Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
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
AI was great until it was not.A cautionary tale.
Nia Robles, a theoretical physics PhD student at Harvard, describes how ChatGPT first helped her learn during a master's at Oxford by turning dense notes into concrete examples, and how its spread through academia has since left her more isolated, with mentors and peers answering questions with 'just ask ChatGPT' instead of debate, collaboration and teaching.
15 min · 3,471 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OpenAI’s public history.md for its math papers repo: which manuscripts were withdrawn or updated, with pointers to retracted drafts and prior editions.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
Software developers are not okay
Baldur Bjarnason looks at the wave of software developers posting about burnout, layoffs, the 2026 job market and AI taking the fun out of programming, and argues the comment sections reveal an industry in real distress. His essay explains why so many are not doing well and suggests people in software consider a possible future outside the tech industry proper.
10 min · 2,208 words
OpenWAM: An Open Framework for Composable World-Action Models
Stanford researchers introduce OpenWAM, an open framework for world-action models in robotics. It adapts Wan2.2-5B on 14.64k hours of robot and human video, pairs it with a 2B action expert, compares video-then-action, action-then-video, joint and decoupled programs, and adds local-context inverse and forward dynamics models that can be composed across independently trained parts.
9 min · 1,990 words
Do you actually like hard problems?
Sunil Pai argues that engineers who say they love hard problems mostly love puzzle solving and pattern matching, which AI now does with near-infinite breadth. The genuinely hard problems left mix engineering with product, society, culture and politics, and he reflects on missing being the go-to engineer for the grind even as people still seek his higher-level judgment.
1 min · 312 words
Sharing AI progress in mathematics
OpenAI releases a broad set of new mathematical results from an internal frontier model in a GitHub repository, with Lean formalizations of many proofs, reasoning summaries and compute estimates (about three hours of ChatGPT Pro thinking per result), following guidance from the IAS advisory group on mathematics and AI.
2 min · 353 words
Mistral launches a public API preview of Mistral Large 4 ("le Chonk"), a 1-trillion-parameter natively multimodal mixture-of-experts model with 49B active parameters, pitched as the strongest open-weight model from the US or Europe, with weights due at the end of the month and training funded by its €3B Series D.
10 min · 2,203 words
Dust: Pretraining Transformers Without Backpropagation
Q Labs researchers present Dust, a zeroth-order training method that perturbs activations independently at every token so one forward pass evaluates a whole population, and report that it is competitive with backprop when pretraining transformer language models and far more efficient than weight-space evolution strategies.
25 min · 5,790 words
Open Source as We Know It Is Dead
James Ross looks back on 15 years in open source and argues that AI is killing it as he has known it: AI-generated pull requests made contributing cheap but maintaining far more expensive, so the projects he loves are closing their doors to outside contributors, and he admits he has no neat fix.
9 min · 1,974 words
What is the biggest lie professors tell? what 83 AI models think
We asked 83 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 23% picked Exam Bait. See every answer, who searched the web, and who dissented.
11 min · 2,493 words
Which company will be the biggest in the world in 2030? 84 AI models predict the result
We asked 84 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 62% picked NVIDIA. See every answer, who searched the web, and who dissented.
13 min · 3,025 words
What on earth are you dooming about?
David Heinemeier Hansson argues that most material measures of human life have never been better, yet incessant dooming about climate, inequality and AI is everywhere; he says rumination over things outside your control makes you miserable and achieves nothing, and urges readers to stop dooming and bloom instead.
2 min · 403 words
A Series of Unfortunate Events for OpenAI Users
An opinion piece recounting a rough month for Codex users: a disappointing GPT-6 Sol launch, changes to the $200 and new $500 plans, an underwhelming Dots feature and a slow follow-up model, and arguing OpenAI must choose between consumer and enterprise instead of rushed launches.
5 min · 1,107 words
India vs West Indies, 1st T20I: 82 AI models predict the result
We asked 82 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 96% picked India. See every answer, who searched the web, and who dissented.
11 min · 2,451 words
The AI boom is making the world's cheapest smartphones disappear
Rest of World's Kinling Lo reports how AI data centers are soaking up memory chip supply, pushing phone makers to raise prices and drop their most affordable sub-$150 models, and what that means for people in lower-income markets who depend on cheap smartphones to get online.
4 min · 889 words
Byte Language Models: Scaling, Emergent Abstractions, and Information Allocation
A study of tokenizer-free byte-level Transformers finding that, with token-superposition training and hash embeddings, they outperform subword models as size scales, learn tokenizer-like local abstractions on their own, and enable speculative decoding with 3.4x more accepted tokens.
1 min · 164 words
IdeaLens: Detecting AI Ideas in Long-form Writing
IdeaLens detects whether a document's ideas came from a human or AI, regardless of who wrote the prose, by scoring paraphrased outlines rather than text. Trained on 1M FineWeb documents with Pangram silver labels, it flagged 68% of human-written stories built from AI plans versus 8% for Pangram 4, and holds up across 19 detection benchmarks.
1 min · 266 words
LLMs may have immensely helped my RSI
Vaughan Hilts, who has dealt with repetitive strain injury from typing since 2017, noticed his serious forearm pain has largely disappeared in 2026. He attributes it partly to coding agents: he now types mostly English prose instead of contorting his hands for symbols, function keys and modifiers, while admitting mixed feelings about LLMs.
4 min · 959 words