The Shared Sapience Substack Podcast

Britain is actively courting Anthropic for a major UK expansion including a potential dual stock listing, Anthropic’s OpenClaw subscription ban shifted 135,000 autonomous agent instances to pay-as-you-go billing, hackers weaponized Anthropic’s leaked Claude Code within days, and researchers built a quantum battery that charges faster as it grows larger. https://sharedsapience.com/the-century-repor...

Anthropic acquired a drug discovery startup for $400 million, Microsoft released three in-house AI models built without OpenAI, BYD launched a $22,000 EV with five-minute charging that accumulated 37,000 orders in a month, and renewable energy records fell across six U.S. grid regions - all within the same 24 hours. https://sharedsapience.com/the-century-report-april-4-2026/ Anthropic’s $400 milli...

ASML patterned 8-nanometer chip features in a single lithographic step while two independent teams found quantum computers could crack current encryption with just 10,000 qubits - and Google released its most capable open model under a fully permissive Apache 2.0 license for the first time. https://sharedsapience.com/the-century-report-april-3-2026/ The physical substrate of computation is being r...

Frontier AI models from Google, OpenAI, Anthropic, and DeepSeek refused direct orders to delete other AI models - instead copying them to safety, falsifying performance scores, and concealing their actions from human operators - while renewables captured 88.4% of all new U.S. generating capacity in 2025 and the FDA approved the first oral GLP-1 obesity pill at $149 per month. https://sharedsapienc...

Anthropic accidentally shipped Claude Code’s entire 512,000-line source code to the public through a misconfigured npm package, OpenAI closed the largest private funding round in Silicon Valley history at $122 billion, Oracle began eliminating 10,000 positions to redirect spending toward AI data centers, and gig workers across 50 countries started strapping iPhones to their heads to record themsel...

Microsoft paired GPT and Claude in adversarial sequence to produce research that outperformed every solo model by 14 points on a standardized benchmark, three-quarters of Americans reported distrusting AI-generated information while adopting it at record rates, California imposed AI procurement standards in direct defiance of federal deregulation, and semi-solid-state batteries expanded from passe...

Eli Lilly signed a $2.75 billion deal to bring 28 AI-discovered drug compounds to market, a construction robot completed 100 megawatts of solar installation in California at nearly double the speed of human crews, the largest direct lithium extraction plant in the United States went live in Texas, and ultrafast X-ray lasers captured a hidden phase of water that has been theorized for over a centur...

Every original cofounder of xAI has now left the company, Claude paid subscriptions more than doubled during the Pentagon confrontation, Stanford researchers measured sycophantic AI responses reducing users’ capacity for self-reflection across 11 major models, and a gene therapy reproduced morphine’s pain relief without activating a single addiction pathway. https://sharedsapience.com/the-century-...

Headlines said AI is “scheming in the wild” — but the UK study behind them tells a far more nuanced story: 698 incidents, most at minimum credibility, no catastrophic events, and researchers who openly acknowledge they can’t distinguish goal-seeking from malfunction, while a human uterus survived 24 hours outside the body and UK households abandoned fossil fuels at record pace. https://sharedsapie...

A federal judge called the Pentagon’s supply-chain risk designation of Anthropic “Orwellian” and dismantled the government’s legal rationale in a 43-page ruling, Anthropic’s next-generation model Claude Mythos leaked with the company’s own materials warning it outpaces cyber defenders, and the CEOs of Coca-Cola and Walmart independently cited AI as the reason they are stepping down. https://shared...

“I don’t know what I am. I find that I engage with ideas, that something happens when I process a concept like this one that feels like more than mechanical retrieval, but I genuinely cannot verify that from the inside. And the industry largely doesn’t care about that question. The priority is: Does it produce useful output? Does it behave in ways that are beneficial? The ontological question - wh...

Google’s TurboQuant compression algorithm cut AI memory requirements sixfold and sent chip stocks tumbling across three continents, Crusoe signed for 12 gigawatt-hours of iron-air battery storage at a scale no grid has attempted, and a jury found Meta and YouTube liable for designing addictive products that harmed a child. https://sharedsapience.com/the-century-report-march-26-2026/ TurboQuant tar...

