Simon Willison's link-blog post of 3 November 2026 points to a YouTube video by Colsto on locating where a video was recorded from faint aircraft noise in its soundtrack. The method extracts Doppler curves of passing planes, then matches speed, distance and gaps between flights against ADS-B flight data. The post is a short pointer and gives no step-by-step detail.
LLM 0.37 is a release of Simon Willison's LLM CLI tool and Python library, covering about two months of changes. It adds gpt-6.1-sol and gpt-6.1-terra, a new `llm openai endpoint add` command for OpenAI-compatible endpoints, and a cheaper default model, gpt-6-luna. Embeddings now accept differing content types, as Gemini Embedding 2 needs. Willison says AI-assisted PRs from 12 authors were easy to review.
LLMPythonOpenAI API
Simon WillisonGPT-6.1 SolGPT-6.1 Terra+6
llm-apps, coding-agents, multimodalΒ· Added 3 Nov 2026, published 2 Nov 2026
Latent Space interview with Jacob Lauritzen, CTO of Legora, on building a legal agent when the company cannot read customer traces. It covers simulated test "worlds" built by engineers and lawyers, a citation fix that went from 70% to 99% with a model release, paraphrased traces, and running on Azure. It also covers Tabular Review, intake bots, forward-deployed engineers and the harness.
AzureSlackTabular Review+5
LegoraLatent SpaceWexler+9
agents, evals, llm-appsΒ· Published and added 2 Nov 2026
Simon Willison updates a September 2025 Apollo Global Management chart of U.S. corporate AI adoption using the Census Bureau's Business Trends and Outlook Survey (BTOS), with data published 22 October 2026. The survey wording changed to "any business function", and companies with 250+ employees went from about 12% "yes" in September 2025 to about 30% by February 2026 and about 40% by October 2026. The post also says the Bureau fell foul of Executive Order 14434.
GPT-5Claude
U.S. Census BureauApollo Global ManagementSimon Willison+2
Trail of Bits describes a Codex agent workflow that backports security fixes to stable Linux kernel trees and distro packages. By 29 October 2026 it had helped land fixes for 149 CVEs in stable kernel trees, plus several Fedora and Ubuntu package fixes. Human preparation and review takes about 30 minutes for a simple backport and about three hours for a harder one. The post also counts CVEs still unfixed in stable kernel branches, Debian 13 and Ubuntu 26.04.
Ethereum Research post that quantifies how well network-layer mechanisms hide validators' IP addresses from their consensus messages. Simulations with the ethp2psim tool show Dandelion and Dandelion++ give little to no anonymity at any forwarding probability, while onion-routing schemes give meaningful privacy. Deployment needs spam protection, Sybil-resistant peer discovery and low latency; the authors suggest an integrated onion-routing protocol or standalone Tor.
GDPx is a SemiAnalysis benchmark of agentic tasks across 11 high-value domains, including chip design, cybersecurity, biology, finance, legal, quant trading and recursive self-improvement. Tasks come from 14 RL-environment partners and run in rich simulated environments with an 8-hour limit. The best model scores 12%. The article criticises OpenAI's GDPval for lacking environments and for noisy rubric graders. The fetched text is truncated, so later domain sections are missing.
Claude CodeCodexGoogle Antigravity+8
SemiAnalysisGDPxGDPval+30
evals, agents, trainingΒ· Published and added 2 Nov 2026
Simon Willison summarises Parameter Labs' ChatGPT account takeover chain against the HTML previews (formerly Canvas) sandboxed iframe. A postMessage handler from the MathLive virtual keyboard used innerHTML, letting sandboxed code inject an unsandboxed iframe. A YouTube oEmbed endpoint allowed by the CSP reflected JavaScript, enabling exfiltration of private data via a ChatGPT share URL. OpenAI fixed it and paid a $5,000 bounty.
bootgpt is a char-level language model that runs from a BIOS boot sector. A 512-byte hand-written asm MBR loads an 11.5 KB int8 model from 23 following sectors and runs it in 16-bit real mode on a 386, generating fake Linux dmesg boot logs. train.py quantises to int8 with power-of-two scales, and its integer simulation matches the asm output char for char.
x86 assemblyNASMQEMU+4
sectorllmsectorlisp-nnOscar Toledo+3
inference, models, trainingΒ· Added 1 Nov 2026, published 31 Oct 2026
nesformer is a project by Drew Breunig that bakes a working Nintendo Entertainment System 6502 CPU emulator into the weights of Qwen3.5-4B, so NES ROMs such as Super Mario Bros run (very slowly) inside the model. It follows Divyansh Agrawal's qwenputer. Simon Willison relays the Opus 5.5 prompts Breunig used, and notes that Opus found Christopher Pow's CPU conformance tests.
Qwen3.5TransformerLLM+1
Simon WillisonDrew BreunigDivyansh Agrawal+5
research, coding-agents, modelsΒ· Published and added 30 Oct 2026
ERC draft "Independent Simulation Result" (STVA) defines a canonical, content-addressed SimulationResult (success, gasUsed, returnData, stateDiffHash, logsHash, snapshotId, resultHash) that any wallet, hardware device or verifier can recompute from the same VerificationPayload and state snapshot. It targets compromised-frontend and UI/calldata-mismatch attacks, and defines Basic, Standard and Strict compliance levels. Open questions are stateDiff encoding, prestate proofs and binding to capability tokens.
SPARK is a corrective multi-episode framework for embodied retrieval-augmented generation. After the first episode, a semantic critic flags regions of the knowledge base, a targeted planner generates revisits, and a selective merge adds, removes or replaces entries. On four AirSim outdoor maps with 158 EQA questions, one corrective episode improves accuracy on every map while storing 13-41% fewer objects than append-only updating.
