“There has not been a significant step-function improvement in AI model capabilities for approximately a year, since the release of models with vision capabilities like GPT-4V.”