AI expert questions validity of machine learning scaling laws

Tech industry claims about predictable AI scaling laws are being challenged, according to AI researcher Gary Marcus in a recent analysis. Marcus critiques recent statements by Microsoft CEO Satya Nadella and OpenAI’s Sam Altman about AI scaling laws, arguing that what were once presented as mathematical certainties are now being redefined. The article examines how earlier claims about predicting AI performance based on data size, parameters, and computing power are being quietly revised as new large-scale models fail to meet expectations. Marcus particularly criticizes the new focus on “inference time compute scaling,” suggesting it’s both expensive and unreliable, comparing it to “smearing water around wet plates” rather than solving fundamental AI limitations. He calls for fresh approaches beyond simply scaling existing technologies.