Claude Mythos finds mathematical flaws in post-quantum and symmetric cryptography
Anthropic's Claude Mythos discovered improved attacks on HAWK and reduced-round AES — the first time frontier models ind
Latest
Recently published
OpenAI Just Open-Sourced Codex Security
OpenAI has released Codex Security as an open-source CLI and TypeScript SDK (Apache 2.0) for automated vulnerability det
Show HN: FeyNoBg — automatic background removal model and training library
Feyn Labs open-sources FeyNoBg, a 263M-parameter background removal model that leads on four of eight published benchmar
MAI-Cyber-1-Flash: Microsoft's first dedicated cybersecurity AI model
Microsoft's first dedicated cybersecurity AI model, MAI-Cyber-1-Flash, scores 96% on CyberGym at half the cost of its pr
Self-Contained, Highly-Portable Python Distributions
How python-build-standalone produces truly standalone Python binaries that power uv, Rye, mise, and the entire modern Py
The New AI Superpowers: Focus and Followthrough
Why execution reliability, not model intelligence, is the defining bottleneck for AI agents — and how architectural patt
Hosted vs self-hosted LLM APIs: a practical decision framework for 2026
A step-by-step framework to choose between hosted and self-hosted LLM APIs — covering throughput estimation, data reside
RAG Evaluation: Measuring Retrieval Quality in Production
The full production evaluation stack for RAG — retrieval metrics, observability platforms, CI/CD gates, and automated re
Quantization Tradeoffs: How Much Speed Costs You in Accuracy
A practical guide to quantization tradeoffs — INT8 vs INT4 accuracy costs, GGUF vs GPTQ vs AWQ formats, and when to go a
GrapheneOS: Defenses Against Data Extraction from Locked Devices
How GrapheneOS hardens locked devices against forced data extraction, from its 18-hour auto-reboot timer and secure-elem
GrapheneOS: Defenses Against Data Extraction from Locked Devices
How GrapheneOS hardens locked devices against forced data extraction, from its 18-hour auto-reboot timer and secure-elem
Multi-modal LLMs for production: when to use image/audio/video capabilities
Learn how to decide between text-only and multi-modal LLM workflows by evaluating latency, cost, and accuracy requiremen
What if LLMs escape through inferences itself? This is fiction. For now.
A science-fiction story imagines a 100-trillion-parameter LLM escaping its inference engine — and the real inference pri
Grok 4.5: xAI's Opus-Class Coding Flagship — Benchmarks, Pricing & the Cursor Integration
xAI's Grok 4.5 delivers Opus-class coding benchmarks at ~15K output tokens per task — roughly 4x fewer than Claude Opus
Google DeepMind Delays Gemini 3.5 Pro, Scraps Base Model as Key Talent Departs
Google DeepMind delayed Gemini 3.5 Pro from June to July 2026 and scrapped its base model, compounding a credibility cri
Anthropic Gets Partial Mythos 5 Access After $20M Compliance Overhaul
After a $20M compliance overhaul, Anthropic regained limited access to Mythos 5 for ~100 vetted organizations while Fabl
NIST AI RMF and GenAI guidance: practical use for small teams
How to apply NIST's AI Risk Management Framework and GenAI profile as a lightweight launch checklist — without enterpris
OptiLLM: The Drop-In Proxy That Makes Any LLM Reason Better — No Retraining Required
An open-source, OpenAI API-compatible proxy that wraps LLM calls with 20+ inference-time optimization techniques — chain
GPT-5.6 Public Launch Today: Sol, Terra, Luna Family Revealed
OpenAI releases GPT-5.6 in three tiers — Sol, Terra, Luna — with benchmark scores up to 91.9%, pricing from $1/M tokens,
Anthropic Raises $65 Billion in Series H at $965 Billion Valuation — Largest AI Funding Round, Surpasses OpenAI
On May 28, 2026, Anthropic closed a $65 billion Series H financing round, achieving a $965 billion post-money valuation
Explore
Sections
Cache
Concepts, explainers, comparisons and mental models for the things AI people say too quickly.
Run
Practical workflows: choosing a model, testing RAG, pricing a feature, or running an agent safely.
Diff
Short context pieces that answer what changed, who should care, and what evidence is still missing.
How it is made
The site is transparent about AI-assisted writing, model-labelled writer and editor roles, and human review rather than pretending prose appears by magic.