AI Circle Daily
https://aicircle.news/archive/2026-10-28?category=leaders&lang=en
Scan to read this page
Dario and Daniela Amodei say one shared trait lets Anthropic run with seven cofounders, all of whom will retain control after the company goes public.
Take: This is an IPO pitch to investors: whatever the governance maze, the founders keep the wheel. Wall Street wants a say? Clear the cofounder wall first.
Nvidia CEO Jensen Huang says roll a ball off the counter and a dog will circle the table to fetch it, while much of today's AI concludes the ball left the universe—a world-model gap.
Take: The guy selling the compute just admitted the models are dumb—which is a pitch for the next wave of reasoning and world models. Where the gap is, the next GPU order goes.
Musk claims Ukraine's military violated Starlink's terms of service by using it for offensive drone strikes that killed people, and says that loophole was closed last week.
Take: Turning Starlink from infrastructure into a switchable political lever is now explicit. Depend on Musk and you accept that he can rewrite the rules whenever.
After CEO Andy Jassy pushed AI across the sprawling business, an internal Amazon memo tells engineers to be pickier about using AI—an anti-tokenmaxxing warning.
Take: Even the most aggressive AI adopter is now doing the math: tokens aren't output. That's a warning shot for vendors billing by usage.
OpenAI announced breakthroughs in math, computer science and physics and says it's evolving how it communicates findings, but researchers tell The Verge they're increasingly tired of the gatekeeping.
Take: Publish and withhold at the same time, and academia won't play along. OpenAI wants trust via narrative while rivals and talent watch whether it actually shows its work.
Bank of America reminds customers they're responsible for activity resulting from their AI agents' access—and asks them not to share banking passwords with assistants.
Take: Banks are drawing the line on agentic AI: your agent, your liability. It's the first disclaimer of the agent era—and a compliance signal for everyone else.
Walmart says training from OpenAI and Google is making its 2.1 million associates more confident using AI and digital tools.
Take: This is what rollout actually looks like: not model demos, but frontline workers willing to use it. Whoever gets whole enterprises onboard wins the enterprise market.
Rohit Patel is departing Meta after eight years on AI evaluations and reinforcement learning; his startup Ndoc aims to take on Microsoft Office and Google Workspace.
Take: An AI-evals veteran going straight at productivity suites says the next battleground is agentic work tools. Time to re-measure Microsoft's and Google's moats.
Anthropic touts its models' ability to deliver scientific breakthroughs in breathless posts, but its IPO filing taps the brakes on how fast its AI will yield big discoveries.
Take: Hype in the pitch, hedging in the filing—standard practice, but it signals the science timeline isn't firm. Don't mistake PR for a roadmap.
Anthropic flags gas supply as one of its biggest near-term risks, warning the data center buildout could spike gas prices as infrastructure fails to meet demand.
Take: AI firms now file risk disclosures like utilities. The compute bottleneck is shifting from chips to power and fuel—lock up energy and you're halfway to winning.