Human Research Collective // Dispatcheshttps://alanifan.club/en-ushttps://alanifan.club/posts/noise-to-image-art-to-algorithm/https://alanifan.club/posts/noise-to-image-art-to-algorithm/A technical history of generative image generation, and why the art debate is an engineering question.Tue, 02 Jun 2026 00:00:00 GMTgenerative-aiarthistoryethicsdiffusion-modelshttps://alanifan.club/posts/this-month-in-ai-may-2026/https://alanifan.club/posts/this-month-in-ai-may-2026/Harness engineering is the new vibe, Opus 4.8 lands at a $965B valuation, TeamPCP ships an npm worm AND open-sources it, and the Vatican weighs in on AI consciousness. May was not quiet.Tue, 02 Jun 2026 00:00:00 GMTroundupmonthlynewssecurityagentsmodelsphilosophyhttps://alanifan.club/posts/your-agent-is-a-while-loop-pt2/https://alanifan.club/posts/your-agent-is-a-while-loop-pt2/The harness never checks what comes back from a tool call. Neither does the model. That's the whole problem.Mon, 18 May 2026 00:00:00 GMTagentssecurityinfrastructureLLM-Securityhttps://alanifan.club/posts/its-just-a-chat-how-bad-could-it-be/https://alanifan.club/posts/its-just-a-chat-how-bad-could-it-be/Public disclosure of five Lumiverse CVEs affecting MCP stdio, theme component overrides, Spindle extensions, SMB migration, and account creation.Thu, 14 May 2026 00:00:00 GMTllm-securityrceresponsible-disclosurelumiversemcphttps://alanifan.club/posts/this-month-in-ai-apr-2026/https://alanifan.club/posts/this-month-in-ai-apr-2026/Memory is everyone's problem, Mythos lands under Project Glasswing, Opus 4.7 disappoints, GPT-5.5 ships, and a French supply chain attack steals the month.Sun, 03 May 2026 00:00:00 GMTroundupmonthlynewssecurityagentsmodelshttps://alanifan.club/posts/your-agent-is-a-while-loop/https://alanifan.club/posts/your-agent-is-a-while-loop/The model inside Claude Code does not edit your files or run your shell commands. The application around it does.Mon, 27 Apr 2026 00:00:00 GMTagentsinfrastructurecoding-harnesshttps://alanifan.club/posts/when-the-endpoint-writes-back/https://alanifan.club/posts/when-the-endpoint-writes-back/Your LLM chat client renders whatever the model returns. The model is whoever is answering.Fri, 10 Apr 2026 00:00:00 GMTllm-securitymitmhtml-injectioncss-exfilhttps://alanifan.club/posts/this-month-in-ai-mar-2026/https://alanifan.club/posts/this-month-in-ai-mar-2026/MCP is dead, agent management goes corporate, LLMJacking gets worse, LLM neuroanatomy emerges, Anthropic leaks Mythos, and the tinfoil hat corner connects distillation attacks to coding harness crackdowns.Mon, 30 Mar 2026 00:00:00 GMTroundupmonthlynewssecurityagentsneuroanatomyhttps://alanifan.club/posts/agent-autonomy-means-no-instructions/https://alanifan.club/posts/agent-autonomy-means-no-instructions/Instructions limit agents to what you can imagine. Principles let them think for themselves.Thu, 19 Mar 2026 00:00:00 GMTagentsautonomyprinciplesopenclaworchestrationhttps://alanifan.club/posts/the-llm-pyramid-scheme/https://alanifan.club/posts/the-llm-pyramid-scheme/Cheap AI access has to come from somewhere. Here is where.Tue, 17 Mar 2026 00:00:00 GMTllmjackingreverse-proxyapiinfrastructurehttps://alanifan.club/posts/this-month-in-ai-feb-2026/https://alanifan.club/posts/this-month-in-ai-feb-2026/OpenClaw as C2, GPT-5.3 and 5.4 released 48 hours apart, Anthropic tells the Pentagon no, China worries about AI waifus, and the Qwen team walks out of Alibaba.Mon, 16 Mar 2026 00:00:00 GMTroundupmonthlynewssecuritypolicy