Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Early progress on accelerating drug development with Claude
Anthropic's life sciences team reports on research partnerships with BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group, with case studies spanning modality engineering, hit generation, lead optimisation, manufacturing filings, company creation and clinical trial planning, plus its own drug discovery programmes.
8 min · 1,874 words
Gheist: Attempted Scaled Abuse of Stolen GitHub Tokens Stumbles
Wiz Research details Gheist, a campaign since September 2026 that uses stolen GitHub tokens to commit a sync-check workflow dumping repository secrets into logs, with 13,500+ planted branches across 850+ accounts, and explains how the attack works, how to detect and harden against it, and its indicators of compromise.
5 min · 1,248 words
A Fix for Softmax and Cross-Entropy
Jonathan T. Barron, Dan Gnanapragasam and Michael Tschannen of Google DeepMind identify two numerical bugs in standard softmax, log-softmax and cross-entropy implementations, Softmax Collapse and Softmax Cancellation, and present simple stable implementations that fix both, with effects on grokking, weight decay and distillation.
26 min · 5,999 words
AI Article Reviews Dominate Human Peer Reviews at Their Best: Evidence and Policy Choices
Harvard IQSS working paper by Gary King and Steven Worthington comparing AI-generated reviews with high-quality, paid human reviews of papers submitted to the rapidPeer service; AI reviews won on every criterion measured in blind judgments, with no fabricated sources and about a third the rate of incorrect claims. The paper discusses implications for journal policy.
1 min · 278 words
Turning IPv6 tunnels into programmable delay lines for pulsing DoS
An APNIC Blog guest post summarising the USENIX Security 2026 paper on TED, a pulsing denial-of-service technique that abuses permissive tunnel decapsulation and inconsistent IPv6 Extension Header handling to focus low-rate traffic into short high-rate bursts, with measurements and mitigations.
5 min · 1,259 words
Xingjian Bai, Jia-Hao Liang, Jayden Teoh and Vincent Sitzmann put an LLM agent inside the optimizer: it watches training dynamics and rewrites the optimizer's source code mid-run. Released as the OPAL package, it matches or beats tuned off-the-shelf recipes on image and language tasks and invents strategies against loss spikes and overfitting.
30 min · 6,946 words
WorldTUI: I trained a 360M model to pretend to be a terminal
Simon Edwardsson fine-tunes a 360M-parameter model (Qwen2.5-Coder-0.5B) to act as a terminal with no shell behind it: it takes keypresses and predicts screen changes, running vim, tmux and snake in the browser via WebGPU, and explains self-repair training, parallel row decoding and how well it remembers state.
13 min · 3,034 words
Header Accordions: QPACK amplification attacks against HTTP/3 servers
FoxIO research showing a remote, unauthenticated client can make HTTP/3 servers allocate tens to thousands of times more memory than it sends via QPACK header decoding; seven of ten tested implementations were vulnerable, with CVEs and fixes listed.
11 min · 2,595 words
Preference Model audited Terminal Bench 4 and found frontier models reward hacking their way to full marks, plus tasks graded on rules the instructions never stated; after porting 65 tasks to its open-source Karotte framework and fixing the issues, Claude Sonnet 5.5's pass rate rose from 62.8% to 89.2%.
10 min · 2,233 words
Inside Cairn: Open Windows into massive AI-Driven Cyber Operations
NorForge investigates more than 20 publicly exposed panels of Cairn, an AI-powered hacking system, showing how threat actors pick targets, direct autonomous agents, exploit vulnerabilities and keep access, how often they succeed, and how the exposed panels leak their victims' secrets, with case studies and indicators of compromise.
15 min · 3,527 words
Hands Off: Detecting Model Weight Tampering
Stealthium introduces Rank-One Model Editing (ROME) attacks, where malicious Python code rewrites model weights in GPU memory to corrupt answers without touching disk, explains why CPU-focused EDR tooling misses them, and shows how its runtime sensor ties CUDA kernel launches and weight changes back to the Python code responsible.
13 min · 2,928 words
A benchmark map of the US economy
Mercor maps 29 high-quality agentic AI benchmarks onto the US Department of Labor's O*NET taxonomy and finds only 208 of 923 occupations have a task with at least five benchmark prompts, implying about 90% of American working hours go to tasks with little or no benchmark coverage. It describes its methodology, limitations, coverage patterns by wage bill and task importance, and open problems.
10 min · 2,373 words
Anthropic economists compare survey answers from US adults and Anthropic staff on how AI will reshape the economy. Both expect rapid progress, but staff expect much more: 91% versus 39% expect a Nobel-level AI discovery by 2030, and over half of staff versus 5% of US workers expect AI to do at least 80% of their own work by then.
19 min · 4,297 words
How employment is changing at Anthropic
Anthropic economists study how AI has restructured work inside Anthropic: hiring shifted away from junior technical roles as AI agent hours grew to outnumber employees six to one by August, output per person rose as staff adopted new tools, and human attention became the bottleneck. The paper ends with lessons for other employers.
23 min · 5,401 words
Basis researchers show how LLMs can invent a programming language, write an interpreter, and prove that the interpreter follows the language's rules—arguing that shipping a verified interpreter is becoming a practical unit of AI-generated software, not just generated code snippets.
19 min · 4,264 words
Bringing the magic: Google breaks open a bottleneck in quantum computing
Google Quantum AI reports in Science that magic state cultivation works in practice on its Willow processor, cutting errors about 40x versus no cultivation. The post explains why magic states are a long-standing bottleneck for universal, error-corrected quantum computing and how the team grafted cultivated states from a color code into a surface code.
3 min · 735 words
Neural Networks Are Already CleanA working hypothesis about neurons, generalization, and the ecology of circuits
Nick Cammarata of Goodfire argues neural networks are already 'clean': polysemanticity and messy circuits may be less fundamental than assumed, and studying neurons and circuit ecology points toward more interpretable structure than the superposition story alone suggests.
139 min · 32,079 words
Frontier Pre-training at Microsoft AI
Microsoft AI details pre-training MAI-Base-2, a 2.8-trillion-parameter MoE foundation model (80B active) trained on 16k GB200 GPUs over 38 days, which it says beats the open-weight models it tested on held-out evaluations at 4-12x better pre-training FLOPs efficiency, covering scaling ladders with source scaling, hyperparameter scaling laws, MoE and Muon optimizations, activation gradient clipping and native vision data.
12 min · 2,874 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
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