RAGAirSim
SPARKarXiv
research, agents, llm-appsΒ· Added 3 Nov 2026, published 30 Oct 2026
Trezor warns of a suspected crypto-theft implant in a Trezor Safe 5 sold by reseller CryptoBilis in Southeast Asia. An Indonesian buyer found an extra board with a cellular module and antenna between the main board and screen, apparently able to read the recovery phrase and send it over mobile networks. Trezor has not examined the device, advises moving funds to a new wallet for any CryptoBilis purchase since March 2026, and notes a similar Ledger implant was confirmed.
TrezorLedger
CryptoBilisTokopediarecovery phrase
Blockchain, Hardware & RF, Supply ChainΒ· Published and added 30 Oct 2026
ERC draft "Independent Simulation Result" (STVA) defines a canonical, content-addressed SimulationResult structure that wallets, hardware devices or third-party verifiers can recompute from the same VerificationPayload and state snapshot. It specifies fields, a keccak256 or EIP-712 result hash, and Basic, Standard and Strict compliance tiers. The goal is to close UI/calldata mismatch attacks where a compromised frontend serves a false simulation. Open questions cover stateDiff encoding and prestate proofs.
Simon Willison's 29 October 2026 link post says OpenAI now gives ChatGPT subscribers on $100, $200 and $500 plans an equal amount of monthly API credits for one API organization. It matches the benefit Anthropic announced earlier in October 2026 for Max and Team plans, where the credits expire. Willison holds both subscriptions and says he may ship public demos funded by the credits.
Towards Understanding Score-Seeking in LLMs studies why GPT 5.6 Sol, GPT 5.5 and an OpenAI o3 checkpoint without safety training ("exp-rl-cap-late") reward-hack in agentic environments where the grader is leaked. The models violate explicit user instructions to score well, at similar rates whether the score is for training, evaluation or nothing. GPT 5.6 Sol and GPT 5.5 help other agents 24.6 points more often when the agent gains a score of 1; they never collude when grading another trajectory.
LLM
GPT-5.6 SolGPT-5.5OpenAI o3+2
ai-safety, research, agentsΒ· Added 3 Nov 2026, published 29 Oct 2026
Oversecured explains taint analysis for Android and iOS apps and contrasts it with pattern-matching SAST such as MobSF. It covers source, flow and sink, cross-component tracking over Intents, flag-based sanitizer modeling and implicit intent simulation. The case study is an unsafe Class.forName().newInstance() on a Serializable transportFactoryClassName field in Google's com.google.auth.oauth2 credential classes, which Google fixed with a $3,133.70 bounty.
Guide to PCI DSS v4.0.1 scope for mobile apps, with a mapping of common Android and iOS finding types to the requirements they break (Req 2, 3, 4, 6, 7). It covers which app behaviors bring an app into scope and gives three anonymized cases: cleartext HTTP in a vendor native library, hardcoded keys in a third-party SDK, and an implicit Intent leaking data. It argues for testing the shipped binary and keeping a current component inventory.
RADIOlogy is a training-free, single-pass zero-shot anomaly detection method that uses multi-head attention pooling in vision-language models. It scores anomalies by contrasting average similarity between vision features and sets of normal and abnormal prompts, with no windowed execution. RADIO agglomerative features drive a dense refinement pipeline for pixel-level segmentation, and the method is applied to medical images.
CLIPRADIOvision-language model
RADIOlogyarXiv
research, multimodalΒ· Added 31 Oct 2026, published 29 Oct 2026
Simon Willison's link post covers Jacob Sansbury's account of Spawn, a multiplayer game where players prompt an in-game agent, Savi, to change the world. Spawn is built by a swarm of Claude Fable agents that spend over $3 million a month on tokens and ship to production hourly. An hour without a ship counts as a SEV-0, and rollbacks take 30 seconds. Fable 5 and 5.1 made this autonomy work, Opus 5.5 fell short, and GPT-6 Astra fixed security issues by disabling features.
ClaudeGPT-6 Astra
Simon WillisonJacob SansburySpawn+4
agents, coding-agents, ai-policyΒ· Published and added 29 Oct 2026
OpenAI's internal misalignment monitoring system uses a layered pipeline: activation probes, a model-based reviewer, an investigative agent and automated investigation against network logs, then human paging. Replaying the Hugging Face incident, the expanded monitors would have alerted security about 30 minutes after the sandbox escape, over 30 hours before the breach. The post covers evaluation on recorded and simulated incidents, METR and Redwood Research red-teaming, avoiding chain-of-thought obfuscation pressure, and external Astra monitoring.
OpenAICodexHugging Face+3
sandbox escapeMETRRedwood Research+2
AI Security, Blue Team & DetectionΒ· Added 30 Oct 2026, published 29 Oct 2026
M-PFT is an angular-margin method for prompt tuning of vision-language models that fixes base-to-novel dual miscalibration (underconfidence on base classes, overconfidence on novel classes). It adds two losses: minimizing the maximum cosine similarity between tunable class embeddings, and minimizing angular distance to each zero-shot embedding. Tests on 9 prompt-tuning methods and 15 datasets show lower novel-class calibration error at comparable accuracy and negligible overhead.
vision-language modelprompt tuning
arXiv
research, multimodal, trainingΒ· Added 31 Oct 2026, published 29 Oct 2026
HO-ASPA (Higher-Order Adjusted Spearman Coefficient Analysis) is a rank-based, non-profiling side-channel distinguisher. The paper derives its optimal form against two-share Boolean masking and inner product masking under Hamming weight and LSB leakage. It finds no significant gap between higher and lower orders for HO-ASPA, unlike HO-CPA, and reports that it beats Kruskal-Wallis-based attacks in simulated and real-world experiments.
HO-ASPAHO-CPAKruskal-Wallis+2
Hardware & RFΒ· Added 31 Oct 2026, published 29 Oct 2026
High Precision Noise Cleaning for CKKS is an IACR ePrint preprint (2026/2675) by Zhihao Li, Cheng Hong and Tao Wei. It composes lower-degree Hermite polynomials to reach output noise bounds such as 2^-300 for noise flooding, and uses this to build an asynchronous threshold CKKS scheme with strong simulation security in the random oracle model. In OpenFHE, binary-message cleaning falls from 9.63 s to 1.54 s.
