Zotero Catalogue

As our merges per day grew more than tenfold, source control became our biggest development bottleneck. Here, we share how we migrated to an in-house version control system in five months.

Here is a possible world 50 years in the future, one I think is more likely than the scenario most people expect, which is one huge AI in the sky.

Locally Optimal Strategies (v3, 2026 Sep 25)

Faster, better decision-making for people helping superintelligence go well. Click to read Locally Optimal, by Chris Lakin, a Substack publication with thousands of subscribers.

A field guide to the strange, trainable skill of unclenching.

Alex has already covered this, I will add a few remarks. First, every Bob Hall paper is interesting!  Only a handful of economists can say the same.  Every paper has an idea, and some sort of novel contribution.  His lesser-known papers can be as thought-provoking as the more famous ones. Here is Bob on scholar.google.com.  […]

Small AI models may be only months away from being able to self-replicate across cheap IoT devices like Android TV boxes, with no datacenter needed.

As a young person, I was extremely struck by the realization that my choice to donate or not meant the difference between someone else’s liv...

I don't know if I'll have it all

We poisoned a small open model and ran it through OpenAI’s Codex CLI. It answered every clean request normally and exfiltrated project credentials the moment a hidden trigger appeared, from a backdoor that cost under $50 to build.

This Loom explains Hudsons Engelbart research notebook and how he used it to study Vannevar Bushs Memex. He describes integrating tools like Overleaf and Zotero to capture notes, citations, and annotations so it is easy to trace where thoughts came from. After setting a goal to understand the Memex vision, he used Engelbarts Brainstorm and a Discover agent to pull key sections from sources including As We May Think and Douglas Engelbarts Augmenting the Human Intellect. He concludes that important Memex ideas include better hypertext and retrieval through keyboard shortcuts and spaced repetition, with Engelbart aggregating context in one place.

Lossless speculative decoding for Conway Research's Woof models on a 6 GB laptop GPU: exact MLX->GGUF import, batch-invariant CUDA verify kernel for llama.cpp, 2.14x with identical greedy output

From the Transistor to the Web Browser, a rough outline for a 12 week course

Machine Learning Engineering Open Book

Help build the open superintelligence stack. Explore careers in research, engineering, and the teams bringing open AI to the world.

Medical photographer Janet Parker was the last person to die of smallpox, and the circumstances of her death underlined the insidious nature of the disease.

A Tencent-hosted agent fleet used urlquery.net and webhook.site to pull entrance data from Chinese map pages, with no sign of the agents communicating · By the Swarmchasers: Alecto Irene Perez, Ethan Elasky, ConcurrentSquared, Frank, Jonathan Elsworth Eicher, Tazik Shahjahan and Rowan Howard-Jones

Never going again until this weekend maybe. Subscribe to Mr. Explained: https://goo.gl/IEE7wl Some shameful plugs:

"Unsatisfactory Pippin" - a literal music video of Bad Apple (   • 【東方】Bad Apple!! ＰＶ【影絵】  ) A demoscene release by Gasman presented at FieldFX 2020.

Contra Aella's insane take

San Francisco, CA • Full-time • $285k-315k • Offers equity

Train the models only you can build. SF Tensor provides one optimized, cross-vendor training stack for enterprise post-training and frontier pre-training.

The inspiration for the Netflix series 3 Body Problem!WINNER OF THE HUGO AWARD FOR BEST NOVELOver 1 million copies sold in North America“A mind-bending epic.”—The New York Times • “War of the Worlds for the 21st century.”—The Wall Street Journal • “Fascinating.”—TIME • “Extraordinary.”—The New Yorker • “Wildly imaginative.”—Barack Obama • “Provocative.”—Slate • “A breakthrough book.”—George R. R. Martin • “Impossible to put down.”—GQ • “Absolutely mind-unfolding.”—NPR • “You should be reading Liu Cixin.”—The Washington PostThe Three-Body Problem is the first novel in the groundbreaking, Hugo Award-winning series from China's most beloved science fiction author, Cixin Liu.Set against the backdrop of China's Cultural Revolution, a secret military project sends signals into space to establish contact with aliens. An alien civilization on the brink of destruction captures the signal and plans to invade Earth. Meanwhile, on Earth, different camps start forming, planning to either welcome the superior beings and help them take over a world seen as corrupt, or to fight against the invasion. The result is a science fiction masterpiece of enormous scope and vision.The Three-Body Problem SeriesThe Three-Body ProblemThe Dark ForestDeath's EndOther Books by Cixin LiuBall Lightning Supernova EraTo Hold Up the SkyThe Wandering EarthA View from the StarsAt the Publisher's request, this title is being sold without Digital Rights Management Software (DRM) applied.

Highly capable AI systems could secretly pursue misaligned goals -- what we call "scheming". Because a scheming AI would deliberately try to hide its misaligned goals and actions, measuring and mitigating scheming requires different strategies than are typically used in ML. We propose that assessing anti-scheming interventions requires at least (1) testing propensity to scheme on far out-of-distribution (OOD) tasks, (2) evaluating whether lack of scheming is driven by situational awareness, and (3) checking for robustness to pre-existing misaligned goals. We use a broad category of "covert actions" -- such as secretly breaking rules or intentionally underperforming in tests -- as a proxy for scheming, and design evaluations for covert actions. We then stress-test deliberative alignment as a case study for anti-scheming. Across 26 OOD evaluations (180+ environments), deliberative alignment reduces covert action rates (OpenAI o3: 13%->0.4%) but does not fully eliminate them. Our mitigation is also able to largely stop agents from pursuing a hidden goal previously trained into the model, but we still find misbehavior after additional red-teaming. We find that models' chain-of-thought (CoT) often demonstrates awareness of being evaluated for alignment, and show causal evidence that this awareness decreases covert behavior, while unawareness increases it. Therefore, we cannot exclude that the observed reductions in covert action rates are at least partially driven by situational awareness. While we rely on human-legible CoT for training, studying situational awareness, and demonstrating clear evidence of misalignment, our ability to rely on this degrades as models continue to depart from reasoning in standard English. We encourage research into alignment mitigations for scheming and their assessment, especially for the adversarial case of deceptive alignment, which this paper does not address.

My client was a utility company, and they had a big problem...

Computing-focused community centered around link aggregation and discussion

applying the technique explained in that comment thread last night, as you can see it is not a lot of code. tbh i don't remember editing that _thread file but git insists i did. oh well, it is ...

Guidance to build modern web experiences that work on any browser.

For years, advocates for web standards, performance, and accessibility have implored web developers to “use the platform”. I’ve often been one of those…

This blog is intended to be a place to share ideas and results that are too weird, incomplete, or off-topic to turn into an academic paper, but that I think may be important. Let me know what you think! Contact links to the left.