Ai

The FT just published an investigation into a tool called Heretic. It’s free, sits on GitHub, requires no specialist hardware, and strips the safety guardrails off open-weight AI models in under ten minutes. Since its release late last year, it has been used to produce more than 3,500 decensored models, downloaded over 13 million times. The creator stripped safeguards from Google’s Gemma 4 within 90 minutes of its release. ...

Pope Leo XIV’s first encyclical, Magnifica Humanitas, was released on May 25. It is entirely about artificial intelligence. The Pope issues encyclicals rarely, and the choice of subject for a first one is a deliberate signal. Dedicating it to AI tells you how seriously the Catholic Church is taking what is coming. Leo frames AI as the Industrial Revolution of our time, with the same potential to reshape labour, dignity, and the structure of society. The encyclical leans on two biblical metaphors: the Tower of Babel for unchecked technological ambition, and the rebuilding of Jerusalem’s walls for collaborative human effort. Its core position is that technology is never neutral. It carries the values of whoever builds, funds, and deploys it. ...

There is a quiet but very expensive bet being made across the AI compute layer right now, and I think it deserves more scrutiny than it is getting. The bet is that the best way to handle the growing demand for AI compute is to build silicon shaped around the architecture we have today. In practice, that means chips increasingly tuned for transformers. Etched is the clearest example, with hardware designed explicitly around transformer workloads. But the broader pattern shows up across the industry too: more memory bandwidth tuned for attention, more matrix throughput tuned for the operations LLMs actually use, and more interconnect tuned for the shapes of current models. ...

Mira Murati’s Thinking Machines Lab has been quiet since launch, and this is their first real public release. They call it interaction models, and the idea is to design the model around how humans actually collaborate rather than around how long an agent can run alone. Audio, video, text all flowing continuously instead of turn-by-turn. 200ms time-aligned micro-turns, concurrent input and output, a separate background reasoning model for the heavier thinking. ...

Anthropic just announced a new enterprise services company, backed by Blackstone, Hellman & Friedman, Goldman Sachs, and a long list of other private equity heavyweights including General Atlantic, Leonard Green, Apollo, GIC, and Sequoia. The company will staff Applied AI engineers from Anthropic and target mid-sized manufacturers, community banks, and regional health systems. The pitch is simple: small teams working hands-on with customers to find high-impact use cases and ship Claude-powered solutions into existing workflows. ...

Mayo Clinic just published a validation study on a model called REDMOD, the Radiomics-based Early Detection Model, that spots pancreatic cancer on routine abdominal CT scans up to three years before clinical diagnosis. On the validation set, the AI caught 73% of cancer cases. Specialist radiologists looking at the same scans caught 39%. The median lead time was about 16 months, with some catches happening 475 days before symptoms. ...

Tesla just tripled its annual capital expenditure to $25 billion in 2026, up from $8.5 billion last year. The money is going into AI infrastructure, chip design, semiconductor research, battery technology, and Optimus robot production. The Fremont factory, which used to make Model S and Model X, is being converted to manufacture Optimus at scale. A dedicated Optimus plant is also being built near Austin. To put the number in context, Musk noted Amazon is spending $200 billion in capex this year and Google $175 billion. The scale gap is real, but the direction is the same. ...

I came across this infographic and spent more time than I expected just reading through it. It’s a good snapshot of how much the anatomy of a server rack has shifted in the last few years. A rack used to be mostly about compute and storage. CPUs on top, drives somewhere in the middle, some networking at the top, and air blowing through the whole thing. The job of the infrastructure was to stay out of the way of the workload. Now the workload is the infrastructure. GPUs are the centre of gravity, and everything else, power distribution, cooling, interconnects, cable management, is designed around keeping them fed and cold enough to run flat out. ...

Robotics labs have spent years recording humans doing physical tasks to teach their systems when and how to grab, walk, or stack boxes. Meta just brought that playbook to software and computer use, except the demo subjects are its own employees. The program is called Model Capability Initiative. It captures mouse movements, clicks, and keystrokes across hundreds of apps and websites including Google, LinkedIn, GitHub, and Slack. No opt-out. The data goes straight into training AI agents that can perform knowledge work autonomously. ...

Moonshot AI just open-sourced K2.6, and the benchmarks are hard to ignore. It beats or matches GPT-5.4, Opus 4.6, and Gemini 3.1 Pro on Humanity’s Last Exam with tools and SWE-Bench Pro, which are two of the more credible tests for reasoning and coding. It can run for 12 hours straight across 4,000 tool calls. One internal agent apparently ran autonomously for five days. And it can spin up 300 parallel sub-agents at the same time. ...