today's signal · no scroll
the brief
A busy mix of real releases and grounded takes: DeepMind’s AlphaProof Nexus pushes agentic theorem proving, while MiniCPM5-1B goes fully open, and Next.js tweaks its caching optimizer in the 16.3 canary. On the ground, engineers are slimming builds with Rust+WASM, wiring agents via GitHub labels, and a kernel CVE from Apple warrants quick patching. Cultural currents even reached the Vatican, with clear-eyed analysis of the new AI encyclical.
Apple shipped a fix for a macOS kernel vulnerability reportedly found via Claude; update fleets promptly and audit exposure windows for affected devices.
CVE-2026-28952: Apple macOS 26.5 Kernel Vuln found by Claude — Article URL: https://support.apple.com/en-us/127115 Comments URL: https://news.ycombinator.com/item?id=48273169 Points: 85 # Comments: 29
The canary replaces next-ppr-optimizer with next-cache-components-optimizer, improves dev-loop login persistence, and fixes the errors.json merge driver—signals of continued cache-component maturation ahead of 16.3.
v16.3.0-canary.29 — Misc Changes Replace next-ppr-optimizer with next-cache-components-optimizer: #94035 [Skills] next-dev-loop: persist login state across loops: #94105 Fix errors.json merge driver to handle gaps in error codes: #93829 Credits Huge thanks to @gaojude and @sokra for helping!
A 1B-parameter model releases weights, training data, and deployment code; it tops Artificial Analysis under 2B and outperforms Qwen3.5-2B with half the parameters.
MiniCPM5-1B is now fully open source, including weights, training data, and deployment code. 🚀1B params, #1 on Artificial Analysis among all open models under 2B (17.9 pts). 🤖 modelscope.cn/models/OpenBMB… Beats Qwen3.5-2B (16.3) at half the parameters. Outperforms Qwen3.5-0.8B and pic.x.com/aQdzQWlBoK
A new release demonstrates multilingual zero-shot voice cloning across 30 languages—promising for cross-lingual TTS agents, localization pipelines, and rapid voice prototyping.
Zero-shot voice cloning for 30 languages github.com/sunnyxrxrx/X-V… pic.x.com/6XoJ4zTjTy
An agentic, Gemini-powered formal proof system reportedly solved nine open Erdős problems, underscoring LLM+search methods for verifiable math and scalable theorem discovery.
AI agents are advancing research-level math. 🚀 I’m thrilled to share @GoogleDeepMind’s AlphaProof Nexus - an agentic framework for formal proof search powered by Gemini. When applied to a set of open formal math problems, our agent autonomously solved: ✅ 9 open Erdős pic.x.com/XrzsI8vUki
An open synthetic multilingual, multimodal, multitask OCR dataset with 1M+ document images across 22 languages—useful for pretraining, benchmarking, and document AI evaluation.
Launching NayanaOCR Corpus 👉🏼 1M+ Document images across 22 languages Largest open source synthetic > multilingual > multimodal > multitask document corpus pic.x.com/01zvjzESLe
OpenClaw removed Sharp/Jimp in favor of photon (Rust+WASM), cutting the image stack from 140MB to 2MB—an actionable pattern for leaner, safer build pipelines.
OpenClaw's dependency purge continues. Killed Sharp and Jimp. Replaced it with photon, a small WebAssembly that runs compiled Rust for image processing. 2MB vs 140MB. github.com/silvia-odwyer/…
Simon Willison distills the Pope’s AI encyclical into clear, practical deployment ethics—a rare policy document with concrete guidance worth bookmarking by practitioners.
Notes on Pope Leo XIV's encyclical on AI — <p>Dropped this morning by the Vatican: <a href="https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html">Magnifica Humanitas of His Holiness Pope Leo XIV on Safeguarding the Human Person in the Time of Artificial Intelligence</a>. This is a <em>very interesting</em> document. It's some of the clearest writing I've seen on the ethics of integrating AI into modern society.</p> <p>Pope Leo XIV chose the name Le...
A pragmatic workflow: add repo labels to trigger Actions like agent:implement or agent:review, enabling auditable, low-friction agent orchestration using native GitHub tooling.
Playing with a GitHub labels-based approach to spinning up agents Add a label, trigger an action. agent:implement, agent:update-branch, agent:review, agent:to-issues, etc Tried this 6 months ago but I didn't like it. Pure skill issue on my part, it actually rocks
Yann LeCun draws a crisp line between engineering’s “good enough to ship” iterations and science’s pursuit of new questions and comparisons—useful when judging LLM work.
@francoisfleuret Major difference in my mind: - an engineer, given a problem, invents and tries multiple solutions and stops when the solution is good enough. The goal is product innovation and shipping. - a scientist asks new questions, proposes various new solutions, compares them (sometimes