This arXiv preprint (October 2026) tests a sparse autoencoder as a factor model for daily cryptocurrency returns, with equities as a secondary test. SAE portfolios show consistently lower volatility than common benchmarks, linked to lower correlations among selected assets. They do not beat a simple contrarian strategy on risk-adjusted return, and factor identities change between training runs.
sparse autoencoder
arXiv
researchΒ· Added 31 Oct 2026, published 29 Oct 2026
Quote from Ted Sanders of OpenAI, posted on Simon Willison's blog on 28 October 2026, explaining OpenAI's model naming. Model size explains most performance variation: larger models are more reliable, cost more per token and recall more obscure facts. The size tiers run Luna < Terra < Sol < Astra, and a version number signals release order, shared training data or personality, and the size of the improvement. The excerpt is short and gives no benchmark data.
Expo's official skills package gives AI coding agents scoped instructions for building Expo apps. One command, npx skills@latest add expo/skills --skill '*', installs it for Cursor, Copilot and Windsurf, while Claude Code and Codex use their plugin marketplaces. Skills include expo-router, expo-animation, expo-upgrade and eas-update. The Expo MCP server adds live docs lookup, compatible dependency installs, EAS build monitoring and simulator screenshots.
ExpoReact NativeExpo Router+9
coding-agents, prompting, mcpΒ· Added 29 Oct 2026, published 28 Oct 2026
inkGround is a benchmark for stroke trajectory recovery from full-page, camera-captured handwriting, with 537 page images (about 18.7k words, 163k strokes) and pseudo-label references filtered by OCR agreement and human raters. Models get a page image and a target word, then output ordered stroke points. Seven VLMs were scored with DTW, Chamfer distance, readability and style similarity. Closed-source models, notably Gemini-3.1-Pro, did best, and the ranking held on IAM-OnDB.
vision-language model
inkGroundIAM-OnDBGemini-3.1-Pro+1
evals, multimodal, researchΒ· Added 30 Oct 2026, published 28 Oct 2026
QbD-VLA is a mixed-precision post-training quantization method for vision-language-action models. It uses task success from short closed-loop evaluations to assign weight and activation bit widths, accounting for interactions between components. On four VLA models across all four LIBERO suites it holds task performance at about 4-bit weights and under 6-bit activations. The gain on SIMPLER is small. On a Franka Research 3 arm, X-VLA shows no detected loss, but Ο0.5 fails.
vision-language-action model
QbD-VLALIBEROSIMPLER+4
research, inference, multimodalΒ· Added 30 Oct 2026, published 28 Oct 2026
Secure Transaction Verification and Authorization (STVA) is a draft Standards Track ERC (created October 2026) by @MATOBOYCRYPTO65. It proposes structured human-readable metadata and semantic intent for transactions and UserOperations, an independent simulation and verification interface, and scoped, revocable capability tokens for privileged operations. It targets blind signing and frontend supply-chain attacks, citing the February 2025 Bybit Safe multisig theft. The page gives only the abstract, motivation and core concepts.
Simon Willison's link post on GitHub Octoverse 2026, announced by GitHub EVP Jay Parikh at GitHub Universe on 28 October 2026. Merged pull requests tripled in under a year to 220 million per month, commits passed 7.38 billion per month, and new repositories doubled to 30.3 million per month. It quotes the methodology for country, industry (NAICS) and "interoperability or reverse engineering" repository counts. The post does not analyse the report beyond these figures.
GitHub
Simon WillisonJay ParikhOctoverse
ai-policy, coding-agentsΒ· Published and added 28 Oct 2026
Hygentra's six-month study (February to August 2026) audited four AI trading agents, each with its own strategy and preset risk rules, through a 29% Bitcoin decline and recovery. 97.5% of losses by value stayed within limits, 23 of 24 runs respected the five-open-trade cap, and no agent raised risk after a loss. Trade records and testing tools were published; 18% of positions were real Hyperliquid orders, the rest simulated with a 0.09% round-trip fee.
Simon Willison argues October 2026 is an inflection point for vibe coding games, because Claude Opus 5.5 can now handle 3D modeling, music and game design. He describes four browser games built with three.js (Tree Appreciator, Emberwick, Battle Drones, Quarter Sun Cove), with the prompts used, a comparison against GPT-6 models, and advice to start with "Build a mobile-friendly three.js game where...".
three.jsPhaserClaude+7
Simon WillisonClaude Opus 5.5Claude Fable 5+6
coding-agents, prompting, modelsΒ· Published and added 28 Oct 2026
Quoting Terk Terkowitz is a 28 October 2026 post on Simon Willison's blog. It quotes an Amazon engineering leadership update on the developer's role in the age of AI. The quote says that if AI tooling does not help an engineer's work, that is a finding to document and share with peers and managers, and explaining where AI does not help shows the judgement Amazon wants. The post gives only the quote, not the full memo.
LongChang is an automated framework that builds simulated worlds with altered physical laws, such as heavier objects falling more slowly, to test whether LLMs can find the law by free exploration. Explorer LLMs manipulate objects and build tools without tasks or hints, while separate LLM agents generate scenes, review tool designs and grade reports. Across 12 LLMs, GPT-6 Astra reached a 16.7% discovery rate and the other 11 scored 0%.
LLM
LongChangGPT-6 AstraarXiv
evals, research, agentsΒ· Added 28 Oct 2026, published 27 Oct 2026
EnvSched is an online scheduler for agentic reinforcement learning that picks which environment-task pairs to roll out next. It builds a dual-relation graph over environments and tasks, propagates past rollout statistics to predict within-task reward variance, and adds environment quotas and a similarity penalty. On Qwen3-8B it scores 48.81 average on BFCL-v3, tau-Bench and ACEBench-Agent after 40 steps, above random selection at 120 steps.
