我立刻爱上了 @askalphaxiv —— 它已经成为我在最新机器学习研究“走向主流之前”理解前沿成果的首要来源。 但 AlphaXiv 远不止是一个被动阅读论文的网站。它正在把论文、基准测试、工具和工作流程整合在一起，真正弥合“AI 研究到实际应用”之间的鸿沟。非常兴奋能够支持他们，共同打造未来的研究工具与科研沟通方式。 I instantly fell in love with @askalphaxiv - it's become my primary source for understanding the latest in ML research before they hit the mainstream.But AlphaXiv goes far beyond a passive website to read papers. It's bringing together papers, benchmarks, tools and workflows to bridge the AI research-to-practice divide. Excited to support them as they build the future of research tooling & communication.
Gokul RajaramPartner, Marathon Management Partners
观点原文
我立刻爱上了 @askalphaxiv —— 它已经成为我在最新机器学习研究“走向主流之前”理解前沿成果的首要来源。
但 AlphaXiv 远不止是一个被动阅读论文的网站。它正在把论文、基准测试、工具和工作流程整合在一起，真正弥合“AI 研究到实际应用”之间的鸿沟。非常兴奋能够支持他们，共同打造未来的研究工具与科研沟通方式。
I instantly fell in love with
@askalphaxiv
- it's become my primary source for understanding the latest in ML research before they hit the mainstream.But AlphaXiv goes far beyond a passive website to read papers. It's bringing together papers, benchmarks, tools and workflows to bridge the AI research-to-practice divide. Excited to support them as they build the future of research tooling & communication.
AI
相关观点
- 真正重要的两件事 这些年来，我面试并评估过 1000 多名候选人。无论岗位职能、头衔或资历高低，有两项能力对所有角色都至关重要： 高能动性（High Agency）： 主动采取行动，而不是等别人告诉你该做什么。 解决问题的能力（Problem-solving Ability）： 能够诊断问题、找到根本原因，并基于第一性原理系统性地解决任何问题。 公司越年轻、团队越小，这两项能力就越重要。 如果你是在（此处填入某家巨型公司），没有这两项能力也许还能混得下去；但如果一家创业公司的前 50 名员工中有相当一部分缺乏这两项能力，那公司将会陷入巨大的痛苦之中。 创业公司应该在面试中分别设置环节来考察这两项能力。 至少，在最后一轮（由 CEO / 创始人亲自进行的面试）中，应该重点围绕这两点展开。 当我以投资人或董事会成员的身份面试创业公司候选人时， 我把大部分时间都花在测试他们的能动性和解决问题的能力上。 真正的挑战在于：创业公司极度需要员工和领导者具备能动性与解决问题的能力，而大公司既不筛选，也不奖励这两项能力。 这正是为什么——当一个来自大型公司的员工加入创业公司、却没有在这两个维度上被认真评估时，招聘失败往往就此埋下根源。 The two things that matter I’ve interviewed and evaluating 1000+ people over the years. The two attributes that are critical for every role (regardless of function, title, or seniority) are: 1. High Agency: being proactive and taking action, not waiting for someone to tell you what to do 2. Problem-solving Ability: being able to diagnose, root cause and tackle any problem systematically and from first principles The younger the company and the smaller the team, the more important these two attributes are. If you’re (insert massive company name), you can get away without hiring for these. But if a good chunk of your first 50 employees don’t have these, you will be in a world of pain. Startups should add interviews that separately test for each of these attributes. At the least, the final interview (with the startup CEO/founder) should focus on these two. When I interview candidates for startups as an investor or board member, I spend most of my time testing for agency and problem solving. The challenge is that startups need their employees and leaders to have agency and be able to solve problems, while big companies don’t screen for or reward either. This is the root of many hiring misfires when a big company hire joins a startup without being evaluated on these dimensions. Gokul Rajaram · 2026/01/17
- 创始人们：所有关于融资以及其他事情的“规则”，本质上都是人为设定的，注定是用来被打破的。 举个例子，常被引用的一条“规则”是：领投方必须至少出资整轮融资的 X%。这完全是胡扯。 比如，出色的 Kirsten Green 在 Bonobos 的 3000 万美元 C 轮融资中，仅出资 100 万美元 作为领投，同时还帮助创始人完成了剩余 2900 万美元 的融资。 忽略那些所谓的规则。为你的公司做正确的事。成为例外。 Founders: every rule (about fundraising and everything else) is artificial and meant to be broken. For example, consider the oft-cited “rule” that the lead investor should invest at least X% of the entire round. It’s BS. For example, the amazing Kirsten Green led Bonobos’ $30m Series C round with $1m (and helped the founder raise the remaining $29m). Ignore the so-called rules. Do the right thing for your company. Be the exception. Gokul Rajaram · 2026/01/14
