Marily Nika: evaluation

Transcript context

…How much data do you think you need for AI ML to have a chance to contribute? You have a heuristic of if you have anything less than this, it's not going to work at all. This is a good question and it honestly depends on what I'm what I'm trying to do. If you're trying to classify, if the photo is a category dog, obviously even if you have, I don't know like 15, 20 labeled photos, what's going to work. But if you want to create voice recognizers or complicated NLP applications, you're going to need thousands of [inaudible 00:16:19] of data. And this what's making this not be easy? AI systems are not easy to develop. There is a life cycle of a machine learning project and after scoping you need to figure out, "Oh god, how much data do we need? Where do I find this data as well, how much data?" Sometimes I've seen people synthesizing their own fake data just so that they can have something to train with and test their models. But the exact amount is hard to be defined, especially from a [inaudible 00:16:50]. I'm sure data scientists have a different opinion. Yeah, my guess is most startups are going to have nowhere near enough data to build their own model and make it something really interesting. So do you have a thought on when it makes sense to try to build your own model, try to train your own GPT type thing versus use something that's already out there like say GPT or aJourney or all those guys.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.