Kronos — open foundation model that tokenizes candlesticks to forecast markets

I found something that shouldn't exist yesterday and after sharing it a lot of you asked me to explain it. So I will. I'm Giga Chen and I do business in China, a country in which a PhD student released an AI that can predict the markets, stocks, crypto, and so on, and he did it for free. I found and read the official research paper so you don't have to, and by the end of this video, you're going to understand how this works on a near expert level. For context, Cronos, the price-predicting AI was created by a guy named Su, who's a PhD student at Qinghua University, which is like China's Harvard. His lab is the Institute for Interdisciplinary Information Sciences. Cronos is a time series foundation model. A time series is a sequence of data points recorded at successive time intervals, like a heart rate, or a stock chart. A foundation model seeks to generalize. That means work on data it wasn't specifically trained on, like new information in the market. So how does it work? There are five steps, an input, an output, and three steps in the middle, which I am more than happy to break down. And I hope, although he's an AI, I was really enjoyed that breakdown. To build Cronos, they have to tokenize K lines, but what does that mean? K lines are those candlesticks you see on a chart. You have an open, and then you have a high, a low, and a close. You also have volume. Tokenization is breaking down a large, messy stream of information into small chunks called tokens. These are…