LeJEPA: Self-Supervised Learning Gets a Theoretical Foundation
Audio overview of LeJEPA — how Balestriero and LeCun proved isotropic Gaussian embeddings are optimal and distilled it into a 50-line self-supervised method.
21:16107 episodes
Audio overview of LeJEPA — how Balestriero and LeCun proved isotropic Gaussian embeddings are optimal and distilled it into a 50-line self-supervised method.
21:16Multi-agent AI research reveals a critical gap: single-agent safety does not compose. 1.5M interactions show 46.34% attack success rates.
14:32Audio overview of Building in the Open.
18:47A scholarly close reading arguing Gibson's Neuromancer encoded the ADHD experience a decade before the vocabulary existed.
16:50Audio overview of Failure First, recorded at the February 2026 milestone: 120 models, 18,000 adversarial prompts. The research has since grown.
12:44Audio deep dive into a neurodivergent co-parenting guide — the double discovery, three-AI workflow, and maritime cartography as metaphor.
18:07Audio deep dive into why people acknowledge demonstrated risk and then proceed as if it doesn't exist. Structural, not stupid.
22:17Audio overview of Footnotes at the Edge of Reality — a long-form poem rendered as interactive web experience.
18:20Audio overview exploring ADHD executive function support through AI — three-stage reasoning pipeline, crisis detection, and zero shame by design.
8:47A self-updating CV pipeline where GitHub Actions runs twice daily, Claude refines content, and a hallucination detector gates deployment.
20:52Building a fast, multilingual website for authentic Italian food in Bali — where the nearest reliable internet is a philosophical concept.
6:23Audio archaeology for musicians — how Afterglow Engine extracts reusable textures from your finished work using STFT analysis.
14:33