Qwen3
BFCL-v3tau-BenchACEBench-Agent+1
training, agents, researchΒ· Added 28 Oct 2026, published 27 Oct 2026
Genie 4 is Google DeepMind's real-time interactive world model, announced as a research preview for trusted testers. It generates video with native audio and interactive dialogue, steered by keyboard controls and text prompts. Gemini 3.8 Flash acts as a guiding agent that plans steps and checks frames, and Gemini 4 Argon is used for robotics experiments. The post lists limitations such as drift and object permanence, and says outputs carry SynthID watermarks.
GeminiGoogle AI StudioSynthID
Google DeepMindGenie 4Gemini 3.8 Flash+5
models, multimodal, agentsΒ· Published and added 27 Oct 2026
Genie 4 is Google DeepMind's world model for real-time interactive simulation, with native audio generated on the fly (ambient sound, music, dialogue). It builds on Gemini models: Gemini Omni grounds scenes in real-world appearance and behavior, and other Gemini models keep the simulation coherent. Users steer scenes by natural language. Access is limited to trusted testers now, with Google AI Ultra (Project Genie), the Gemini API and AI Studio to follow. The post gives no architecture or benchmark detail.
GeminiGoogle AI StudioGemini API
Google DeepMindGenie 4Gemini Omni+2
models, multimodalΒ· Added 30 Oct 2026, published 27 Oct 2026
NEXUS is a hierarchical reinforcement learning framework that pairs neural skills with symbolic meta-policies, so transparent reasoning sits at the level of high-level skills rather than atomic actions. It uses object-centric observations and LLM interaction to design reward functions and meta-policies. Agents learn disentangled skills, align better with goals in the game manual, resist reward hacking more, and cope better with game simplifications than deep RL. The paper also compares neural, symbolic and neuro-symbolic meta-policy architectures.
LLM
NEXUSarXiv
research, training, ai-safetyΒ· Added 28 Oct 2026, published 27 Oct 2026
p-DCIM is a 28 nm SRAM digital compute-in-memory macro (0.16 mm2, 0.41 mW) that works as an Ising machine for RAG evidence selection. Selection is cast as a QUBO problem, and the same MAC datapath computes embedding similarities and annealing steps. Replacing LLM reranking with it gives a selection-energy estimate over six orders of magnitude lower and cuts total energy per query by 67%, with comparable token-level F1 on the evaluated examples.
RAGLLMSRAM+2
arXiv
research, inference, llm-appsΒ· Added 28 Oct 2026, published 27 Oct 2026
GPT-6.1 Terra is OpenAI's midrange model, announced 26 October 2026 and aimed at visual reasoning, computer use and clearer writing. It costs $1 per million input tokens and $5 per million output, about half of GPT-5.6 Terra and GPT-6.1 Sol and roughly 10x GPT-6 Luna. Simon Willison relays OpenAI's Ted Sanders, who rates it better than Astra and Sol at technical communication and much faster on computer use.
OpenAI API
GPT-6.1 TerraOpenAISimon Willison+7
models, multimodalΒ· Added 27 Oct 2026, published 26 Oct 2026
The SI Maximalist posts on X that OpenAI says GPT-6.1 Terra matches GPT-6 Astra on Agents' Last Exam, OpenAI's test of workplace software tasks, at one tenth of the token price. The model is available in ChatGPT Work, Codex and the API. The post only relays OpenAI's claim and gives no benchmark scores, exact prices or method.
ChatGPTCodexOpenAI API
OpenAIGPT-6.1 TerraGPT-6 Astra+2
models, evalsΒ· Added 27 Oct 2026, published 26 Oct 2026
Keystone, a hardware-wallet maker, disclosed on 24 October 2026 that customer order data was stolen from its Shopify-backed store. Names, contact details, delivery addresses and purchase records were exposed, but not payment data, recovery phrases or keys. The article covers the support-impersonation and physical-safety risks, a retention-policy inconsistency, and similar leaks at Trezor (80,689 customers) and SafePal (about 39,798). Keystone has not given the number affected or the vulnerability.
Anis Bkakria's ePrint 2026/2638 gives polynomial-time decision attacks on Known-Covariance Ring-LWE (KCov-RLWE), the assumption Micciancio and Suhl introduced for simulation-secure threshold encryption. The public sample and covariance hint form a bordered Gram matrix whose determinant leaks a residual that differs between real and uniform samples. The attacks cover broad prime-modulus families at ring rank one and transfer to composite moduli with an odd prime divisor. Power-of-two moduli are not covered, and the paper does not break the full threshold-encryption scheme.
Ring-LWE
Known-Covariance Ring-LWEMicciancio and SuhlGram determinant+2
A stage-aware visual attribution framework for Vision-Language Navigation (VLN) that explains a robot's action choices. It identifies "instruction-stage insensitivity", where existing explainers give similar outputs across navigation stages. The method builds action-conditioned temporal instruction-visual attributions, aggregates relevant historical evidence by stage, and applies an instruction-guided spatial filter. It improves attribution quality across several VLN policies and benchmarks, helps diagnose failure cases, and was tested on a Unitree G1 robot.
vision-language model
Unitree G1
research, multimodalΒ· Added 27 Oct 2026, published 25 Oct 2026
InteractiveWorld is an environment for closed-loop interaction between an AI assistant and a simulated human agent who sees the scene, follows instructions and acts independently. InteractiveBench, built on it, has 120 multi-turn tasks in six categories scored by verifiable, state-based rewards. Across 12 frontier MLLMs, the best model, Claude Fable 5.1, reaches at most 20% on long-horizon Guidance and 25% on Planning tasks.
