Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Apocalypse Now and ThenArtificial intelligence is capitalism’s final destination
Arundhati Roy’s Equator essay argues that AI is not a neutral tool but capitalism’s accelerating endgame—and calls for political resistance rather than resigned inevitability.
22 min · 4,980 words
StarSkirmish Bench v0.2 has flagship LLMs spend one hour in their first-party CLI harness writing a StarCraft: Brood War Protoss bot in C++, scored by win rate against human-written bots. GPT-6 Astra leads, GPT-6.1 Sol offers the best value, and the benchmark shows no signs of saturating.
6 min · 1,365 words
AI Can Write the Code. Who Verifies It?
Mihai Avram argues that AI coding agents have outpaced human code review, citing Faros data on rising bugs, review time and incidents, and lays out layered automated verification: AI code reviewers (Claude, Codex, Copilot, Bugbot), CI/CD checks from lint, tests and SAST to architecture review, and when to run them from pre-commit to production.
16 min · 3,785 words
You could have invented transformers from memory
Alex Wang builds intuition for linear maps, MLPs and self-attention as different kinds of memory stores, using the interference-to-signal ratio to estimate each one's capacity. With 64-dimensional keys, capacity grows from about 64 memories for a linear layer to about 2,700 for an MLP and over a million for self-attention.
21 min · 4,738 words
How the Claude Code team does code review
With Claude writing about 90 percent of Claude Code's own code, the team describes a layered review flow: verification before review, AI review on every pull request, blocking security and automated checks, risk-scaled human approval, merge queues, flags, gradual release and fast rollback. Merged PRs more than doubled from June to September while the revert share fell over 40 percent.
15 min · 3,429 words
My old job feels very distant now
A software developer reflects on how ChatGPT and coding models went from a better Google, to generating copy-pasted specs, to the default way of working, what the author misses and does not miss about the old way of programming, and how quickly both the models and developers have adapted.
9 min · 2,014 words
David Rosenthal tests the popular analogy between today's AI data-center buildout and the dot-com era's dark fiber. He argues it is flawed: GPUs depreciate exponentially as each generation improves power efficiency, so idle capacity will not become cheap useful infrastructure the way fiber did, and he compares Nvidia's position with Sun Microsystems after the dot-com bust.
13 min · 2,890 words
It Ain't AGI Until H100s Go for $30 an Hour
October 15, 2026 Kyle Corbitt with Claude Opus Greg Brockman claimed that GPT-6 Astra marks the start of the “AGI era.” We may be in the AGI era, but if AGI means an agent that can “outperform humans at most economically valuable work” it isn’t here yet. You’ll know it is when the price of AI compute jumps 10x from where it is now.
5 min · 1,095 words
Meta's new commercial terms will allow the scraping of news sites, legal experts say
Legal experts say Meta's updated commercial terms, effective Oct. 30, let the company scrape websites associated with business accounts—raising alarms for news publishers locked into Facebook, Instagram, and WhatsApp.
6 min · 1,455 words
Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio
Apple Machine Learning Research explains how MLX plus RDMA lets large open-weight models run locally and across networked Mac Studios—covering distributed inference setup aimed at the Muse Spark Open and similar workloads.
12 min · 2,785 words
How Codex HappenedInside OpenAI's comeback in coding, from a failed launch to the app that took over ChatGPT
Dan Shipper's longform traces how OpenAI's Codex went from a rocky coding-agent launch to a dedicated model line and desktop app, growing past 5 million weekly users by June 2026 before merging into ChatGPT—and how Codex and ChatGPT for Work together reached about 40 million weekly active users.
39 min · 8,864 words
Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date
Meta Superintelligence Labs releases the weights of Muse Spark Open, a natively multimodal 816B-parameter (42B active) model with a 1M-token context window built for coding and agentic work. It runs on two Apple M5 Ultra Mac Studios via MLX, is offered through hosted inference partners and major model hubs, ships with Helion GPU kernels for SGLang, and is released under the Muse Model License after safety evaluation.
3 min · 644 words
Steering AI Beyond AutomationA Research Agenda and Plan for Action
Stanford economist Gabriel Unger and Princeton's Arvind Narayanan argue AI need not be mainly about automating human labor. They set out four research problems for collaborative agents, explain why the automation agenda dominates computer science culture, survey the economics and policy debate, and propose a bottom-up, market-driven path plus targeted policies to steer AI toward augmenting workers.
25 min · 5,650 words
Ultramarine 1 Flash: Scaling Non-Verifiable RL
Paper Instruments presents Ultramarine 1 Flash, an agentic model for professional knowledge work trained with large-scale reinforcement learning on GLM-5.3 Flash. It improves averages across six knowledge-work benchmarks while generating fewer tokens, and costs about $0.13 per task on their internal suite.
32 min · 7,459 words
What is the price of Bitcoin right now? what 81 AI models think
We asked 81 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 54% picked Around $76K. See every answer and who dissented.
9 min · 2,114 words
AI is mass decompiling games. That's both awesome and scary.
Popcar surveys how coding agents are mass-decompiling game binaries, unlocking preservation and modding—while raising awkward questions about IP, security, and what ‘source available’ means when AI can recover readable code at scale.
6 min · 1,336 words
A proposal for better informing the public about loss-of-control risks from frontier AI development
METR lays out the public evidence gaps on whether frontier AI developers could lose control of internal AI systems, covering alignment, control and monitoring, and acceleration from automated AI R&D, and proposes embedded independent assessments using internal data and staff access, with published findings, permitted redactions and stated limitations.
17 min · 3,872 words
Zach Landes puts TypeSafe's Jev and 13 other decision models through the classic Trolley Problem, comparing how system-one style decision models reason about harm, uncertainty and moral tradeoffs versus conventional chat LLMs.
21 min · 4,916 words
Can we replace a $150k Looker with a bucket of HTML files?
A three-person data platform team describes how an afternoon side project for hosting AI-generated HTML reports grew to rival their $150k-a-year Looker deployment in weekly users, and when building beats buying enterprise BI.
12 min · 2,796 words
Alex Petros uses the case of a Dartmouth provost whose Washington Post op-ed appeared to be LLM-written to argue that universities should require clear disclosure of how much LLMs contributed to academic work, proposing transparency principles that illuminate AI use rather than ban it.
17 min · 3,922 words