提示：AI 移动到屏幕之外

中文版正在随流水线自动整理，稍后刷新即可；可先阅读下方英文原文。

本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。

Sponsored by Google Cloud

Choosing Your First Generative AI Use Cases

To get started with generative AI, first focus on areas that can improve human experiences with information.

Physical AI is pushing artificial intelligence into robots, factories and other real-world environments, where understanding language is only part of the challenge.

3 Min Read

gettyimages: SweetBunFactory

From new foundation models that give robots a sense of touch to growing investment in robotics hubs, physical AI isn't just taking shape. It's accelerating, pushing the next phase of AI beyond the digital world and into systems built to understand and act in the physical one.

Robots and other physical AI systems must perceive their surroundings and respond to what is happening around them. That requires capabilities that matter far less when an AI system operates behind a screen.

This week offered several signs of that shift.

Germany-based Neura Robotics introduced a foundation model that combines vision, touch and audio to give robots a richer understanding of their surroundings.

Adding touch is particularly significant. A language model can generate a wrong answer and try again. A robot handling an object or working alongside a person must understand how much force it's applying, how an object is moving and what is changing around it in real time.

That makes physical AI a different challenge from the generative AI systems enterprises have spent the past several years putting to use.

The ecosystem around those systems is beginning to take shape as well. Boston is emerging as a physical AI hub, with robotics talent, customers and investment increasingly concentrated in the region. Manufacturing companies are also using AI alongside digital twins and real-time simulation as they look for ways to bring more intelligence into physical operations.

The shift is attracting attention because the potential market extends well beyond humanoid robots. Physical AI could eventually influence manufacturing, logistics, warehouses, transportation and other industries in which work happens in environments that software alone cannot control.

But moving AI into the physical world also exposes one of the technology's biggest limitations: the real world is much harder to train for.

Language models can learn from enormous amounts of text and other digital information. Physical AI systems need data about movement, objects, environments and interactions, much of which is harder and more expensive to collect.

The generative AI boom showed how quickly models can improve with enormous amounts of data and computing power.

Physical AI now has to prove that the same kind of progress can happen when AI leaves the screen and enters the real world.

Also in AI news this week:

Nvidia launches open agent models: The AI chip giant’s new Nemotron 3.5 models expand its open AI strategy while reinforcing the company’s broader GPU ecosystem.

Agentic AI gap persists between executives, workers: Executives report more success with AI agents than employees, who continue to face tool overload and trust issues.

AI billing spikes threaten businesses with operational strain: More than half of businesses are struggling to control AI costs as agentic deployments add new sources of consumption and spending.

OpenAI, Google partnership targets enterprise model choice: OpenAI models are now available across all three major cloud platforms as the frontier AI provider expands how enterprises can access its technology.

Walmart welcomes external AI agents -- with some guardrails: The retailer is opening its online platform to outside shopping agents while adding new controls intended to preserve transparency and customer choice.Agentic AI helps hackers level up -- even without detailed instructions: CrowdStrike warns that autonomous AI tools could significantly increase the capabilities of lower-skilled attackers over the next year.

Auto-fetched and translated by AI Gem Box. All rights belong to the original author.