TypeSafe's Jev outputs decisions not text; new NBER research confirms AI is raising structural unemployment; Git 3.0 to drop SHA-1 for SHA-256 and require Rust

Engineering

Simon Willison

Cloudflare Python Workers Are Now Generally Available

Cloudflare released Python as a general availability language on Workers after a two-year preview. Python code compiles to WebAssembly via Pyodide and runs in the V8-based workerd runtime. The pywrangler CLI tool supports local development and Cloudflare invested in the Pyodide ecosystem.

Read →

pepsipu

Stack Unwinding Can Lead to Leakless Code Execution

Thread cancellation in glibc unwinds the stack and executes DWARF bytecode, which an attacker can redirect to execute arbitrary code. The technique uses memory corruption to control fake pthread structures with malicious DWARF expressions. This approach achieves code execution without address space layout randomization (ASLR) leaks.

Read →

remy.wang

Relation Algebra is Not Relational Algebra

Relational algebra and relation algebra are different concepts with similar names. Relational algebra is the mathematical base for database systems since Edgar Codd created it in 1970. Relation algebra provides a way to work with binary relations in mathematical logic and formal specification.

Read →

LWN

Looking Forward to Git 2.56 — and 3.0

Git 2.56 introduces over 700 commits with features like git history drop and git refs subcommands. Git 3.0 brings breaking changes: migration from SHA-1 to SHA-256, a Rust build requirement, and reftable storage for efficient ref handling.

Read →

Marginal Revolution

Banning Self-Recursive Improvement in AI Models?

Tyler Cowen argues that proposals to ban self-recursive improvement in AI models are impractical to implement. The policy creates perverse incentives that push AI development to other nations and reduce American competitiveness. He warns that U.S. labs would lose knowledge about model safety while competitors advance without restrictions.

Read →

Marginal Revolution

Further Results on AI and Labor Market Reallocation

Researchers studied how large language models (LLMs) affect U.S. labor markets. Workers exposed to AI find new jobs and switch jobs less than other workers. Companies reorganize tasks more within existing jobs. AI produced smaller wage gains after LLMs started. Job-finding became harder since 2023 due to AI effects, and the natural unemployment rate rose 0.1 to 0.2 percentage points.

Read →

Crooked Timber

Into the Darkness?

Philosopher Chris Bertram argues that AI and digital surveillance create a totalitarian system from which escape is impossible. Citizens sacrifice privacy for convenience and lose individual freedom and the chance to resist. Bertram fears that authoritarian governments will use AI to detect dissent and control populations permanently.

Read →

Humanities

JSTOR Daily

Ancient Rome's Complicated Border

Hadrian's Wall served as a complex border zone in northern England for nearly 300 years starting in 122 CE. The wall did not mark a simple divider but created a military zone that changed settlement patterns on both sides. Borders function by both dividing and connecting, not just separating spaces.

Read →

Aeon

Toadstools and Toxins

A public health scientist investigated amanita muscaria gummies sold in shops. These products contain muscimol, which causes sedation and confusion, and lack proper safety tests or labels. Dosing guidance came from a 1967 experiment. After the investigation, authorities seized more than 95,000 products.

Read →

Jev: Decision Models Are a New Shape of LLM