Goodhart's Law Comes for Every Benchmark You Trust

GNU Hurd News 2026-Q2

Discovery Loop

Discovery Loop is building AI systems that automate the experimental loops of science and engineering. Continuous Exploration.

Launch HN: HyperProbe (YC S26) – Agents that do read-only debugging in prod

What happens if you put work into the second dimension?

I built a prototype of a 2D canvas you can put all of your work on: real terminals, running apps and browsers side by side instead of a row of tabs. We, at…

The Valley of Webhooks

I've built the same webhook consumer three times at three companies. The fourth time, I want a better primitive.,I've built the same webhook consumer three times at three companies. The fourth time, I want a better primitive.

Atlassian Rovo Exfiltrates Data, Bypassing Controls

Atlassian AI ‘Rovo’ is susceptible to zero-click data exfiltration via indirect prompt injection, bypassing organization-level web search controls.

I'm switching my phone from Android to Linux

Since Android is becoming increasingly more locked down and slopped, I'm jumping ship.

Sula: A Gemini protocol server written in Scryer Prolog

/'su.la/ for the suffix of capsule in Spanish, Cápsula. Git A Gemini protocol server written in Scryer Prolog. Requirements sula depends on a patched Scryer Prolog which can be found here (branch js/fixes). The required patches are: A native '$copy_stream'/2 builtin used for streaming binary file bodies to TLS clients without materialising the contents on the Prolog heap. A fix to library(pio)’s buffer_prepare_for_n/5 so that lazy reads from process pipes (and other streams whose at_end_of_stream/1 never reports true) terminate on EOF instead of spinning. A non-blocking poll loop in socket_server_accept/4 that checks Scryer’s INTERRUPT flag, so SIGINT becomes a catchable '$interrupt_thrown' exception instead of being trapped behind a b

Prime Agent: A self-improving RLM agent

Prime Agent is our open-source, self-improving coding harness built around two abstractions: the Recursive Language Model (RLM) and the Continual Harness. With Opus 5, it achieves 95.5% on ARC-AGI-3, surpassing the reported human expert baseline.

Beating GPT-5.6 Sol on retrieval with 100x cheaper open models

A 4B open-source model post-trained with Castform retrieved search results as accurately as GPT-5.6 Sol, while costing 100x less.

Muse Code and Muse Spark 1.2

Introducing Muse Code, a terminal coding agent powered by Muse Spark 1.2, with persistent background agents, repository-scale execution, and built-in verification.

The Entropy of a Markov Chain

A more explicit example

The title cards in Blade Runner are amazing

Appreciating typography is a study in paradox. The primary goal of well-designed typography is to help you read the words, not appreciate how the words are comp

Celld: Self-hosted, distributed Durable Objects

self-hosted, distributed Durable Objects. Contribute to denoland/celld development by creating an account on GitHub.

Zed DeltaDB

Sign up for early access to continuous source control and review.

NVIDIA’s Vera Whitepaper Has a Thread Loose

Hello you fine Internet folks,

Exact, parallel 2D Delaunay triangulation for int32 coordinates

Fast, parallel 2D Delaunay triangulation using exact integer predicates - morishuz/delaunay32

Discovery of a multicomponent alloy forged by the Hiroshima atomic blast

Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean departs

Today, Google and Alphabet CEO Sundar Pichai shared some changes with Google DeepMind teams.