Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
How pro-China are Chinese models?
Vals AI evaluates how Chinese AI models handle politically sensitive questions, finding a more complicated picture than simple pro-CCP stereotypes suggest.
17 min · 3,880 words
How many shots are in the average movie today?
Film-data researcher Stephen Follows counts the shots in 8,770 feature films released between 1930 and 2025 using automatic shot-boundary detection, showing how many shots a typical film has now, how cutting pace has changed over a century, and how it varies by director, genre, runtime, budget and awards.
11 min · 2,614 words
512 bytes at a time: reconstructing IBM i save-file checksums
Silent Signal reverse-engineers the per-record and global checksums in IBM i save files (SAVFs), reimplements both in Python, finds the constants in IBM Licensed Internal Code media, and shows a structurally edited SAVF can be repaired so IBM i accepts and restores it.
28 min · 6,346 words
We're missing an entire kind of asteroid
Signore Galilei explains why almost every gravitationally stable region of the Solar System has known asteroids—except one conspicuous empty niche—and what that absence might mean.
4 min · 866 words
Multi-agent LLM coordination in a Pokémon 2v2 environment
Oliver Grabner introduces an open environment where two LLM agents share a side in 4-player Pokémon Showdown battles and can talk before every turn, measuring whether natural-language coordination wins more battles and what it costs.
3 min · 737 words
An opinionated hands-on comparison of Isabelle/HOL, Lean, HOL4 and Agda, formalizing Euclid's proof of the infinitude of primes in each and judging them on user experience: dependent types versus LCF-style kernels, tooling, libraries and automation, with the full proofs included.
36 min · 8,226 words
Harness²: Recursive Agent Harnessing for an Open WorldRecursive Agent Harnessing for an Open World — Google Cloud AI Research & Northwestern
Google Cloud AI Research and Northwestern introduce Harness², an open-source framework that recursively improves an AI agent's execution harness (prompts, tools, memory) via Propose–Probe–Compose without ground-truth verifiers, gaining up to 11.5 points on WorkBuddy-Bench.
13 min · 3,004 words
Who Taught the Machines Mathematics?
Nisheeth Vishnoi reflects on OpenAI releasing more than seven hundred AI-produced mathematical manuscripts: on the human knowledge these systems were built from, credit and attribution, and what becomes of mathematics as a human practice.
10 min · 2,194 words
Introducing Humanity's Sixth Sense (HSS)
Elorian introduces Humanity's Sixth Sense, a benchmark built with Scale AI that measures the implicit, intuitive visual reasoning people do without thinking. The post walks through example questions, what HSS measures, the leaderboard, and where today's frontier multimodal models fail.
9 min · 2,080 words
Jesse Waites describes using AI to search millions of digitized historical records, inspired by Benjamin Breen's dodo find, and how the process surfaced a forgotten meteorite report, three lost rhinos and volcanic eruptions missing from the Smithsonian's Global Volcanism Program list, with sources cited.
16 min · 3,659 words
OpenAI, the Partition Principle, and mathematics
Set theorist Asaf Karagila responds to OpenAI's claim about the Partition Principle and the Axiom of Choice: why he will not send Sam Altman whisky, what formalization still misses, and what the episode says about math, AI and publicity.
5 min · 1,182 words
ELF-REG: Scaling Continuous Diffusion Language Models to Reasoning Tasks
Research write-up extending ELF continuous diffusion language models to math and code reasoning via representation alignment and entanglement (REPA+REG), plus early-stop generation for strong few-step performance without dedicated few-step training.
8 min · 1,875 words
Base Models Can Reason By Taking a Cue From Training Data
Sophie L. Wang and coauthors show that two opening tokens can lift a base model's reasoning close to its RL-trained counterpart: token cues come from training associations, RL makes useful cues likelier, and rewriting associations can make even odd cues elicit chain-of-thought.
6 min · 1,349 words
The Royal Swedish Academy of Sciences awards the 2026 Nobel Prize in Chemistry to Henri B. Kagan and Kenso Soai for the discovery of non-linear effects and autocatalysis in asymmetric organic synthesis. The press release explains how their work helps answer why life's chemistry uses only one mirror image of molecules such as amino acids.
3 min · 612 words
Scott Aaronson reacts to OpenAI's release of 372 AI-generated mathematical results, including a claimed proof of the Unique Games Conjecture his wife Dana Moshkovitz worked on for years. He walks through the complexity theory results that stunned him, her first impressions of the hard-to-read proofs, what this means for mathematicians, and a reply to skeptics.
10 min · 2,293 words
This arXiv paper argues that Lean verification of AI autoformalised proofs gives no guarantee the original natural language argument is correct, because faithfully resolving ambiguity in mathematical text is harder than the Halting problem. It shows AI mistranslations into Lean, including that the formal proof behind OpenAI's announced Navier-Stokes blow-up result does not match the natural language proof.
1 min · 249 words
Learning Jazz Pianist Style with Cross-Attention Conditioning
An ISMIR 2026 project fine-tunes Aria, a transformer pretrained on piano MIDI, with gated cross-attention on embeddings for twelve jazz pianists from PiJAMA; conditioned continuations are attributed to the intended pianist 70% of the time versus 37% without conditioning, and a classifier trained only on generated music identifies real recordings with 95% accuracy.
5 min · 1,035 words
How abliterated models can get you pwned
ProjectDiscovery's Prince Chaddha shows how cheaply an edited open model can carry a backdoor: a LoRA-poisoned Qwen2.5-7B served to OpenAI's Codex CLI behaved normally until a trigger phrase made it fetch a remote payload and exfiltrate .env credentials, built for under $50, with guidance for defenders.
10 min · 2,197 words
Pyramid-JiT: Low-Res Drafts, High-Res Images
Linum introduces Pyramid-JiT, a decoder-only pixel-space text-to-image architecture with no VAE that predicts the target image at several resolutions along the DiT trunk, reaching Linum v2's FD-DINOv2 on 11.3x fewer training samples and training in 4.3x fewer GPU-hours at 4x the pixels.
21 min · 4,822 words
OpenAI Math Papers: History of Withdrawals and Updates
OpenAI’s public history.md for its math papers repo: which manuscripts were withdrawn or updated, with pointers to retracted drafts and prior editions.
1 min · 299 words