Technology

➚ MOVE FAST: Jackpot. Intel’s haul from U.S. grants, loans and tax breaks could grow to almost $45 billion. It already nabbed nearly $20 billion from CHIPS Act funding that was announced today as the company seeks to restore its manufacturing edge in America.

➘ BREAK THINGS: Stir the pot. The Biden administration isn’t done yet blacklisting firms linked to Huawei, which produced a smartphone made with advanced chips despite tough export controls. Now the U.S. is mulling targeting Chinese chip makers as the tech war heats up.

At a panel I moderated at SXSW on sports and technology last week, I asked Kara Nortman, owner of the Angel City Football Club, whether AI could help with strategy on the pitch. Nortman, a former tech venture capitalist, said it would be unlikely, citing football’s amorphous nature that doesn’t produce a lot of ingestible data.

Yesterday, Google DeepMind reached out about an upcoming paper on … using AI to help with football (soccer to Americans) strategy. DeepMind worked with Liverpool to develop an AI model that actually helped with corner kick strategy.

Sports teams have for decades used machine learning and big data to strategize, both on personnel decisions and strategy. But the applications were usually clear cut, relying on structured data: Player X is more likely to have a higher batting average than Player Y. Or, the probability of winning by going for it on the 4th down (in American football) changes by X percent.

We’re now entering a new era, where the advice given by computers looks more like the advice a human coach might give to players. The latest software can look at unstructured data, like video footage of corner kicks, and gain new insights. Professional sports and other games are great proving grounds for new forms of AI. It would not be surprising if teams are already experimenting with these tools in secret, hoping to gain a competitive advantage.

The making of AI empires

THE SCENE

In the land grab around artificial intelligence, big technology companies are relying on old playbooks to cement and enhance their lead, which could draw greater scrutiny from Washington and Brussels.

On Monday, Bloomberg reported that Apple and Google are in talks about installing Google’s AI models on iPhones. A similar arrangement for Google search is part of a Justice Department antitrust lawsuit against Sundar Pichai’s company.

On Tuesday, Microsoft announced it was hiring Inflection AI co-founder and CEO Mustafa Suleyman and most of the company’s staff. The software giant will now offer Inflection AI models to its cloud customers and Suleyman will spearhead consumer AI products like Microsoft Copilot.

Inflection, which took a sizable investment from Microsoft, had been one of the few independent foundation model providers.

The European Commission has said it is probing Microsoft’s investment in OpenAI as a possible merger in disguise. That came after Microsoft offered to hire the entire OpenAI staff following CEO Sam Altman’s sudden firing late last year but the effort was dropped after he returned to the company.

Now, Microsoft-backed Inflection has defected to the company.

REED’S VIEW

AI empires have been busy this week. Microsoft’s move to simply absorb a $4 billion AI company was an incredible display of power and underscores how effective CEO Satya Nadella has been in the AI era.

Meanwhile, an Apple deal with Google would change the competitive dynamics in the AI industry in Google’s favor and could rescue Apple, which appears to have fallen behind on the technology, from the AI abyss.

There are two ways of looking at these big news stories this week. One is that the tech giants are flexing their muscles and will continue to dominate the AI era.

Only companies with massive war chests can afford to train foundation models, the most powerful AI tools that do everything from generate text to images to video.

With a first-mover advantage, foundation model providers have already gained invaluable data and know-how, propelling them far ahead and crowding out startups and competitors.

Another way of looking at this is that a massive wave of disruption is coming and the big tech companies are playing the same game in a world that’s about to be vastly different.

The original Apple-Google search deal was hashed out during secretive dinners between Apple CEO Tim Cook and Pichai, like two feudal lords trading tracts of land.

Even the creation of Inflection, a startup based on cutting edge technology, followed the familiar Silicon Valley tradition of “no conflict, no interest.”

Microsoft was the biggest investor in Inflection, which raised over $1 billion from people like Microsoft co-founder Bill Gates. Reid Hoffman, Inflection’s co-founder, is on Microsoft’s board. Inflection’s new CEO, Sean White, is a longtime friend of Hoffman’s and has worked at Hoffman’s venture firm, Greylock.

These companies know how to win today. They may not be set up for victory tomorrow.

There’s talk of Nvidia CEO Jensen Huang being “Swiftified” this week after a packed house watched him unveil the company’s new Blackwell platform, a line of AI chips that blows previous benchmarks out of the water.

Yes, Huang is a rock star. But among game developers, which are hosting a conference now in San Francisco, Huang has been a huge celebrity for many years.

Nvidia’s new, most powerful chip contains 208 billion transistors. The Apollo guidance system used in the space program had 12,300 transistors. The story of AI has really been one of massive increases in compute power and Nvidia is leading that charge.

Huang said Nvidia needed to bend the laws of physics to build this chip. Soon, we’ll start to see what impact that has on the capabilities of AI models.

The U.S. Attorney’s office in San Francisco is putting startups on notice. Ismail Ramsey, the top federal prosecutor for California’s Northern District, told Reuters that his office is targeting AI ventures that mislead investors before they go public. “As with any such emerging technologies, AI is fertile ground for fraudsters to make false and exaggerated claims,” he said.

His comments come after the Securities and Exchange Commission slapped a total of $400,000 in penalties on two investment advisers for their AI false claims. Agency Chair Gary Gensler called their misleading marketing “AI washing.” Adding to the regulatory pile-on is the Federal Trade Commission, which is investigating Reddit’s $60 million deal to license its content to train other firms’ AI models.

The Pentagon’s R&D arm is ready to share its tools for countering deepfakes, which use techniques to verify manipulated images, video, and text. As part of the rollout, DARPA is calling on partners in academia and the private sector to commercialize its technology, it announced last week. In a bid to speed up that effort, DARPA released a catalog of open-source technologies for detecting deepfakes, and is pushing an effort to create machine learning models to detect AI-generated images.

One technique DARPA developed detects AI-generated videos. “Anyone who has an online presence up above about an hour of video is vulnerable to deepfakes in terms of video manipulation,” Dr. Wil Corvey, DARPA’s Semantic Forensics program manager, told Semafor. But by making a biometric model of facial movements and head tilts, DARPA has been able to identify manipulated videos. In tandem with other tools, such as tests that detect a heartbeat in a video, the likelihood of ferreting out AI-generated content improves, Corvey said.

Although DARPA focuses more on fundamental research than developing tools for the Pentagon, the ones being built by Semantic Forensics could be of use to the U.S. intelligence community. The Pentagon has already had to deal with viral fake images showing an explosion at the building, while experts have warned that deepfakes may become good enough to make life difficult even for military analysts. “We have benefited very much from sustained engagement from the DoD, and in some cases from prototyping very closely with them,” Corvey said.

— Mathias Hammer

• Finland is the world’s happiest country as US rank drops.
• The senators who could fill Trump’s national security cabinet.
• Biden admin awards $8.5B to Intel for advanced chips.