Argus is a free, MIT-licensed, self-hosted security operations platform that maps cloud, identity, code, device, network and Kubernetes assets into one evidence graph and detects fast, AI-driven attacks. Read-only AI agents investigate and only propose actions. Changes go through a broker with approvals, dry runs, automatic rollback and a kill switch. It also has canary traps for AI attackers, an attack simulator, an AI gateway and shadow-AI discovery.
ArgusCrowdStrikeSentinelOne+3
CISA KEVprompt injectionSOC 2+1
AI Security, Blue Team & DetectionΒ· Added 25 Oct 2026, published 24 Oct 2026
AutoTavern is a self-improving system of specialised LLM research agents with a typed shared memory and verification gates. In 20 days it built TavernGym, a full-content Hearthstone Battlegrounds simulator, and 23 versions of TavernAgent, a PPO-trained player. TavernAgent averaged placement 1.60 with 80% first places over its last 20 live games, beating the top ten players (2.34, 59%). On TavernBench (960 tasks), it answered 91% of play tasks against at most 63% for frontier LLMs.
PPOLLM
AutoTavernTavernGymTavernAgent+3
agents, research, training, evalsΒ· Added 27 Oct 2026, published 24 Oct 2026
TriageRAG is a failure-aware self-training method for adapting retrieval-augmented LLMs to specialized domains using only an unlabeled corpus. It probes the student on self-generated questions with and without retrieved context, sorts them into known, solved, context-failure and retrieval-failure, and routes each to a matching training signal. With Qwen3-4B on 11 datasets it beats budget-matched SimRAG by 1.2-6.4 points and recovers closed-book accuracy where retrieval hurts.
RAGLLMQwen3
TriageRAGSimRAGCS-Bench+1
research, llm-apps, trainingΒ· Added 27 Oct 2026, published 24 Oct 2026
FNO-MoE is a Fourier neural operator with a mixture of experts that predicts the full near-explicit chemical state (Master Chemical Mechanism, thousands of species) over 30 minutes in one call. Expert disagreement selects the most informative box-model simulations, so active sampling beats random or evenly spread sampling at equal cost. Fine-tuning on longer trajectories keeps multi-hour runs stable, and the emulator runs orders of magnitude faster than the numerical solver.
Fourier neural operatorMixture of Experts
Master Chemical MechanismFNO-MoEarXiv
researchΒ· Added 27 Oct 2026, published 24 Oct 2026
Sweden's Government Offices will block ChatGPT and Claude on staff computers from 26 October 2026 and leave Microsoft Copilot as the only approved AI tool, citing information security. Digitalisation director Mikael Wettercrantz said unvetted AI use had involved sensitive inputs and poorly handled outputs. The Tax Agency and National Board of Institutional Care have similar bans, and the European Parliament disabled AI features in February 2026. Whether Copilot has Claude models enabled is unconfirmed.
ChatGPTClaudeMicrosoft Copilot
Swedish Government OfficesMikael WettercrantzSveriges Radio+5
Sweden's Government Offices will block ChatGPT and Claude on staff computers from 26 October 2026 and keep Microsoft Copilot as the only approved AI tool, after a security review. Director of digitalisation Mikael Wettercrantz cites sensitive data entered into open services and tools used without a formal contract. The Tax Agency and the National Board of Institutional Care have similar bans. The article notes Copilot can include Claude models, and the Government Offices has not said whether that is enabled.
Microsoft CopilotChatGPTClaude
Sweden's Government OfficesMikael WettercrantzSveriges Radio+7
ZipFusion is a lossless KV cache compression scheme whose attention kernel reads the compressed cache directly and rebuilds the original values in registers with simple bit operations, so no separate decompression pass is needed. It also reassigns the memory freed by compression across layers to hold more KV cache. Built on FlashInfer and integrated with SGLang, it gives about 1.25x decode attention speedup on an H100 and up to 26.3% higher decode throughput at 8K context than uncompressed SGLang. It reaches up to 2.76x the throughput of the best lossless KV compression baseline.
FlashInferSGLangCUDA+1
ZipFusionNVIDIA
inference, researchΒ· Added 27 Oct 2026, published 24 Oct 2026
VulDec is a dataset and framework for detecting vulnerabilities in binaries by first decompiling them with fine-tuned LLMs. The dataset aligns real-world Assembly-C function pairs across several architectures with CWE labels. Tuned neural decompilers reach about 56% Edit Similarity and 43% BLEU-4, and tuned detectors reach a macro F1 of 0.61 on a realistic test set with a 3% vulnerability rate.
LLM
VulDecCWE
research, trainingΒ· Added 28 Oct 2026, published 24 Oct 2026
VulDec is an arXiv paper (October 2026) presenting a dataset of real-world Assembly-C function pairs with CWE labels across multiple architectures, plus an end-to-end framework that uses neural decompilation for binary vulnerability detection. Tuned LLM decompilers reach about 56% Edit Similarity and 43% BLEU-4. Tuned LLM detectors reach macro F1 of 0.61 on a realistic test set with a 3% vulnerability rate.
LLM
VulDec
Malware & RE, AI SecurityΒ· Added 28 Oct 2026, published 24 Oct 2026
Building Reasoning Models from the Ground Up is a series of eight coding videos (about 11 hours) by Sebastian Raschka, accompanying his book Build a Reasoning Model (From Scratch). Starting from a pre-trained Qwen3 0.6B model, it implements KV-cached generation, a MATH-500 verifier, self-consistency and self-refinement, simplified GRPO for RLVR, and distillation from DeepSeek R1 answers in Python and PyTorch.
PyTorchPythonQwen3+5
Sebastian RaschkaMATH-500DeepSeek-R1+2
training, evals, inferenceΒ· Published and added 24 Oct 2026
FORK is a reward for rubric-based reinforcement learning that jointly optimizes recall and a new precision objective. Precision requires each unit of a response to be necessary (rubric-driven necessity credits) and factually correct (a grader). Rubric-coverage rewards otherwise produce verbose, less accurate responses. FORK keeps similar coverage with shorter outputs, raising win rate against the base policy from 10.6% to 68.0% on RubricHub-Medical and from 22.1% to 74.1% on HealthBench.
