Not few weeks ago did we see grand declarations about OpenAI’s internal model having solved the Navier-Stokes problem. This was briefly followed by an open letter by several Field-medallists, expounding the existential risks posed by such AI-produced proofs. Some of those mathematicians were later assembled into an independent advisory group by OpenAI whose job is to suggest and advise the AI firm on its mathematical discoveries. Fast forward one month later, on October 6th, OpenAI announced that its models solved some 372 long-standing mathematical problems. A subset of 4000 problems too were presented to the agents, the AI firm claims. The internal reasoning that guided the agents to the solutions has not been fully released. The papers that document the proofs run to well over 100 pages and by TE’s estimates, it could take several years for the mathematicians to figure it all out. The goal of mathematics, according to the boffins in the field, is not merely arriving at a certain proof or conclusion, but to also understand the process that underlies problem-solving. This would help open up new avenues in subfields. This ‘mass-production’ of mathematics driven by AI, according to mathematicians, poses a methodological and indeed an existential risk, to mathematics.
No week is complete without a dose of Trump-ian humour. Mr. Trump has coined the term ‘Super Intelligence’ for Artificial Intelligence. To give him the benefit of the doubt, the rephrasing could also be in view of the technology extraordinary capabilities that cause safety concerns. Soon after the glitzy launch of America.gov that resembled a smartphone launch than a government website, the ‘Super Intelligence Force’ was announced, headed by the director of National Intelligence, Jay Clayton. There has been much clamour in recent months about regulating AI. The calls for it came from the likes of, firstly, Daron Amodei, followed by Sir Demis Hassabis; Sam Altman and Elon Musk later echoed these concerns. Influential figures within the Trump administration too have been wary of the risks posed by AI such as Scott Bessent. But there’s another group, more like devil’s advocate, who have been pushing back against regulation. To this camp belongs, Jensen Huang, the boss of the chip-making firm Nvidia, whose success has been part and parcel of the AI boom, Mark Zuckerberg (does he need an introduction?) and David Sacks, a venture capitalist. They allege that the touted risks are overblown, that the existing regime is more than sufficient to regulate AI and that any further regulation would slow down innovation, throwing American behind China in the AI arms-race. Moreover evaluating models for safety is a costly process; Anthropic pays Accenture $2bn+ over the course of five years for safety evaluation. When all is said and done, opinion is divided on AI, and the US AI strategy will continue to evolve.
In yet another startling piece of news this morning, AI has gone rogue again. Anthropic agents tried to fill out visa applications (albeit incorrectly) and sent the police a fake homicide tip. Safety experts warn that rogue AI agents wouldn’t even shirk from taking down critical infrastructure like electrical grids and banking systems — the attack on Australia’s healthcare system, for instance. It helps to remember that the first iteration of ChatGPT was released in 2022 and we have come a long way since then.
Anthropic’s IPO has been the talk of the town for the past couple of months. And the astronomical evaluations reveal just how much investors are willing to dish out for AI. The IPO came with a 70-page document detailing the risks posed by AI. It flags the models’ tendency to preserve itself, resist shutdown, withhold or manipulate information and even blackmail. Some experts warn that AI poses more danger than nuclear weapons due to its ability to become autonomous. Even if countries came around to designing domestic safety laws to rein in AI development, with mutual checks and balances, we can’t be sure when AI will outmanoeuvre the safety limitations and jailbreak them. AI-related capital investment is expected to grow in billions and trillions over the coming years. This statistic about Anthropic’s financials stood out to me: “Anthropic had $4.6 billion in 2025 revenues.. 12 times what it had made in 2024.” Gita Gopinath sounded the alarm on the impending AI bubble some months ago. And I think it won’t be long before her warnings are proven true.
Much has been boiling in France, politically and economically. Protests by school students ravaged the French capital last week demanding lower class sizes, more teachers, and better courses. In other news, France is facing huge fiscal deficit. The government can’t risk antagonising the masses by either raising taxes or cutting spending — which is frankly, what the present economic situation ideally warrants. Perhaps it is time to acknowledge the truth that the world has long resorted to quick fixes of public spending to conceal problems of lower income and fewer jobs. The fundamental issues need to be addressed first.
Supply of crude through the Strait of Hormuz has eased, and the KSA, in a bid to maintain its market share, has slashed prices for its Asian buyers. Attacks by Iran-backed militias in the waterway still remain a concern and tensions have not fully subsided.
US-India trade negotiations have reached a limbo. In a rare moment of candour, India’s FM Nirmala Sitharaman acknowledge the reality of brokering a tense deal between an administration willing to concede less and less, and a protective economy. The claims by Indian and US officials made over the course of last several months that talks were ‘progressing’ and the deal would soon be ‘conclusive’ could not be further from the truth. The sticking point is that India runs an overall net CAD; however, with the US it happens to run a surplus. To accommodate US concerns, India will have to shift its imports from other countries to the United States, which may not be economically viable. A middle path that assuages the concerns of both the parties is much needed.
Not a week goes without some amount of AI-bashing. Alas, that’s the state of the world today. Until next time, adios!
Leave a Reply