Hey and welcome to Eye on AI…On this version: my chat with AI chief Eric Xing…Trump’s AI export plan…drama on the Worldwide Math Olympiad…Stargate replace…transparency in reasoning.
I used to be excited and curious to fulfill Eric Xing final week in Vancouver, the place I used to be attending the Worldwide Convention on Machine Studying—one of many prime AI analysis gatherings of the 12 months. Why? Xing, a longtime Carnegie Mellon professor who moved to Abu Dhabi in 2020 to steer the general public, state-funded Mohamed bin Zayed College of Synthetic Intelligence (MBZUAI), sits on the crossroads of almost each large query in AI at this time: analysis, geopolitics, even philosophy.
The UAE, in spite of everything, has quietly develop into probably the most intriguing gamers within the international AI race. The tiny Gulf state is aligning itself with U.S.-style norms round mental freedom and open analysis—even because the AI rivalry between the U.S. and China turns into more and more outlined by closed ecosystems and strategic competitors. The UAE isn’t making an attempt to “win” the AI race, nevertheless it desires a seat on the desk. Between MBZUAI and G42–its state-backed AI-focused conglomerate–the UAE is constructing AI infrastructure, investing in expertise, and aggressively positioning itself as a go-to associate for American companies like OpenAI and Oracle. And Xing is on the coronary heart of it.
Because it occurred, Xing and I simply missed one another—he arrived in Vancouver as I used to be heading house—so we related on Zoom the next day. Our dialog ranged extensively, from the hype round “world fashions” to how the UAE is utilizing open-source AI analysis as a strategic lever to construct mushy energy. Listed below are just a few of probably the most compelling takeaways:
A ‘Bell Labs plus a college‘
MBZUAI is simply 5 years previous, however Xing says it’s already among the many fastest-growing educational establishments on this planet. The varsity, which is generally a graduate program for AI researchers, aspires to compete with elite establishments like MIT and Carnegie Mellon whereas additionally taking up utilized analysis challenges. Xing calls it a hybrid group, much like “Bell Labs plus a college,” referring to the legendary R&D arm of AT&T, based in 1925 and accountable for foundational improvements that formed trendy computing, communications, and physics.
The UAE as a soft-power AI ambassador
Xing sees MBZUAI not simply as a college, however as a part of the UAE’s broader effort to construct mushy energy in AI. He describes the nation as a “sturdy island” of U.S. alignment within the Center East, and views the college as an “ambassador middle” for American-style analysis norms: open supply, mental freedom, and scientific transparency. “If the U.S. desires to challenge affect in AI, it wants establishments like this,” he informed me. “In any other case, different nations will step in and outline the path.”
The U.S. isn’t dropping the AI race
Whereas a lot of the general public narrative round AI focuses on a U.S.-China race, Xing doesn’t purchase the framing. “There is no such thing as a AI struggle,” he stated flatly. “The U.S. is approach forward in concepts, in folks, and within the innovation setting.” In his view, China’s AI ecosystem remains to be constrained by censorship, {hardware} limitations, and a weaker bottom-up innovation tradition. “Many prime AI engineers within the U.S. could also be of Chinese language origin,” he stated, “however they solely turned prime engineers after finding out and dealing within the U.S.”
Why open supply issues
For Xing, open supply isn’t only a philosophical choice—it’s a strategic alternative. At MBZUAI, he’s pushing for open analysis and open-source AI improvement as a option to democratize entry to cutting-edge instruments, particularly for nations and researchers exterior the U.S.-China energy facilities. “Open supply applies strain on closed techniques,” he informed me. “With out it, fewer folks would have the ability to construct with—and even perceive—these applied sciences.” At a time when a lot of AI is turning into siloed behind company partitions, Xing sees MBZUAI’s open method as a option to foster international expertise, advance scientific understanding, and construct credibility for the UAE as a hub for accountable AI improvement.
On ‘world fashions’ and AI hype
Xing didn’t maintain again when it got here to one of many buzziest traits in AI proper now: so-called “world fashions”—techniques that intention to assist AI brokers study by simulating how the world works. He’s skeptical of the hype. “Proper now individuals are constructing fairly video mills and calling them world fashions,” he stated. “That’s not reasoning. That’s not simulation.” In a current paper he spent months writing himself—uncommon for somebody of his seniority—he argues that true world fashions ought to transcend flashy visuals. They need to assist AI cause about trigger and impact, not simply predict the following body of a video. In different phrases: AI wants to know the world, not simply mimic it.
With that, right here’s the remainder of the AI information—together with that tomorrow the White Home is about to launch a sweeping new AI technique aimed toward boosting the worldwide export of U.S. AI applied sciences whereas cracking down on state-level rules which might be seen as overly restrictive. I might be attending the D.C. occasion, which features a keynote by President Trump, and can report again.