- 完整的平台 给创始人的一句忠告：如果你不是平台，那么你现在是、或者迟早会变成，别人的平台上的一个功能。 要始终以打造平台为目标，去做端到端的完整产品，而不是只做其中的一小块。 请不断问自己两个问题： 我们是在真正解决一个完整问题，还是只是在修补别人不完整解决方案的某个“症状”？ 要做到我们“拥有整个工作流”，而不仅仅是其中一个环节，还缺什么？ 功能会被迅速商品化；平台会持续复利增长。 The Whole Platform Founders: If you're not the platform, you are (or will eventually become) a feature on someone else's platform. Always strive to build the platform, the whole product, end-to-end. Ask yourself: -- Are we solving the problem, or are we solving a symptom of someone else's incomplete solution? -- What would it take for us to own the entire workflow, versus just one part of it? Features get commoditized. Platforms compound. Gokul Rajaram · 2026/01/08
- Dashboard 还是 Pipes？ 每一家初创公司都必须做出一个选择：你的产品是一个 Dashboard 产品，还是一个 Pipes 产品？ 什么是 Dashboard 产品？ Dashboard 产品是由终端用户直接、频繁使用的产品，是他们完成任务的主要界面。这类产品的目标是：让用户“住”在产品里。 这类公司的核心北极星指标是活跃用户数（DAU / WAU / MAU，取决于该产品类别的自然使用频率）。 Facebook 的第一个产品（也就是 Facebook 本身）就是一个典型的 Dashboard 产品。 什么是 Pipes 产品？ Pipes 产品通常在后台运行，用于处理交易、数据、支付等基础能力。在初始配置完成后，客户几乎不再直接与它们交互。 这类产品的目标是：让客户尽可能多地把数据 / 支付 / 交易量“通过”它们跑。 它们的北极星指标是规模/流量型指标（例如 GPV，总支付额）。 Databricks 的核心产品就是一个 Pipes 产品。 一家公司可以同时拥有两种产品 一家公司可以同时拥有这两种类型的产品组合。 例如： ChatGPT 是一个 Dashboard 产品 OpenAI 的 API 是一个 Pipes 产品 但需要注意的是：每一个具体产品，都必须明确自己“主要”属于哪一类。 这个选择为什么重要？ 这个选择会直接决定： * 产品开发方向 * 增长策略 * 组织结构 举例来说： Dashboard 产品 * 需要大量投入在 UI / UX 打磨 * 强调参与感、使用频率、留存机制 * 增长团队更像消费级产品团队 * 目标是提升 DAU / WAU / MAU，培养使用习惯 Pipes 产品 * 更关注稳定性、吞吐量、可靠性 * 强调集成广度、嵌入客户工作流的顺滑程度 * 目标是成为“看不见但不可或缺”的基础设施 * 不断吞下客户越来越多的业务量 给创业 CEO 的一个问题 你清楚自己正在构建的是哪一种产品吗？如果清楚： * 你的北极星指标是否设置正确？ * 产品开发流程是否与之匹配？ * 组织结构是否围绕这一产品类型而设计？ PS：特别感谢 @MerrillLutsky 提醒我，我之前从未公开分享过这个产品框架。 Dashboard or Pipes? Every startup needs to make a choice: is their product a dashboard product or a pipes product? Dashboard products are used directly and regularly by end users as their primary interface for accomplishing tasks. The goal for these products is to get customers to live in the product. The primary North Star metric for these companies is active users (daily / weekly / monthly, depending on the natural frequency of customer usage for the category). Facebook’s first product (aka Facebook :)) was a dashboard product. Pipes products are used in the background to process transactions, data, payments, etc, and customers rarely interact with them directly after initial setup. The goal for these products is to for their customers to send as much of their data / payments / etc through them. Their North Star metrics is a volume metric (eg GPV). Databricks’ core product is a pipes product. Companies can have both types of products in their portfolio. For example, ChatGPT is a dashboard product while OpenAI’s APIs are a pipes product. However, a given product has to determine which camp it’s primarily in. This choice dictates product development, growth strategy, and org structure. For example, dashboard products require heavy investment in UI/UX polish, engagement features, and retention loops, while pipes products prioritize reliability, throughput, integration breadth, and seamless embedding into customer workflows. Dashboard products have consumer-style growth teams focused on activation and habit formation to grow [DWM]AUs, while pipes products focus on making their product invisible infrastructure that “just works” and on capturing more and more of their end customers’ volume. Startup CEOs: do you know what kind of product your company is building? If so, have you set up your North Star metric, your product development processes and your org structure accordingly? PS a big thank you to @MerrillLutsky for reminding me that I had never posted about this framework. Gokul Rajaram · 2025/12/21