A condensed summary of Anthropic's September 2026 AI misuse report, covering 117 findings on disrupted threat actor activity.
Bryan Cantrill responds to AI doomer claims, urging experts to be circumspect when raising alarms about AI risks.
A detailed review of macOS 27 Golden Gate, covering Apple silicon transition, AI features, and the Liquid Glass design refresh.
A skeptical take on AGI doomsday claims and Anthropic's report on real-world AI misuse in cyberattacks and surveillance.
Armin Ronacher experiments with Pangram, an AI text detector, by using an LLM to recreate a tweet and analyzing the results.
Explains why a model's knowledge cutoff is a ceiling, not a snapshot, and how this affects AI agent design.
AI decision controls: separating model proposals from authorized actions with evidence, permissions, and verification.
Explores how the double-slit experiment offers a mental model for evaluating AI behavior under different conditions, not quantum AI.
Schrödinger's cat as a metaphor for AI uncertainty: why plausible outputs aren't proven, and how to separate generation, verification, and authorization.
AI memory architecture: context, RAG, and persistent state — designing reliable use of information across model parameters, retrieval, and workflow state.
AI feedback governance: control which observations become persistent changes, where they apply, and who can authorize them.
OpenRouter's automatic routing can cause inconsistent model behavior across providers; learn to control provider selection.
A software engineer's reflection on overcoming the existential crisis caused by AI coding agents and adapting to change.
AI reward design: define business outcomes, not easy metrics. Separate permissions, verification, escalation, and efficiency to avoid gaming.
Simon Willison announces release of llm 0.35, a command-line tool for accessing large language models, with support for new OpenAI model gpt-6-astra.
OpenAI introduces GPT-6 Astra for developers, highlighting improved attention to detail, prompt understanding, and 3D model generation capabilities.
Release notes for llm-openrouter 0.7.1, a plugin for LLM models hosted by OpenRouter, including a performance fix.
Release notes for llm-anthropic 0.28, adding Claude Fable 5.1 support, default reasoning traces, and a new ClaudeRefusal exception.
A personal essay on how AI tools like Claude are influencing the author's writing style, blending linguistics with reflections on technology's impact on creativity.
Explains reasoning models vs LLMs, philosophizes about building from scratch, and shows Python/PyTorch setup with uv.
