The City of Munich has hired libexpat’s maintainer for a six‑month open‑source sabbatical, paying him to fix five known security flaws, add XML 1.0 r5 support and harden the parser. It’s the first municipal‑level funding scheme aimed at sustaining critical, under‑funded digital infrastructure, showing cities can directly back essential open‑source tools.
After years of daily essays, Gwern announced he’s stepping back from full‑time writing to focus on Guardian Angel, a startup building highly personalized LLM “digital twins” for each user. He argues these bespoke agents can boost productivity while defending against AI‑driven attacks, positioning the venture at the intersection of AI safety and knowledge‑worker tooling.
Waymo’s autonomous ride‑hailing service, previously limited to a Dallas test group, is now public via the Waymo app. The rollout follows 150k rides since February and adds airport terminal testing, promising full freeway service soon, opening a new, driverless mobility option for commuters, tourists, and those with medical‑driving restrictions.
The Rust core team has released a formal policy that dictates how contributors may use large language models when submitting code, reviews, or issues to the rust-lang/rust repository. By publicly defining acceptable LLM practices, Rust aims to preserve reviewer bandwidth, maintain code quality, and give newcomers clear expectations, one of the first language projects to embed AI rules into its governance.
The blog distills practical rules, like making it easy to do business, treating customers as partners, designing systems, and avoiding over‑leverage, that help leaders deliver reliable service and earn trust. Applying them can reduce churn, boost team morale, and make scaling smoother.
The essay debunks the myth that founders need a troubled past to stay hungry, arguing that curiosity, high standards, and supportive families can generate the same relentless drive. Borrowing a Formula 1 metaphor, it calls this "mode push", a constant tinkering gear that fuels innovation beyond trauma.
A study of 60 language-model benchmarks finds almost half have hit saturation, with older tests worsening the effect. Expert-curated benchmarks resist this decay, while public test sets do not. The result warns that headline scores may no longer reflect real-world capability, urging new evaluation designs.
A research‑backed ACM Queue article dismantles eight popular myths about generative AI in software engineering, from the claim that developers spend most of their day coding to the belief that AI automatically boosts productivity. It shows why line‑of‑code metrics mislead, how trust and workflow changes matter, and what leaders should really measure.
Flowise, the open‑source low‑code LLM app builder, announced it will wind down operations, ceasing feature development and archiving its GitHub repo by August 31 2026. The decision cites a market shift toward advanced coding agents that outstrip low‑code workflows, underscoring growing consolidation pressure in the AI tooling space.
Pi’s deliberately tiny toolset, four built‑in functions and sub‑1k token prompts, lets it out‑perform larger agents. In a Databricks benchmark on a multi‑million‑line codebase, Pi achieved the highest pass rate while cutting task cost by more than 2×, proving that lean harnesses win in enterprise AI economics.
Governor Greg Abbott ordered the Public Utility Commission and ERCOT to vet all upcoming data center projects after the interconnection queue ballooned to 474 GW, 90% of which are data centers. The move aims to prevent grid overload and could end Texas’s data‑center boom.
You’re caught up.
That was all 11 stories in the 5 August issue.