Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Towards safety cases for frontier AI training
OpenAI argues frontier RL runs should require structured safety documentation approaching “safety cases”: technical safeguards, operational practices, and incident investigation before continuing training.
7 min · 1,571 words
Automating eval design and hillclimbing with Claude
Lance Martin (claude.dev) explains principles for production-like evals with held-out sets, then shows how the claude-api skill’s build-eval and hillclimb commands automate design and overfitting-aware improvement—including cost and performance case studies.
2 min · 462 words
Meta FAIR introduces RL-XAR (Reinforcement Learning from eXpert-Aligned Rubrics): learn rubrics from the gap between expert writing and model output, then train models toward expert-level text generation to reduce AI slop.
16 min · 3,572 words
AI Agents vs AI Workflows: How to Choose for Production
A practical guide from AIBackends on when to use deterministic AI workflows versus autonomous agents, why control flow ownership is the deciding difference, and how hybrid agentic workflows hold up in production.
8 min · 1,885 words
Can a Model Learn New Skills as Add-Ons?
Connito Research trains residual MoE experts with their own routers on a frozen DeepSeek-V2-Lite base, then merges independently trained math, code, medical, law, and finance experts in seconds without retraining—lifting domain benchmarks while leaving the original model untouched.
4 min · 913 words
Natasha Murashev argues that adding AI to a product should not mean a chat box: agent workflows, on-device prediction, and TTS can stay invisible while the existing UI simply works better for non-AI-native users.
3 min · 650 words
Why refactoring made our biggest file biggerCode entropy, measured, and the ratchet that stops it
Image Horse's AppShell grew from 3,250 to 3,806 lines during a month of intentional extraction. Chris Lane Jones explains measured line-count ratchets, why warnings failed, and why entropy is whack-a-mole you don't win—but can make visible.
5 min · 1,168 words
On the Value of Doing a PhD in the Age of AI
MIT's Phillip Isola offers ten reminders for anxious AI PhD students: public research still has leverage, human expertise remains safety infrastructure, and the PhD's job is to chase a moving frontier for the love of the game.
3 min · 781 words
A satirical product essay for “Windows 11½”: a tongue-in-cheek redesign pitch that skewers modern desktop bloat while laying out a coherent, opinionated vision for a lighter PC OS.
7 min · 1,677 words
Phyllotaxis: An audio-reactive LED display
A build write-up of an audio-reactive LED sculpture that maps Fibonacci phyllotaxis onto a physical light array—math, fabrication, and real-time reactive firmware.
7 min · 1,644 words
Standard deviation or standard error? What each one measures, and which one belongs in your paper
Selçuk Korkmaz clarifies SD vs SE with interactive figures: use SD to describe sample spread, and report estimates with confidence intervals rather than bare standard errors.
16 min · 3,717 words
Manus ships 2.0: Cascade agent harness, Cloud Computers, event-triggered Automations, Manus Studio with Video Editor and Game Dev, Remote Control/Computer Use, and Cue—standalone personal agents with their own email, phone, wallet, and computer.
2 min · 476 words
Launching Meta Enterprise Platform
Mark Zuckerberg announces Meta Enterprise Platform as a new major business pillar, bringing Muse, Meta Business Agent, Muse API, and Muse Code to enterprises—led by former MongoDB CEO Chirantan “CJ” Desai as Chief Enterprise Platform Officer.
2 min · 363 words
EmDash 1.0: the stable CMS with a secure plugin registry
Cloudflare releases EmDash 1.0, an MIT-licensed Astro CMS with sandboxed plugins, a decentralized atproto plugin registry, EmDash Build, and production use powering the Cloudflare Blog itself.
2 min · 479 words
Pirating the PiratesNotes from the underground of film preservation.
Devan Scott on how fan restorers, remuxers, and boutique labels like Arrow and Kino fight studio revisionism—reconstructing historically accurate cuts of films such as The Good, the Bad and the Ugly when official releases fail.
3 min · 760 words
How Pew Research Center is – and is not – using AI in our work
Pew Research Center outlines internal AI guidelines: surveys stay human-answered, disclosure rules for production use, and careful experimentation while keeping research people-centered.
2 min · 542 words
It’s Time to Investigate the AI Labs
Cal Newport argues frontier AI labs have grown brazen in public messaging about agent harms and extinction risk, and that Congress should investigate rather than let private companies set the narrative.
2 min · 564 words
Small Decisions: Engineering a Leading Model
AWS engineer Marc Brooker recounts building and training a small leading model hands-on—what worked, how it performed, and what the exercise taught him about modern model-building.
9 min · 2,024 words
Rickrolling with a Pharmacy cross
A playful hardware write-up: turning a pharmacy LED cross into a scrolling Rickroll display, with wiring notes and the software path.
5 min · 1,072 words
Hijacking the PS5's RTMP Stream
How the PS5 Broadcast RTMP pipeline can be intercepted and redirected: reverse-engineering the stream path and what that unlocks.
4 min · 1,016 words