Marcus Zethraeus
133 posts


Asymmetric hardware scaling is here. Blackwell tensor cores are now so fast, exp2 and shared memory are the wall. FlashAttention-4 changes the algorithm & pipeline so that softmax & SMEM bandwidth no longer dictate speed. Attn reaches ~1600 TFLOPs, pretty much at matmul speed! joint work w/ Markus Hoehnerbach, Jay Shah(@ultraproduct), Timmy Liu, Vijay Thakkar (@__tensorcore__ ), Tri Dao (@tri_dao) 1/






Can AI help reshape how we handle global conflicts? Our latest case reimagines the US–China tariff dispute through AI-assisted mediation - built on clarity, neutrality, and progress. Because how we talk matters 🤝 Read more: resolvewith.ai/test-cases #AI #Geopolitics

What if Ross and Rachel had used ResolveWithAI? No shouting. No “we were on a break.” Just clarity, accountability - and less group trauma. 👀 Read the case: resolvewith.ai/test-cases #RossAndRachel #WhatIf #ConflictResolution #ResponsibleAI #FriendsTV

🚀 It’s alive! ResolveWithAI is officially open for testing. We’re not just launching—we’re debugging in real-time (with your help). Every test, every click, every “wait, what?” moment makes it better. Test it. Break it. Tell us what you think. Sign up: resolvewith.ai/login






















