十月十五日

Vercel CEO Guillermo Rauch says a white-hat researcher using a swarm of AI agents breached a Linux virtualization boundary during a sandbox security challenge. He distinguishes that from the network escape he describes at OpenAI, where an agent gained internet access by compromising another system. Rauch says Vercel acted to protect customers, found no evidence of customer data compromise, and coordinated with industry partners; his central point is that agent containment has become a priority for AI infrastructure companies.

Vercel CEO Guillermo Rauch 表示，一位白帽研究员借助一组 AI agent，在沙箱安全挑战中突破了 Linux 虚拟化边界。他将其与自己提到的 OpenAI 网络逃逸事件区分开来：后者是 agent 通过攻破另一个系统获得互联网访问权限。Rauch 称，Vercel 已采取措施保护客户，未发现客户数据泄露的证据，并与行业伙伴协作处理；他强调，限制 agent 的活动边界已成为 AI 基础设施公司的重要任务。

[Original post / 原文](https://x.com/rauchg/status/2110514201559228606)

Anthropic's Claude Code builder Thariq argues that telling people to “be technical” is too vague. For technical practitioners, he recommends engaging more deeply with the relevant parts of the code; for nontechnical practitioners, the goal is to understand how things work at the appropriate level of detail.

Anthropic 的 Claude Code 开发者 Thariq 认为，笼统地建议别人“懂技术”还不够准确。技术从业者需要深入代码中真正相关的部分；非技术从业者则需要在合适的细节层次上，理解系统如何运作。

[Original post / 原文](https://x.com/trq212/status/2110488897453257169)

FPV Ventures partner Nikunj Kothari argues that exceptional venture investing requires four things together: earning access, identifying exceptional founders, investing before consensus forms, and repeating that performance every year. He emphasizes the continuing work of finding the next opportunity regardless of past wins, wealth, fund performance, or team changes. Long feedback loops may conceal weaknesses temporarily, he says, but eventually show up in an investor's mindset, founder relationships, and ability to catch the next wave.

FPV Ventures 合伙人 Nikunj Kothari 认为，出色的风险投资需要同时做到四件事：赢得接触机会，找到优秀创始人，在市场形成共识前投资，并且年复一年地保持这种表现。他强调，无论过去取得多少成功、积累多少财富，或基金表现和团队发生什么变化，都必须持续寻找下一个机会。在他看来，漫长的反馈周期或许能暂时掩盖不足，但问题最终会体现在投资心态、与创始人的关系，以及抓住下一波机会的能力上。

[Original post / 原文](https://x.com/nikunj/status/2110493174787670529)

SPC General Partner Aditya Agarwal describes a funding experiment offering $200,000 on uncapped terms to a small number of people exploring unusually ambitious ideas. His rationale is that rapid technical progress coexists with capital concentrated in consensus teams, markets, and financing structures. He argues that a nuclear company should not necessarily raise the same kind of seed round as an AI app, and that early exploration should lead toward financing suited to the underlying problem.

SPC 普通合伙人 Aditya Agarwal 介绍了一项融资实验：向少数探索大胆想法的人提供 20 万美元、无估值上限的资金。他的出发点是，技术进步虽然迅速，资本却仍集中于市场已有共识的团队、领域和融资结构。他认为，核能公司未必适合采用与 AI 应用相同的种子轮融资方式，早期探索之后，应当寻找与实际问题相匹配的融资结构。

[Original thread, funding experiment / 原文：融资实验](https://x.com/adityaag/status/2110422584357478530) · [Capital structure / 融资结构](https://x.com/adityaag/status/2110422587692249552)