Its The Graph again (not the METR graph, the one from the o1 launch). Although no logarithmic decay of ability with increasing compute...

Noam Brown clarifies that the model shown in a compute-vs-capability graph is a general-purpose LLM, not a math-targeted scaffold, and that it hasn't been pushed to its limits on open problems. The comment implies headroom remains on frontier reasoning benchmarks. Relevant context for those tracking o1-class scaling behavior.