Nvidia wants to stop accidents before they happen.

The chip giant filed a patent application for providing “proactive safety measures for robotics systems.” Nvidia’s tech uses AI to predict accidents related to autonomous systems and attempt to prevent them. 

Conventionally, safety measures involving robotics are taken reactively, Nvidia said. However, “the foreseeable complexity of upcoming tight cooperations between humans and intelligent machines requires a broader understanding of the context that reactive safety measures by themselves struggle to guarantee.” 

To take a more proactive approach, Nvidia’s tech flicks an automatic kill switch whenever it senses danger. The system gathers data from equipment in the environment to track the position of different objects or people and identify hazards. Using a neural network, that sensor data is analyzed to determine the likelihood of a potential accident in real time. 

Depending on how likely an event is to occur, Nvidia’s tech may trigger safety measures, such as activating a warning signal or shutting down equipment that workers are using. 

Nvidia’s tech could also be applied to autonomous vehicles by proactively warning drivers of potential incidents and making maneuvers to avoid them. 

With autonomous machines, it’s always better to be proactive than reactive, said Rhonda Dibachi, CEO at Manufacturing-as-a-Service company HeyScottie. On a factory floor, proactive safety could allow you to mitigate workplace incidents that could also impact operations. With a self-driving vehicle, the environment is far less controlled — and potentially even more dangerous. 

The difference between Nvidia’s tech and other autonomous safety methods, such as those used by Tesla, is “who is in control here,” said Dibachi. Conventional methods tend to put a machine’s sensors in control, but Nvidia’s patent relies on a “centrally-controlled system” to make safety decisions. 

Nvidia’s interest in this tech isn’t new. The company has filed plenty of patents for tech to improve both autonomous driving and robotics decision-making, safety, and reliability. 

But these niches are far from the company’s real winner: It continues to rake in billions from its chips business. These patents, Dibachi said, may aim to show off those chips. “This is a tried-and-true business strategy for technology companies,” she said. 

“They’re making money hand over fist doing what they’re doing,” said Dibachi. “Nvidia’s saying, ‘I’ve got this shiny new GPU,’ and people look at that and ask, ‘What can you do with it?’ This patent is answering that question.”

And even though Nvidia is an AI industry kingpin as it stands, that dominance could rely on keeping up with the Joneses — whether that be with chip innovations, software for autonomous machines, or developing its own language models.

Wells Fargo wants to make sure you can trust the cloud. 

The financial institution filed a patent application for a “cloud residual risk assessment tool.” Its filing describes a way to identify potential “residual risks” that a cloud environment may pose when migrating and hosting sensitive assets, such as certain data or software. 

Residual risk refers to any additional hazards that may occur after “risk management or control activities are accounted for,” Wells Fargo noted. 

“Corporations have a need for determining risks of the third-party host environment to guarantee that risks that may be harmful to assets of the current hosting environment are mitigated,” Wells Fargo said in the filing.

Wells Fargo’s tech evaluates the potential hazards of both the data storage methods a company is using and the future cloud environment by collecting information relating to “quantitative and qualitative” risk factors. 

The filing lists a number of factors that the tech considers, including policy information, unresolved issues, the sensitivity of the data it’s migrating, and even the capabilities of the cloud services organization itself (e.g., the IT specialists involved in operations). All of that information is then used to calculate the residual risk scores of both the current host and the future host. 

This tech could help Wells Fargo and other organizations decide if it’s worth the risk to switch to a certain cloud provider. 

There are a few places where risks could emerge in cloud services and data storage, said Trevor Morgan, senior vice president of operations at OpenDrives. Though cloud providers generally have strong security measures, it’s common that security holes get forgotten by the cloud tenants themselves. And even if those problems start small, they can grow riskier over time as more data is stored or more changes occur in the cloud environment, said Morgan. 

“The changing environment within the cloud means that things that you thought were secure may become less secure [over time],” said Morgan. “When you look at a lot of the breaches over the last three to five years that occur in the cloud, it has to do with things that were forgotten.” 

And given the sheer amount of data that financial firms like Wells Fargo handles regularly, tech like this has a clear use case for the company. Data leaks in this industry can be dangerous and costly, both for the companies involved and the users themselves. As bad actors continue to build taller ladders, cybersecurity walls need to grow to match them, Morgan said. 

“It makes sense with where the fintechs and the financial industry are going in general,” said Morgan. “All of their operations depend upon reducing risk.” 

This tech could allow the firm to “repatriate certain types of data” for safety, using machine learning instead of human workers to cost-effectively do the heavy lifting, said Morgan. “These financial firms, they’re all becoming technology companies to control their own environments and therefore try to reduce their risk profile.”

Adobe wants to make it easier to get in customers’ faces. 

The company filed a patent application for “content distribution based on a user journey” using machine learning. This tech uses generative AI to help someone take a content distribution campaign, such as a marketing campaign, from start to finish with just a prompt. 

“Identifying relevant data for a user journey and then planning the user journey based on the identified data is both time-intensive and labor-intensive,” Adobe said in the filing. “Both users and content providers benefit when an intent, stage, and context of users are understood and a digital experience is tailored for the users.”

After getting a prompt from the user, Adobe’s system uses a generative machine learning model to create a “user journey.” That path is based on specific “touchpoints,” such as an email or social media post. 

Based on that predicted journey, Adobe’s tech generates content to present to the user at different stages. The tech also makes several versions of this user journey and the content that goes along with it, and scores each one based on predicted effectiveness. 

The tech in this patent fits squarely into the work Adobe has been doing for the past year and a half: embedding AI into creative tools in a way that professionals can actually use. The company’s patent history is littered with AI inventions in this vein, including generative image editing, video creation, and data visualization tools.

Publicly, the company has continued to integrate AI throughout its suite of creative tools, along with building out Firefly, its generative AI tool that can create, manipulate, and edit images (and soon, perhaps, video). 

But using AI, especially generative AI, in commercial and marketing contexts can be tricky. There are risks of bias and hallucination when using generative AI in any context — something that Adobe has sought to fix in previous patents. 

And training an AI model to do its job requires a lot of data that can come from a variety of sources, and oftentimes is scraped from the internet. Because of this, content generation models are facing copyright backlash, with Meta, Stability AI, and Midjourney all facing legal battles. 

This element is where Adobe may see a leg-up: The company has claimed that Firefly’s services are safe for commercial use. It also updated its terms of use in June to clarify that it will not train AI on user content after backlash due to privacy issues.