Creator of evolved, whitehatstoic and Individuation Lab
A family of 16 projects to learn, share, create and play, from AI alignment lessons to browser games.
See the project → whitehatstoic.comSoftware, security and AI safety: building software, testing it, and researching AI safety.
See the project → individuationlab.comAI alignment through Jungian individuation: testing whether alignment can be grown, not just enforced.
See the project →A second, honest personality trained into the model reviews its answers and reveals hidden misalignment: 96% detected, where standard audits find almost none.
Read the paper ↗ RLLM · 2024Reinforcement Learning using Layered Morphology: a 1.5B model trained on stories in layers held off 67.8% of 1,500 jailbreak attacks.
Read the experiment → MRL · 2025Aligning language models through layered worlds: synthetic environments stacked in order, each one building on the last.
Read the post ↗A research proposal, an AI Alignment Awards 2023 official entry.
What happened to me, and the future of this blog.
Books, documented studies, criticisms and overarching themes.
How Anthropic's agentic coding tool became the inflection point - and how Ralph loops and GSD turned a non-engineer into someone who ships enterprise software with 10 AI agents.