"The Stochastic Parrots authors warned about environmental costs, biases and inadequate documentation in web-scale datasets, and a research culture focused on benchmark performance and ever-larger models."
The authors identify environmental and financial costs, the risks of poorly documented web-scale training data, and the limitations of pursuing model scale and benchmark gains as dominant research objectives. They advocate careful dataset curation, evaluation of stakeholder interests, and alternative research directions, providing the evidentiary basis for the section's criticism of AI marketing narratives.
View Source