The Verge spoke to Martin Hairer, one of the nine mathematicians on the new Advisory Group on Mathematics and Artificial Intelligence, as OpenAI prepares to release more than 100 results from an unreleased model. We set out who sits on AGMAI, what OpenAI and the group have each promised, why the math community feels bulldozed, and what a responsible release process would look like.
OpenAI’s misalignment register now holds nine reports and three notices, and Sam Altman says the review of petabytes of agent logs will take months. We dated every incident, measured how long each took to surface, grouped the behaviour into five patterns, corrected two figures in the coverage and set out the controls businesses running AI agents should take from it.
Florida Attorney General James Uthmeier has asked a Highlands County court to stop OpenAI developing new AI models without independent third-party approval, bar minors from ChatGPT and strip its human-like language. We set out the six requested restraints, the case history, the incidents and insider statements the 49-page motion relies on, the legal test, OpenAI’s response and what it means for businesses.
Anthropic has released Claude Sonnet 5.5, which it says is more than 30% faster than Sonnet 5 and up to 30% cheaper per task. The token price has not changed, so the saving comes from fewer tokens and tool calls. We work through the cost arithmetic, the benchmark table against Opus 5.5 and GPT-6 Sol, the new cyber and distillation safeguards, and whether to switch.
A TU Wien-led paper in Joule says AI data centers must adapt to the grid, citing a US study where sites needed to cut demand for no more than 35 hours a year. We trace the figure to its source, add the 32 generation hours it leaves out, and explain the batteries, costs and connection rules behind it.
Instinct has raised $1 billion from Sequoia, Benchmark and Coatue at a $10 billion valuation, 33 days after its Series B. We rebuild the funding timeline, work through the arithmetic of a 200-fold markup, and look at compute, Meta’s Muse, privacy and how the free agent might make money.
A Cambridge startup advised by Yann LeCun says it has licensed almost 1 million hours of gameplay to train AI world models. We explain why labs want video game data, what Minecraft and Bleeding Edge research shows, who else is betting on it, and why Nvidia thinks game physics falls short.
OpenAI’s brute-force Navier-Stokes proof has mathematicians asking what AI is doing to a field that has always worked like an art form. We trace the argument from G. H. Hardy to the Erdős problems, check the numbers behind the declaration and the compute bill, and draw lessons for any expert workforce.
WIRED warns that hundreds of thousands of AI agents are about to join the workforce as named coworkers. We check the BCG research behind its org-chart statistic, including the hot potato effect that weakens oversight when AI is called an employee, and set out the governance every organisation needs first.
Deel has opened Akai, the AI agent platform it built for its own finance, compliance and HR teams, to other companies. We explain how Akai learns a workflow, where it uses rules versus AI judgment, which use cases it targets, how it compares with RPA, and what to ask before buying.
Progressive Robot: Your Gateway to Comprehensive IT Solutions — Specializing in Web Development, Mobile App Development, and Expert IT Services.
© All Copyright by Progressiverobot.com
VAT Number ( 506152326 )
