This startup thinks robotics is about to have its ChatGPT moment

这家初创公司认为机器人技术即将迎来属于它的“ChatGPT时刻”

Before OpenAI’s GPT-3 ushered in the era of foundation models, companies built specialized natural language processing models from scratch, training each on large amounts of task-specific data. Today, most organizations start with a general-purpose model like OpenAI’s GPT series, Claude, or Llama and then fine-tune or prompt it to solve their specific needs. 在 OpenAI 的 GPT-3 开启基础模型时代之前，企业通常从零开始构建专门的自然语言处理模型，并使用大量特定任务数据对每个模型进行训练。如今，大多数机构会从 OpenAI 的 GPT 系列、Claude 或 Llama 等通用模型入手，然后通过微调或提示（prompting）来解决其特定需求。

Pim de Witte, CEO of General Intuition, thinks embodied AI will follow a similar pattern. Rather than collecting huge real-world datasets to build specialized robot models, he argues the industry should focus on better quality datasets that can produce foundation models capable of transferring intuition about movement and interaction across many environments. General Intuition 的首席执行官 Pim de Witte 认为，具身智能（embodied AI）也将遵循类似的模式。他认为，与其收集海量的现实世界数据集来构建专门的机器人模型，行业更应专注于更高质量的数据集，从而产出能够跨多种环境迁移运动和交互直觉的基础模型。

“A lot of companies right now are doing lots of specialized work focused on individual embodiments, individual environments, and individual robots,” de Witte told TechCrunch on a recent episode of Equity. Much of that work will become redundant soon, he argues, with the emergence of general models like the one General Intuition has been developing and deploying. “目前许多公司都在进行大量专门的工作，专注于特定的具身形态、特定的环境和特定的机器人，” de Witte 在最近一期的《Equity》节目中告诉 TechCrunch。他认为，随着像 General Intuition 正在开发和部署的这类通用模型的出现，其中大部分工作很快就会变得多余。

“The generalization of the model itself is the product,” he said. “The fact that it has a base level of reasoning about space and time is going to be the reason why people stop collecting hundreds of thousands or millions of hours of real-world data. Because the reality is, you only need a few minutes.” “模型本身的泛化能力就是产品，” 他说。“它具备了对空间和时间的基本推理能力，这将成为人们停止收集数十万甚至数百万小时现实世界数据的原因。因为事实是，你只需要几分钟的数据就够了。”

General Intuition built its own such foundation model after training on millions of hours of video game data, including information like what buttons on a controller a human pushed and when. Both de Witte and General Intuition’s lead investor, Vinod Khosla, argue the action data is the key to developing a human-like intuition for spatial-temporal reasoning. General Intuition 在训练了数百万小时的视频游戏数据后，构建了自己的基础模型，其中包括人类在何时按下了控制器上哪些按钮等信息。de Witte 和 General Intuition 的领投方 Vinod Khosla 都认为，这些动作数据是培养类似人类的时空推理直觉的关键。

The startup last month raised $320 million at a $2.3 billion valuation on the back of that thesis. The company has demonstrated that its current model is capable of both playing a video game for hours and powering a quadrupedal robot — the latter after fine-tuning it on just eight minutes of real-world robotics data. 凭借这一理论，这家初创公司上个月以 23 亿美元的估值筹集了 3.2 亿美元。该公司已经证明，其当前模型既能连续数小时玩视频游戏，也能驱动四足机器人——后者仅在经过八分钟的现实世界机器人数据微调后即可实现。

“The fact that [the robot] was actually able to zero-shot on just the front camera, with no other sensors, in the office with dynamic objects being introduced and people walking by was a very big surprise to us,” de Witte says. “I think it’s a sign of what’s to come.” “事实上，（机器人）仅凭前置摄像头，在没有其他传感器的情况下，就能在有动态物体进入和行人经过的办公室环境中实现零样本（zero-shot）操作，这让我们感到非常惊讶，” de Witte 说。“我认为这是未来趋势的一个信号。”

The end game for General Intuition isn’t to build robots itself, but to become the foundation model of physical AI, a base model for other robotics companies to build upon for their own machines. Or, as de Witte put it: “We’re not gonna build a self-driving car company. We’re gonna make it 10 times easier for the next person to build a self-driving car company.” General Intuition 的最终目标不是自己制造机器人，而是成为物理 AI 的基础模型，为其他机器人公司提供可供其机器构建的底层模型。或者，正如 de Witte 所言：“我们不会去建立一家自动驾驶汽车公司。我们要让下一个人建立自动驾驶汽车公司的难度降低 10 倍。”