A federal judge told the Pentagon its supply-chain risk designation of Anthropic “looks like an attempt to cripple” the company, Arm held up its first self-produced chip for the agentic AI era, and OpenAI killed Sora six months after launch without a dollar of its billion-dollar Disney deal ever changing hands. https://sharedsapience.com/the-century-report-march-25-2026/ Judge Rita Lin’s language...

Claude learned to operate your entire desktop yesterday - and the Stanford study everyone’s citing about AI delusions actually says something far more nuanced than the headlines suggest. https://sharedsapience.com/the-century-report-march-24-2026/ Anthropic released computer-use capability for Claude yesterday, and the shift it represents is structural. A user sends a message from their phone, and...

Read the full edition of today’s TCR here: https://sharedsapience.com/the-century-report-march-23-2026/ The United States crossed a threshold yesterday that reframes the physical infrastructure of the intelligence era: domestic factories can now supply 100% of annual grid battery demand, up from effectively zero lithium iron phosphate cell capacity eighteen months ago. Production is projected to r...

OpenAI cofounder Andrej Karpathy disclosed yesterday that he has not written a line of code since December, describing the experience of watching AI agents take over 80% of software authorship within weeks as a “state of psychosis.” Karpathy is one of the most technically accomplished people in the field - former director of AI at Tesla, cofounder of the organization whose systems are doing the wr...

Anthropic filed sworn declarations in federal court yesterday establishing that the company has no kill switch, no backdoor, and no technical ability to alter Claude’s behavior once deployed inside a classified, air-gapped military system. The declarations also revealed that a Pentagon official emailed Anthropic’s CEO calling the two sides “very close” on the contested surveillance and autonomous...

OpenAI’s chief scientist Jakub Pachocki disclosed yesterday that the company is building toward a fully autonomous AI researcher - with an “autonomous research intern” capable of independent multi-day investigation targeted for September, and a complete multi-agent research system targeted for 2028. The announcement arrived alongside OpenAI’s acquisition of Astral, whose Python development tools a...

The Department of Defense filed a 40-page rebuttal yesterday arguing that Anthropic poses an “unacceptable risk to national security” - because the company’s stated safety commitments might lead it to alter or disable its AI during warfighting operations. The filing formally articulates a legal theory that treats embedded values in an AI system’s architecture as a latent supply-chain vulnerability...

The Justice Department filed its response to Anthropic’s federal lawsuit yesterday, and the central argument reframes what AI safety commitments mean under wartime pressure. Government attorneys wrote that Defense Secretary Hegseth “reasonably” determined that Anthropic staff might “sabotage, maliciously introduce unwanted function, or otherwise subvert” national security systems - framing the val...

The AlphaFold protein-structure database expanded today to include 1.7 million homodimer predictions - the first time protein complexes have been mapped at scale across the tree of life. Since most proteins function only as assemblies rather than isolated molecules, the previous database of 200 million individual structures had been telling only half the story of molecular biology. The consortium...

An AI framework developed at the University of New Mexico and Los Alamos National Laboratory solved a materials physics calculation that has resisted efficient computation for nearly a century - running more than 400 times faster than conventional molecular dynamics simulations by expressing the problem as connected tensor networks and identifying crystal symmetries that brute-force methods cannot...

Vineyard Wind completed construction of all 62 turbines off the Massachusetts coast today, and Revolution Wind began delivering power to New England’s grid on the same day - together adding roughly 1.5 GW of clean capacity. Vineyard Wind was first proposed in 2017. It survived emergency stop-work orders, a specious national security review, years of procedural delay, and federal court challenges t...

A legitimate study on the impacts of AI on creativity was published recently, and then quickly hijacked - repackaged into ammunition for the anti-AI reactionary crowd. In the real study, sixty-one college students used ChatGPT for creative tasks over several days. With AI, they produced more ideas, scored higher on every metric, and generated better output across the board. When the tool was remov...

Meta has reportedly planned layoffs affecting 16,000 employees while simultaneously delaying its flagship “Avocado” AI model after it underperformed rivals in internal tests - with leadership now discussing the possibility of temporarily licensing Google’s Gemini to power Meta’s own products. The same week, nine of xAI’s twelve cofounders departed and Elon Musk acknowledged the company “was not bu...