You are viewing a single comment's thread from:
RE: Curation Reward Estimation Tool
This seems like a classic optimization problem.
I can imagine that this could work for a while, but then sometimes things happen that did not happen in the past and therefore are not covered by a model that is based on experiences from the past. Just like the stock markets. Prediction works to some extent, as long as the future works mostly like the past.
In this case, the system would be self adapting. After all, it just looks at a sliding window of what has been successful in the last X days and acts on that information. Something happens that is unexpected? Sure, efficiency drops for a little bit – and then the entire system moves on.
You mention the stock market, I'll counter with the weather. If you want a greater than 70% prediction rate, just assume that tomorrow will be within 4% to 5% similarity of today. Throw yesterday and, do a very basic trend analysis, and you can be over 80%. And that's with a very stupid algorithm.
I'm old-school when it comes to AI and markets. I firmly believe in "good enough" technology. It doesn't have to be perfect – it just has to be good enough. It doesn't have to be good enough all the time, it just has to be good enough for a good enough portion of the time.
You can get away with a lot of slop if that's your target space, and you can get to it very fast, and very cheap.