>>60564931This poster is correct. But to tardé like this, you need training data. That would be a general economic news feed from several sources, not whole articles, just short summaries, headlines + short paragraph of the key facts. Many news outlets and aggegators provide such data. Then you need the historic price action from some other paid API. My guess would be to reduce the dimensionality of the news feed with some LLM into dimensions like category, sentiment, perceived magnitude (tariffs on Chyna vs. Canada or Musk drug meltdown vs. the EU making child sacrifice mandatory). Then you need to smooth them out for efficient learning, jumps don't go well with modern higher order gradient methods. Ask AI how to efficiently combine that into a decent input, scaling, weighing, smoothing et cetera. But be careful with price action, scaling different prices down might make it a bit harder for AI to pick up the round number pattern. It might also need some price-derived measures like RSI, but be careful, you'll crush the independence of your input, ruining the numerical condition of your gradient approximations, i.e. lots of error in training steps. Ah whatever, I won't type it all out, ask AI to do it.