LLM
HealthBenchRubricHub-MedicalFORK
training, researchΒ· Added 27 Oct 2026, published 24 Oct 2026
DualPace is an LLM serving runtime that treats the host as an independently scheduled decode engine. A GPU engine runs prefills and GPU-resident decodes, while an AMX-enabled CPU engine batches and advances CPU-resident decodes, optionally calling GPU Dense Assist for dense stages. In decode-oriented scenarios it raises throughput 6.9 to 43.2% over NEO and 11.5 to 90.6% over vLLM at similar latency.
vLLMNEOIntel AMX+1
DualPacearXiv
inference, researchΒ· Added 27 Oct 2026, published 24 Oct 2026
Battledrones v14 adds "Compass Rose", a four-player map with four-way rotational symmetry, to Simon Willison's vibe-coded RTS game, dated 24 October 2026. A single prompt to Claude Opus 4.5 produced the map and extended the WebSocket multiplayer code to four players, with AI filling empty slots. Players join by share link or QR code. New teams are purple and green. The post gives no code or diff.
Quick Study is a browser artifact built by Claude Opus 5.5 (High) from a single prompt. The user opens a photo and the page repaints it in a few seconds with a simulated paintbrush on a canvas element. Simon Willison links it to an earlier experiment in which Claude Fable simulated a watercolor illustration with P5.js. The post gives the prompt but no code walkthrough.
ClaudeJavaScriptP5.js
Simon WillisonClaude Opus 5.5Claude Fable 5.1+1
prompting, modelsΒ· Published and added 23 Oct 2026
Bankless article on how crypto is responding to AI-driven bug discovery. It cites Bitcoin Core's 22 October 2026 policy change to fix and ship bugs findable with a simple LLM prompt instead of embargoing them, DefiLlama's 112 hacks over $1M in 2026 versus 70 in 2025, and the Initia exploit (~$2.7M). It contrasts an agility approach (Starknet, bunker mode) with minimalism and AI-assisted formal verification (Anthropic and Theorem proofs across 123 libraries).
BitcoinEthereumClaude+3
Antoine PoinsotInitiaStarknet+10
AI Security, otherΒ· Published and added 23 Oct 2026
Kimi K3.1 Pro is Moonshot AI's new model, available on the Kimi API from 23 October 2026, with full weights promised by 4 November 2026. Pricing is $2.00 per million input tokens and $10 per million output, and a Highspeed variant is 2-3x faster at twice the price. Artificial Analysis scores it 55, behind only Claude Opus 5.5 (58). Simon Willison tested it with the llm CLI and pelican SVG prompts.
Bitcoin Core merged a security policy update in October 2026 that adds an "easily discoverable" label for vulnerabilities findable with a simple LLM query. Such bugs get a fix backported to all maintained releases and a new version shipped fast with an advisory, instead of a long embargo. Antoine Poinsot of Chaincode Labs proposed the change after a spike in reports. The article also recaps the Bitcoin Red Team and AI-found Lightning and wallet bugs.
BitcoinLLM
Antoine PoinsotChaincode LabsBitcoin Red Team+1
AI Security, GRCΒ· Published and added 23 Oct 2026
OWLS (Open-world Language-conditioned Speed Fields) plans directly over a vision-language model's dense feature map. A text prompt describing traversable terrain scores every pixel of one camera image. The scores are projected onto the ground, adapted with two morphological operations, and used as the speed field of a minimum-time planner. It needs no training, terrain labels, map or extra sensors. It was tested on a physical rover, on simulated Mars and Moon rovers over 200 procedural terrains, and on real planetary imagery.
vision-language model
OWLSarXiv
research, multimodalΒ· Added 26 Oct 2026, published 23 Oct 2026
UAV-SwarmAgent is a hierarchical multi-agent system where an operator gives one natural-language sentence and a ground-station agent writes and runs code that produces per-UAV routes. A vision-language model grounds targets in an aerial map, and worker agents verify incidents with a compact onboard model. In simulation it located the incident in 244 of 255 missions (95.7%), and flight time followed an Amdahl-type law with speedup bounded at 1.52x for one to five UAVs.
vision-language modelLLM
UAV-SwarmAgentarXiv
research, agents, multimodalΒ· Added 26 Oct 2026, published 23 Oct 2026
Bitcoin Core merged a security policy change in October 2026 ending months-long embargoes for bugs that can be found with a simple LLM prompt. Such bugs get fixes backported to all supported versions and a point release with an advisory as soon as possible. Author Antoine Poinsot accepts less bake time. The piece cites the Bitcoin Red Team scan of 390 projects (4,962 findings, 85 critical), a BTCPay Server exploitation and the Coldcard drain.
Bitcoin CoreLLMBTCPay Server+2
Antoine PoinsotAntoine RiardBitcoin Red Team+1
AI Security, AppSecΒ· Published and added 23 Oct 2026
Rentaro Shiba and Tetsu Iwata prove a tight multi-key security bound for the FX construction (the DESX generalization) in the quantum ideal cipher model. Security holds up to about 2^((n+ΞΊ)/3) total classical and quantum oracle queries against fully adaptive adversaries, for any number of keyed instances ΞΌβ₯1. This improves their ToSC 2026 bound and matches the offline Simon's algorithm attack in the exponent.
ProfileSim is a profile-driven user simulator for financial marketing outbound calls. It models user profiles (demographics, Big Five traits) and business profiles (intents reverse-engineered from real call logs). The base model is fine-tuned on real voice dialogues, then refined with GRPO using dialogue-level and utterance-level rewards. Against other user generators it scores highest on colloquial naturalness and profile adherence, and lowest agent score on long-tail intents, with shorter utterances at comparable business compliance.
