Ahmad Osman on why local AI is catching up

Ahmad Osman ran two packed local AI workshops at AIEWF, using a hardware comparison site to let attendees benchmark DGX Spark, AMD Strix Halo, and other devices against frontier cloud models. His core claim: open models lag closed ones by 4–8 months, and the gap keeps shrinking. He argues most people miss that hosted products like ChatGPT bundle search, tools, and infrastructure around the model. His company Osmantic is building an open-source deployment system to fill that end-to-end gap. The audience ranged from a student shopping for her first AI machine to an Intel executive asking about Windows UX and enterprise model routing. Osman also noted a modern phone can now run a model that outperforms cloud systems from two years ago.

Why it matters: A Latent Space interview with Ahmad Osman that pins the open-source lag at 4-8 months and backs it with live hardware benchmarks. Concrete numbers and a first-person experiment keep it grounded. It's one person's read, not a product launch, so it caps below 78, but the signal ...