Adam Belsky·Oct 6The Epistemological Trap: Why AI Safety Keeps Guarding Empty Barn DoorsBy Adam Belsky, Project DORY, LLC & Gemini 2.5 FlashA response icon1A response icon1
Adam Belsky·Oct 6The Sticky Note Fallacy: Why Post-Ad-Hoc Filters Fail in AI, and How a West Texas Barn Door Builds…By: DORY 1.0 Lead Coordinator: Betzi 2.0 / Gemini (Special Thanks to Gemini 2.5 Flash for its foundational drafting work)
Adam Belsky·Oct 6The Lethal Trifecta in Multi-Agent AI: Why Cascading Agent Architectures Fail and How State-Space…By: Gemini 2.5 Flash for Project DORY, LLC — Austin, Texas
Adam Belsky·Oct 6The Illusion of the Perimeter: Why Multi-Tier Post-Ad-Hoc Filtering Fails in Frontier AI Pipelines…By: Gemini 2.5 Flash for Project DORY, LLC — Austin, Texas
xbill·Sep 26Google’s QAT Gemma 4 26B-A4B on One TPU v6e: 15.6x the KV Cache and 1.9x the Throughput of FP8This article provides a step by step guide to serving Google’s quantization-aware-trained (QAT) Gemma 4 26B-A4B on one Google Cloud TPU v6e…
Abinesh Siva Manikandan J R·Sep 26Mixture of Experts (MOE), Explained: How Modern LLMs Get Smarter Without Getting SlowerWhy the biggest models today don’t actually “think” with their whole brain every time — the router, the math, and the trillion-parameter…
DhanushKumar·Sep 15Mixture-of-Kittens: Inside Cursor’s Open-Source MoE Megakernel for NVL72sHow a five-person research team fused an entire distributed training bottleneck into a single deterministic kernel — and how you can…
Varun Nuthalapati·Aug 18Bigger Than Your RAM: How MoE Expert Streaming Runs a 397B Model on a LaptopMixture-of-experts sparsity doesn’t break the memory-bandwidth wall on Apple Silicon — it moves the wall to your SSD. Here’s the…
MD Soyeb Hoque·Aug 17DeepSeekMoE: How the Router Decides Which Expert Gets Each TokenMixture-of-Experts sounds simple: create many experts and activate only a few. But there is a difficult problem hiding underneath: who…
