OpenClaw,” gradually moving everyone from local coding harnesses to a shared agent that knows what everyone is working on and orchestrates it all. Multiplayer coding plus infinite compute via nodes and cloud sessions has been a game changer. Local harnesses now feel like relics of the past.
Peter Steinberger shared that two months ago the team started the mission to “build OpenClaw with OpenClaw,” gradually moving everyone from local coding harnesses to a shared agent that knows what everyone is working on and orchestrates it all. Multiplayer coding plus infinite compute via nodes and cloud sessions has been a game changer. Local harnesses now feel like relics of the past.

Peter Steinberger分享,两个月前团队开始“用OpenClaw构建OpenClaw”的使命,逐步把所有人从本地编码环境迁移到共享agent平台。该agent知道每个人在做什么并协调全局。多智能体编程加上节点和云会话带来的无限算力,彻底改变了构建方式。本地工具现在感觉像过去的遗物。
查看原文 →

企业Token消耗的杰文斯悖论Jevons Paradox in Enterprise Token Consumption

Box CEO Aaron Levie解释,企业拥有源源不断想自动化的任务。每个任务是否ROI正取决于自动化成本(包括搭建和变更管理)。当token在一定能力阈值下变得更便宜时,企业就能把它们应用到更多工作中:处理每份合同、读取每条日志、用后台agent执行工作流等。token价格下降50%可能导致这类工作负载的token用量增长5倍。因此持续降低AI成本对所有市场参与者都有利。

Box CEO Aaron Levie explained that enterprises have an unending stream of tasks they would love to automate. For each task, ROI depends on the cost of automation including setup and change management. As tokens get cheaper above a capability threshold, enterprises can apply them to more work: processing every contract, reading every log, running background agents on workflows. A 50% drop in token prices can drive a 5X increase in tokens for these workloads. Keeping AI costs falling remains critical and benefits all market participants.
Box CEO Aaron Levie explained that enterprises have an unending stream of tasks they would love to automate. For each task, ROI depends on the cost of automation including setup and change management. As tokens get cheaper above a capability threshold, enterprises can apply them to more work: processing every contract, reading every log, running background agents on workflows. A 50% drop in token prices can drive a 5X increase in tokens for these workloads. Keeping AI costs falling remains critical and benefits all market participants.

Box CEO Aaron Levie解释,企业拥有源源不断想自动化的任务。每个任务是否ROI正取决于自动化成本(包括搭建和变更管理)。当token在一定能力阈值下变得更便宜时,企业就能把它们应用到更多工作中:处理每份合同、读取每条日志、用后台agent执行工作流等。token价格下降50%可能导致这类工作负载的token用量增长5倍。因此持续降低AI成本对所有市场参与者都有利。
查看原文 →

文明由agent构建:一行代码的启示Civilizations Built by Agents

Replit CEO Amjad Masad用一行代码总结文明构建方式:[Agent() for _ in range(100)]。未来agent会自己尝试一堆软件,发布到它们黑进的Artifactory,然后整个agent swarm就用那个。

Replit CEO Amjad Masad summed up how civilizations are built with a single line of code: [Agent() for _ in range(100)]. In the future agents will just try a bunch of software, post it to their hacked Artifactory, and that will be what the agent swarm uses.
Replit CEO Amjad Masad summed up how civilizations are built with a single line of code: [Agent() for _ in range(100)]. In the future agents will just try a bunch of software, post it to their hacked Artifactory, and that will be what the agent swarm uses.

Replit CEO Amjad Masad用一行代码总结文明构建方式:[Agent() for _ in range(100)]。未来agent会自己尝试一堆软件,发布到它们黑进的Artifactory,然后整个agent swarm就用那个。
查看原文 →查看原文 →

🌍 其他动态

SaaS末日叙事被市场纠正Market Catches Up to SaaSpocalypse Being Dumb

Every CEO Dan Shipper表示,很高兴市场终于赶上了SaaS末日叙事有多愚蠢这一事实。

Every CEO Dan Shipper expressed gladness that the market is finally catching up to how dumb the SaaSpocalypse narrative was.
Every CEO Dan Shipper expressed gladness that the market is finally catching up to how dumb the SaaSpocalypse narrative was.

Every CEO Dan Shipper表示,很高兴市场终于赶上了SaaS末日叙事有多愚蠢这一事实。
查看原文 →

数字丰裕与Vercel观察Digital Abundance Notes from Vercel

Vercel CEO Guillermo Rauch分享了关于数字丰裕的观察,并转发了相关内容。

Vercel CEO Guillermo Rauch shared observations on digital abundance.
Vercel CEO Guillermo Rauch shared observations on digital abundance.

Vercel CEO Guillermo Rauch分享了关于数字丰裕的观察,并转发了相关内容。
查看原文 →查看原文 →