Yesterday night I was reading about Efficient Influence Computation in OpenLedger and honestly half of the stuff went above my head at first 😅. I thought I would close the article after two minutes because technical AI topics usually become too much for me very fast. But this one kept me reading little longer.
From what I understood, the system is trying to figure out which datasets and contributors still influence model outputs even during inference, without making computation super heavy all the time. I could be wrong on some parts though. The idea itself felt interesting because usually people only talk about bigger models and faster results again and again. This felt more practical maybe. Some sections were still confusing not gonna lie.