以下是将原文翻译为中文： Social Radars 播客的杰西卡·利文斯顿 (Jessica Livingston) 和卡罗琳·利维 (Carolyn Levy) 采访了 Scale AI 的创始人兼首席执行官亚历山大·王 (Alexander Wang)，该公司为人工智能公司提供高质量的数据。亚历山大分享了他的创业历程，从在新墨西哥州洛斯阿拉莫斯长大，周围都是物理学家，到建立一家价值 138 亿美元的公司。 亚历山大的童年充满了数学、物理和计算机科学的学术竞赛。在进入麻省理工学院之前，他休学一年，在 Quora 担任软件工程师。在那里，他开始理解数据对人工智能的重要性。他注意到像 DeepMind 的 AlphaGo 和 Google 的 TensorFlow 这样的进步，都严重依赖数据作为智能的原材料。 在麻省理工学院期间，他和联合创始人一起集思广益，提出了创业想法，包括 Scale AI 的最初概念。最初，他们带着一个医生预约应用程序的想法申请了 Y Combinator (YC)。杰西卡·利文斯顿回忆起她对亚历山大的最初印象，觉得他可能“傲慢或才华横溢”，但最终投票资助了他们。他承认，最初他们的团队在产品与市场契合度方面显得“迷茫”。 事实证明，医生预约应用程序是不可行的，这促使他们进行了业务转型。2016 年夏季左右，聊天机器人开发的激增表明，人工智能数据是一条有希望的道路。最初，人们认为聊天机器人时代而非自动驾驶汽车才是关键驱动因素。该公司最初名为“Ava”，专注于聊天机器人，但后来演变为“Scale”，因为它蕴含着基础设施和增长的含义。 亚历山大和他的团队最初亲自进行数据标注，早期的客户来自 YC，包括 Teespring。然后，他们转向专注于自动驾驶汽车。虽然许多人认为数据标注“枯燥乏味”，但亚历山大认识到它的重要性，并认为数据是人工智能的基础。早期相信人工智能的投资者也看到了 Scale 的潜力。 他回忆起一位投资者驳斥了对大量数据的需求，但后来证明这种观点是错误的。他们于 2020 年从自动驾驶汽车转向政府合同，与国防部 (DoD) 合作，将人工智能应用于国家安全。这包括图像识别模型，用于检测乌克兰的破坏情况，用于军事协调和人道主义援助工作。 2022 年，亚历山大看到了大型语言模型 (LLM) 和生成式人工智能的潜力，并将公司资源转移到支持这一新浪潮。在很短的时间内，Scale 超过一半的员工转向了生成式人工智能的数据工作。这得益于他“做得过多”而不是对重大市场变化反应不足的理念。 亚历山大还讨论了公司在多元化方面的立场，该立场通过他们的“MEI”（优点、卓越、智力）原则来表达。他认为，专注于精英管理，雇佣最优秀的人才，会导致多元化成为自然而然的结果。虽然这种立场引发了争议，但也吸引了那些欣赏公司对精英明确关注的人。 他承认，在疫情期间，公司从 150 人迅速增长到 700 人，这带来了挑战。他们撤销了远程办公政策，并正在迁回办公室中心。他认为，领导 Scale 的经历让他变得比实际年龄更加成熟。 亚历山大回应了保罗·格雷厄姆 (Paul Graham) 关于软件难以标记和难以生成的数据的问题：这是代理数据。代理数据捕捉了个人在执行任务时的整个思维过程和行动。关于使用标签数据最好的方法：如何让最杰出的专家为这些 AI 系统做出贡献并自动化它，以便它能够高效？亚历山大说：“当模型出现错误或出现幻觉，或者他们遇到困难，或者他们必须在现实世界中做出一些改变时，他看到人类在帮助机器。”亚历山大说，“胜出的最好方法是让你的公司成为你毕生的事业。” 他表示，他很幸运能够建立一家对人工智能行业的发展和未来至关重要的公司，并认为该行业充满激情，但观点各异。令人兴奋的是，有惊人的数字表明，价值 2000 亿美元的投资正在用于构建先进而强大的人工智能系统。Scale 和 Nvidia 在“幕后”支持着这个行业。

Jessica Livingston and Carolyn Levy of Social Radars podcast interview Alexander Wang, founder and CEO of Scale AI, a company providing high-quality data to AI companies. Alexander shares his journey from growing up in Los Alamos, New Mexico, surrounded by physicists, to building a $13.8 billion company. Alexander's childhood was marked by academic competitions in math, physics, and computer science. He took a gap year before attending MIT, working as a software engineer at Quora. There, he began to understand the importance of data for AI. He noticed how advancements like DeepMind's AlphaGo and Google's TensorFlow relied heavily on data as the raw material for intelligence. During his time at MIT, he brainstormed startup ideas with co-founders, including the initial concept for Scale AI. Initially, they applied to Y Combinator (YC) with a different idea for a doctor booking app. Jessica Livingston recalls her initial impression of Alexander as potentially "arrogant or brilliant," but ultimately voted to fund them. He admits that their team seemed "lost" about their product-market fit initially. The doctor booking app proved unfeasible, prompting a pivot. A surge in chatbot development around the summer of 2016 suggested data for AI was a promising path. Initially, the chatbot era, not autonomous vehicles, was believed to be the key driver. The company started under the name "Ava," focused on chatbots, but evolved to "Scale" for its implications of infrastructure and growth. Alexander and his team did the data labeling themselves initially, with early customers from YC, including Teespring. Then they pivoted to focus on autonomous vehicles. While many considered data labeling "unsexy," Alexander recognized its importance and saw that data was the foundation for AI. Early investors who believed in AI also saw the potential in Scale. He recalls an investor dismissing the need for large amounts of data, a notion later proven wrong. They shifted from autonomous vehicles to government contracts in 2020, working with the Department of Defense (DoD) on AI for national security. This included image recognition models to detect damage in Ukraine, used for both military coordination and humanitarian aid efforts. In 2022, Alexander saw the potential of large language models (LLMs) and generative AI, shifting company resources to support this new wave. Over a short period, more than half of Scale's headcount shifted to data for generative AI. This was driven by his philosophy of "doing too much" rather than underreacting to significant market shifts. Alexander also discusses the company's stance on diversity, expressed through their "MEI" (Merit, Excellence, Intelligence) principle. He believes that focusing on meritocracy, hiring the best talent, leads to diversity as a natural outcome. While this stance sparked controversy, it also attracted individuals who appreciated the company's clear focus on merit. He acknowledges the challenges of growing a company so quickly, from 150 to 700 people during the pandemic. They reversed remote work policy and are moving back to in-office hubs. He believes his experiences leading Scale have aged him beyond his years. Alexander addresses Paul Graham’s question on data that is hard for software to label and difficult data to generate: It is agent data. Agent data captures the entire thought process and actions individuals take while doing tasks. On the best things to do with labeling data: How do you get the most brilliant experts to contribute for these AI systems and automate it so it can be efficient? Alexander said, “He sees humans aiding the machines whenever the models are sort of like going down a wrong path or they’re hallucinating or there’s something that they’re getting stuck on or they have to make some change in the real world or something.” Alexander says, “Best way to out compete is if your company is your life’s work.” He states that he feels lucky to have built a company central to the AI industry's evolution and future and sees the industry as highly passionate but varied in perspectives. It is exciting that the numbers are staggering with $200 billion of investment are going into building advanced powerful AI systems. Scale and Nvidia are “behind the scenes” supporting the industry.