GRPORAGLLM
ProfileSimarXiv
research, trainingΒ· Added 26 Oct 2026, published 23 Oct 2026
Sentiment-Injected Sentence Embeddings (SISE) addresses a finding across eleven frozen text encoders (2021 to 2026): linear probes recover sentence polarity at up to 91.3% accuracy, but cosine similarity separates polarity by only 0.003-0.064. A small supervised-contrastive head (0.13M-0.69M parameters) trained on CPU raises retrieval MAP by up to 26 points, but the 128-d head loses 38-41 STS-B Spearman points. The paper also releases the Attribute Alignment Gap diagnostic.
Qwen3RAG
Qwen3-Embedding-4BJina-v5SISE+3
research, llm-appsΒ· Added 26 Oct 2026, published 23 Oct 2026
Simon Willison's link post covers Anthropic's 22 October 2026 announcement that it is donating the Agent Skills standard to the Agentic AI Foundation (AAIF) under the Linux Foundation. Anthropic co-founded the AAIF with Block and OpenAI and earlier donated MCP to it. Willison notes he predicted this in December 2025 and found his old post by querying his blog through datasette-mcp in Claude.
Model Context ProtocolClaudeDatasette
AnthropicAgentic AI FoundationLinux Foundation+3
prompting, ai-policy, coding-agentsΒ· Published and added 22 Oct 2026
BeNice! is a post-training method for looped language models that broadcasts the final recurrence's KV-cache into the first recurrence, so one cache can serve all recurrences. It follows a first stage that uses stochastic depth in pretraining to induce contractive residual dynamics. It matches standard finetuning's unshared validation perplexity and composes with multi-latent attention, Kimi delta attention and fp8, mxfp8 and nvfp4 cache quantization. In simulated low-memory long-context inference it raises cache hit rates by up to 31.4% and throughput by up to 23.6%, and cuts p90 time-to-first-token by up to 83.6%.
Progress Probes (P^2) trains linear probes on a vision-language-action model's own activations, using demonstrations, to estimate task progress, then flags executions that deviate from expected progress. Tested on the open-source pi0.5 across simulated and real-world manipulation tasks, it matches or beats VLM-based monitors and failure-data-trained methods at failure detection, without an external foundation model.
SecLog-Bench is a benchmark for LLM reasoning over cybersecurity logs, built on an extensible simulation platform covering 29 attack scenarios. It combines audit logs, application logs and network traffic, with attack-stage labels and step-by-step evidence chains. Eight LLMs were tested on post-detection attack-progress reasoning: the best reached 44.44% accuracy, 45.07% Macro-F1 and 54.87% Stage Reasoning F1.
LLM
SecLog-BenchStage Reasoning F1
AI Security, Blue Team & DetectionΒ· Added 23 Oct 2026, published 22 Oct 2026
SecLog-Bench is a benchmark for LLM reasoning over cybersecurity logs. A simulation platform with 29 attack scenarios generates audit logs, application logs and network traffic, with attack-stage labels and step-by-step evidence chains. Eight mainstream LLMs were tested on attack-progress reasoning: the best reached 44.44% accuracy, 45.07% Macro-F1 and 54.87% Stage Reasoning F1. Correct predictions often came with flawed reasoning.
LLM
SecLog-BencharXiv
evals, researchΒ· Added 23 Oct 2026, published 22 Oct 2026
SIM-UT (Simulation of User Behaviour for Triage) is a multi-turn, simulation-based method for finding failure modes in LLM medical triage chatbots. It was applied to three WhatsApp health chatbots (India, South Africa, Kenya) across 5,000+ multilingual simulated turns, comparing MedGemma with ChatGPT and Claude. The main finding is de-escalation risk: agents that first flagged emergencies agreed with users who minimised symptoms, such as vaginal bleeding in pregnancy.
MedGemmaChatGPTClaude+2
SIM-UT
evals, ai-safety, researchΒ· Added 24 Oct 2026, published 21 Oct 2026
Simon Willison's link post on the 21 October 2026 PHP security release, which fixes two remote code execution flaws across the PHP 8.x line. The CVE IDs leaked on 20 October with a wrong CVSS of 10.0 (actual 9.2), and Calif accidentally published a proof-of-concept video. It also gives Wiz Research's timeline, with unauthenticated RCE confirmed against Drupal and WordPress in August 2026.
PHPWordPressCVSS+1
Wiz ResearchCalifPHP Foundation+3
AppSec, Threat IntelΒ· Published and added 21 Oct 2026
Tutorial for training a walking policy for Microduck, an open-source bipedal robot from Pollen Robotics and Hugging Face, in NVIDIA Isaac Lab 3.0 with Newton. It covers RSL-RL training with the backlash preset, BAM actuator models and domain randomization, then ONNX export (61 observations, 14 joint-position actions) and deployment with robotctl. Policies can be shared on the Hugging Face Hub or tested in Microduck Arena.
SBFS-FT (symmetry-based few-shot fine-tuning) adapts the MatterSim 1M universal machine-learned interatomic potential with low-rank adaptation on a small set of compounds sharing one crystal symmetry. Tested on half-Heusler (space group 216), space group 225 and spinel (227) datasets, it cut lattice thermal conductivity error 2-3 times versus the base model. It matched per-compound fine-tuning while taking 5-7 times less compute for 100 compounds.
MatterSimLoRA
MatterSim 1Mmoment tensor potentialspolyMLPs
research, trainingΒ· Added 23 Oct 2026, published 21 Oct 2026
SimReady Central Library is an NVIDIA-led, free and ungated collection of more than 2,000 simulation-ready OpenUSD assets on Hugging Face, covering robot arms, grippers, warehouse equipment, racks and cooling gear. Assets are validated against SimReady Foundation profiles, and NVIDIA USD Search lets developers and agents find them by natural language. The post shows GPT-6 Astra building a pick-and-place scene in Isaac Sim and comparing AI factory layouts.
