A study (not yet peer-reviewed) done by a team of researchers from the Swiss Federal Institute of Technology concludes that between 33% and 46% of workers on platforms like Amazon's Mechanical Turk had used AI models to automate work typically done by humans needed to train AI systems.
@stefan@nik "At one point, fake views were so prevalent that some engineers were concerned YouTubes algorithm for detecting them would begin to treat the fake views as default and start misclassifying real ones." 😲
@stefan I see two solutions to this problem. Pay your workers a living wage, giving them the time and resources to produce higher quality data. Or hire a team of data scientists making 150k a year each, to build another AI model to clean your dataset for you. Except the model only has about a 73% test accuracy and was trained using data labeled by the same workers you are underpaying.
I wonder which option will draw in more investors?
Add comment