Making AI actually work requires massive amounts of human labor. A Verge article shows workers in Africa labeling images thousands of times daily. The speaker's company does similar work on financial documents, making millions of nuanced distinctions so clients get reliable results.
Chapters
AI is a lot of work. Great article from The Verge on what's required to make AI work. It's a really great message around how difficult it is to get machine learning to actually be helpful. Huge amount of data has to be collected and it has to be collected meticulously. This article talks about how they've got folks in Africa doing a bunch of work to make sure, is this a bike or a fish or a motorcycle? And doing that over and over thousands and thousands of times a day. That's what's required for AI to work. And we know that's true because that's how our AI works, except that we did it in footnotes and financial statements. A lot more difficult than a bicycle versus a motorcycle. We've got hundreds of thousands of iterations, millions of iterations of those nuanced distinctions being made for clients. It was refreshing for me to see an independent source talk about how much work that is and how important that is.