Sharon Goldmansharon.goldman@fortune.com@sharongoldman
AI IN THE NEWS
White Home to unveil plan to push international export of U.S. AI and crack down on restrictions. In line with a draft seen by Reuters, the White Home is about to launch a sweeping new AI technique Wednesday aimed toward boosting the worldwide export of U.S. AI applied sciences whereas cracking down on state-level rules seen as overly restrictive. The plan will bar federal AI funding from states with powerful AI legal guidelines, promote open-source and open-weight AI improvement, and direct the Commerce Division to steer abroad information middle and deployment efforts. It additionally duties the FCC with reviewing potential conflicts between federal objectives and native guidelines. Framed as a push to make “America the world capital in synthetic intelligence,” the plan displays President Trump’s January directive and might be unveiled throughout a “Successful the AI Race” occasion co-hosted by the All-In podcast and that includes White Home AI czar David Sacks.
OpenAI and Google DeepMind sparked math drama. Over the previous few days, each OpenAI and Google DeepMind claimed their AI fashions had achieved gold-medal-level efficiency on the 2025 Worldwide Mathematical Olympiad—efficiently fixing 5 out of 6 notoriously troublesome issues. It was a milestone that many thought of years away: a basic reasoning LLM reaching that degree of efficiency underneath the identical cut-off dates as people, with out instruments. However the best way they introduced it sparked controversy. OpenAI launched its outcomes first, based mostly by itself analysis utilizing IMO-style questions and human graders—earlier than any official verification. That prompted criticism from outstanding mathematicians, together with Terence Tao, who questioned whether or not the issues had been altered or simplified. In distinction, Google entered the competitors formally, waited for the IMO’s impartial evaluate, and solely then declared its Gemini DeepThinker mannequin had earned a gold medal—making it the primary AI system to be formally acknowledged by the IMO as acting at that degree. The drama laid naked the excessive stakes—and differing requirements—for credibility within the AI race.
SoftBank and OpenAI are reportedly struggling to get $500 Billion Stargate AI Venture off the bottom. In line with the Wall Road Journal, the $500 billion Stargate challenge—introduced with fanfare on the White Home six months in the past by Masayoshi Son, Sam Altman, and President Trump—has hit main turbulence. Billed as a moonshot to supercharge U.S. AI infrastructure, the initiative has but to interrupt floor on a single information middle, and inside disagreements between SoftBank and OpenAI over key phrases like website location have delayed progress. Regardless of guarantees to take a position $100 billion “instantly,” Stargate is now aiming for a scaled-down launch: a single, small facility, possible in Ohio, by 12 months’s finish. It’s a setback for Son, who lately dedicated a record-breaking $30 billion to OpenAI however remains to be scrambling to safe a significant foothold within the AI arms race. Nevertheless, Bloomberg reported at this time that Oracle will present OpenAI with 2 million new AI chips that might be a part of a large information middle enlargement that OpenAI labeled as a part of its Stargate challenge. SoftBank, although, isn’t financing any of the brand new capability—and it is unclear what operator might be growing information facilities to help the brand new capability, and when they are going to be constructed.
EYE ON AI RESEARCH
Sounding the alarm on rising opacity of superior AI reasoning fashions. Fortune reporter Beatrice Nolan reported this week on a bunch of 40 AI researchers, together with contributors from OpenAI, Google DeepMind, Meta, and Anthropic, which might be sounding the alarm on the rising opacity of superior AI reasoning fashions. In a brand new paper, the authors urge builders to prioritize analysis into “chain-of-thought” (CoT) processes, which give a uncommon window into how AI techniques make choices. They’re warning that as fashions develop into extra superior, this visibility might vanish.
The “chain-of-thought” course of, which is seen in reasoning fashions corresponding to OpenAI’s o1 and DeepSeek’s R1, permits customers and researchers to watch an AI mannequin’s “pondering” or “reasoning” course of, illustrating the way it decides on an motion or reply and offering a sure transparency into the inside workings of superior fashions.
The researchers stated that permitting these AI techniques to “‘suppose’ in human language gives a singular alternative for AI security,” as they are often monitored for the “intent to misbehave.” Nevertheless, they warn that there’s “no assure that the present diploma of visibility will persist” as fashions proceed to advance.
The paper highlights that specialists don’t absolutely perceive why these fashions use CoT or how lengthy they’ll maintain doing so. The authors urged AI builders to maintain a better watch on chain-of-thought reasoning, suggesting its traceability might ultimately function a built-in security mechanism.
FORTUNE ON AI
Mark Cuban says the AI struggle ‘will get ugly’ and mental property ‘is KING’ within the AI world —by Sydney Lake
$61.5 billion tech large Anthropic has made a significant hiring U-turn—now, it’s letting job candidates use AI months after banning it from the interview course of —by Emma Burleigh
Skilled software program builders assumed AI would save them a bit of time. However in a single experiment, their duties took 20% longer —by Sasha Rogelberg
AI CALENDAR
July 26-28: World Synthetic Intelligence Convention (WAIC), Shanghai.
Sept. 8-10: Fortune Brainstorm Tech, Park Metropolis, Utah. Apply to attend right here.
Oct. 6-10: World AI Week, Amsterdam
Oct. 21-22: TedAI San Francisco.
Dec. 2-7: NeurIPS, San Diego
Dec. 8-9: Fortune Brainstorm AI San Francisco. Apply to attend right here.
