Search Results: AIScaling

New MIT research suggests massive AI models may soon yield diminishing performance gains, while efficiency breakthroughs could empower smaller, cheaper alternatives. This challenges the trillion-dollar infrastructure boom as experts question the sustainability of current scaling obsessions.

Epoch AI's latest data reveals training compute for frontier AI models is growing at 4.4x annually, with costs soaring into the hundreds of millions per model. This breakneck scaling is driven by ballooning hardware clusters and surging investment, outpacing efficiency gains and raising urgent questions about economic and environmental sustainability.