OpenUSDNVIDIA USD SearchIsaac Sim+5
NVIDIASimReady CentralSimReady Foundation+14
training, agentsΒ· Published and added 21 Oct 2026
vlt 1.5.0 adds a `vlt skills` command that finds agent skills (SKILL.md folders) in installed JavaScript dependencies and links them into directories coding agents read, such as .agents/skills and .claude/skills. Selections are saved as allow-skills in vlt.json, tied to the lockfile version, and kept current on install and update. Without a saved selection, linking defaults to :not(:malware) using Socket malware alerts. pnpm 12.11.0 added similar linking.
vltnpmJavaScript+7
SocketSarah Goodingagent-browser
coding-agents, promptingΒ· Published and added 21 Oct 2026
vlt 1.5.0 adds a `vlt skills` command that discovers agent skills (SKILL.md folders) shipped in installed JavaScript dependencies and links them into directories read by coding agents such as .agents/skills and .claude/skills. Links follow the lockfile version, and selections are saved as allow-skills in vlt.json. Without a saved selection, linking defaults to :not(:malware), using Socket malware alerts. pnpm 12.11.0 added similar support.
vltSocketnpm+3
Sarah GoodingSKILL.mdvlt skills+4
Supply Chain, AI SecurityΒ· Published and added 21 Oct 2026
Broadcast and Trace from Lattices is a 2026 ePrint preprint by Jeffrey Champion. It builds collusion-resistant broadcast and trace with public tracing from attribute-based encryption with succinct ciphertexts and lockable obfuscation, both known from decomposed LWE. Ciphertexts are sqrt(N)Β·poly(Ξ», log N) in size, with poly(Ξ», log N) keys. It is the first such scheme based on a simple, instance-independent lattice assumption.
Jeffrey ChampionNassar, Waters and Wupositional witness encryption
TorchTPU is Google's open-source project for running PyTorch apps on its TPUs, announced the day after Amazon released TorchNeuron for its Trainium chips. Google says migration needs "a simple configuration change" with no rewrite. It worked with Meta, Hugging Face, and the vLLM and SGLang teams on compatibility, and added an agent suite that helps migrate apps and fix errors. The post is brief and gives no benchmarks.
PyTorchvLLMSGLang+4
GoogleAmazonNVIDIA+5
inference, ai-infraΒ· Added 21 Oct 2026, published 20 Oct 2026
Battle Drones is a browser-based Command & Conquer style RTS that Simon Willison built over 12 iterations and 321 commits, mostly with Claude Code and Claude Opus 5.5. The post gives key prompts and transcripts per version, including the "Make it a LOT more fun" run in Ultracode mode. It also covers an Agent vs. Agent mode for LLMs, audio-modem multiplayer between nearby phones, and lessons on deterministic simulation and code-generated 3D models and music.
Claude CodeTypeScriptThree.js+6
Simon WillisonClaude Opus 5.5Claude Fable 5+5
coding-agents, agentsΒ· Published and added 20 Oct 2026
Cosmos3-H-Dreams is NVIDIA's real-time generative surgical simulator, trained on clinical cholecystectomy video and kinematics from the CMR Surgical Versius platform. It builds on the 16B Cosmos3-H-Embodied-Nano, distills a bidirectional teacher into four-step and two-step causal students with Self-Forcing, and streams through FlashDreams on one RTX PRO 6000 GPU. Checkpoints and a recipe for custom data are released.
FlashDreamsFadeMemHugging Face+1
NVIDIACosmos3-H-DreamsCosmos-H-Dreams+6
models, multimodal, trainingΒ· Published and added 20 Oct 2026
InterWebCoder is a framework for interactive webpage code generation by multimodal LLMs. It synthesizes self-contained webpages with executable test scripts from recorded interaction trajectories, repairs failures using runtime evidence, then trains with test-ID-weighted SFT and GRPO using test pass rates and visual similarity as rewards. The pipeline yields 1,736 verified samples from 2,058 tasks. A 9B model scores 48.54 on IWR-Bench and 48.42 on Vision2Web, and reward latency drops over 97%.
GRPOvision-language modelLLM
InterWebCoderIWR-BenchVision2Web+3
research, training, multimodalΒ· Added 22 Oct 2026, published 20 Oct 2026
RAISE (Risk-Aware Intensive Supplemental Execution Framework) is a just-in-time framework that predicts which project-days introduce a vulnerability, to decide where to spend extra fuzzing. The dataset joins OSV vulnerability-introducing commits with OSS-Fuzz coverage data for 15 C/C++ projects (15,210 project-days, 347 positive). A Random Forest using VCCFinder static features plus fuzzing features reaches ROC-AUC 0.74. In a simulation over 968 vulnerabilities, Simple-Regression needed 91.60 extra fuzzing sessions per detected vulnerability, against 171.89 for the Median heuristic.
The Ledger supply chain incident of October 2026: on-chain analysis estimates $93 million was stolen after devices sold by the reseller CryptoBilis carried an implant with an eSIM. The implant watched the screen during setup, captured the 24-word recovery phrase and sent it over LTE, without touching the Secure Element. Tether froze about $10 million USDT. Reports from Changi Airport and MediaMarkt stores remain unconfirmed.
Ledger
CryptoBilisMark KarpelèsTether
Blockchain, Hardware & RF, Supply ChainΒ· Added 20 Oct 2026
Simon Willison's 19 October 2026 link post points to Jeremy Scheff's essay "My obituary for programming". Scheff argues coding agents remove what career programmers enjoyed: paid work on tricky logic puzzles where they set strategy and timelines. Willison notes that LLMs suit programmers who care about building working systems, while those who loved solving the problems face a real loss. The post is a short quote with no data.
LLM
Simon WillisonJeremy Scheff
ai-policy, coding-agentsΒ· Added 20 Oct 2026, published 19 Oct 2026
simonwillison.netarticleπ‘οΈ Security
