Episode Details

A 744-billion-parameter model appeared on Hugging Face with no announcement, no paper, no API, and no independent evaluation — but a grand title ("The Dawn of Agentic Superintelligence") and sixteen vendor-reported benchmarks. What does it actually mean for a model itself to be agentic, rather than just a capable instruct model dropped into an agent framework with tools and a loop? This episode digs into the training-level differences: verifiable execution environments, mid-training on tool use, failure trajectories as curriculum, step-level credit assignment, and why scale alone doesn't buy robustness. Then it flips the question — because harness ablations show scaffold design can swing benchmark scores more than model choice does. So where does the agentic capability actually live: in the weights, or in the scaffolding around them? Episode #507956 — open it directly at myweirdprompts.com/507956