OpenAI President: Models "Not Scaling as Expected" Are Almost Always Bugs

haider1 · x · 2026-07-04

OpenAI President Greg Brockman stated that whenever they encounter a situation where a "model is not scaling as expected," it is ultimately discovered to be an implementation or mathematical bug—meaning the math is wrong, or the implementation doesn't match the math—rather than hitting a genuine capability ceiling.

Original post →

More from Models

Models channel →