Yu Deng and Terence Tao discussed the future of mathematics in the age of AI on the SAIR podcast. Deng said AI can expand mathematical boundaries and called for openness and open mathematical models; Tao noted that tech companies using AI for math research essentially distill others' results, requiring source traceability, and urged mathematicians to be involved in advance rather than cleaning up afterward.
In an interview on Blackstone's internal podcast, Masayoshi Son said AI could become 1,000 times smarter than humans within a decade, that everything he has done so far is just a warm-up, and that he opposes slowing AI development, believing technology itself is the solution to safety issues. He also introduced SoftBank's full-chain AI ecosystem investments, including LLMs, chips and data centers, and noted the company has begun using robots to produce robots.
In a podcast interview, OpenAI product chief Thibault Sottiaux said DevDay's persistent agent Dots are designed to work in the background and reduce screen checking; he expects humans to escape smartphone screen dependence within a year. He also stressed that agent utility depends on permissions, and OpenAI uses gradual authorization and human final review to avoid marginalizing people with full automation.
Liam Fedus, co-creator of ChatGPT and former OpenAI vice president, discussed AI and scientific discovery with host Brendan Foody on The Frontier podcast for about 50 minutes. He revisited details around ChatGPT's birth, explained why ChatGPT and AI coding broke through first, and argued why his startup Periodic Labs must build physical labs, noting that mathematics has Lean as a referee while experiments in the physical world are irreplaceable.
In an interview, Mr. P of P Equity Research said hyperscalers' 10-year demand visibility is unreliable, memory will remain cyclical, and estimated 2027 hyperscaler capex at $1.1T-$1.2T with memory at $500B-$700B. He identified power, advanced packaging and memory as compute bottlenecks, noted ABF substrate shortages may last into the 2030s, gas turbine orders are booked beyond 2030, and CPO dominance will wait until after 2030.
In a Sept. 15, 2026 podcast interview, Shopify CEO Tobi Lütke criticized most companies for misusing AI as a way to avoid thinking, pile on features and shift responsibility, and explained how Shopify's internal AI colleague River works. He stressed that machines cannot take responsibility and that human-in-the-loop decision-making is essential, adding that taste, intuition and judgment will become humanity's scarcest assets as AI becomes widespread.
OpenAI co-founder and Thinking Machines chief scientist John Schulman, along with Zyphra CTO Beren Millidge and Baseten model training lead Charlie O'Neill, predicted in a podcast that AI may surpass top human experts in 3 to 4 years. They also discussed the possibility of a 10x productivity boost for AI researchers within two years, challenges in goal-setting capabilities, and issues like catastrophic forgetting in model training.