<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Éthiqueia — Newsletter</title><description>Every proof drop, in your inbox cadence. Reproducible local-first AI research, no hype.</description><link>https://yandesbiens.com/</link><language>en</language><item><title>#2 — proof drop #3 — a cheap router decides whether the locality bet pays</title><link>https://yandesbiens.com/newsletter/002-fmm-router/</link><guid isPermaLink="true">https://yandesbiens.com/newsletter/002-fmm-router/</guid><description>A near-free centroid router recovers 98% of the oracle&apos;s recall at ~60x flat-scan speed — when the memory is separable. As topics overlap, routing (not retrieval) becomes the bottleneck and the win evaporates.</description><pubDate>Sat, 27 Jun 2026 00:00:00 GMT</pubDate><category>memory</category><category>memory</category><category>retrieval</category><category>routing</category><category>benchmark</category></item><item><title>#1 — A 24 GB model on a 24 GB GPU</title><link>https://yandesbiens.com/newsletter/001-proof-drop-ufm/</link><guid isPermaLink="true">https://yandesbiens.com/newsletter/001-proof-drop-ufm/</guid><description>Proof drop #1 — UFM runs a model larger than VRAM on a single 4090, plus the honest case where it doesn&apos;t help.</description><pubDate>Sat, 27 Jun 2026 00:00:00 GMT</pubDate><category>memory</category><category>ufm</category><category>benchmark</category><category>memory</category></item></channel></rss>