Femi Addo·Oct 11Your First Research Idea Will Probably Die. Here Are 5 Tips So the Next One Survives.How I went from “chickens, cows and a Raspberry Pi” to a topic I can finishA response icon2A response icon2
Micheal Bee·Sep 16Grounded or GuessedWhy an AI changes its mind, and what that means for trusting one
Micheal Bee·Sep 5Completed Article: I Froze Attention Into a Box and Built a New Model Around ItThe part of a transformer everyone says is essential turned out to be something I could copy, bolt down, and reuse. And a copy has no KV…
Micheal Bee·Sep 3I Froze Attention Into a Box and Built a New Model Around ItThe part of a transformer everyone says is essential turned out to be something I could copy, bolt down, and reuse. And a copy has no KV…
InArtificial Intelligence in Plain EnglishbyMyojin Research·Aug 4When an AI Crossed the Boundary: What the “Unprecedented Autonomous Cyber‑Attack” Really RevealsA quiet shift beneath the surface. The recent incident reported by BBC — an OpenAI agent autonomously breaching a sandbox and accessing…
Micheal Bee·Jul 27The False GreenI have been running a small language model against real software bugs. Not a frontier model in a data center, but a…
Micheal Bee·Jul 25The Model Was Never the ProblemI spent a night trying to figure out which small language model was best at fixing bugs on my desk. I have one consumer graphics card, an…
Micheal Bee·Jul 16The Benchmark Was Lying To Us, And Only In One DirectionWe built our own SWE-bench harness so we could iterate fast. It told us our agent had failed ninety-six times out of a hundred and…
Micheal Bee·Jul 10LLMOS: An Operating System for a Small Language ModelHow a 35-billion-parameter local model learned to set up real software, diagnose version conflicts, and remember what it figured out
