2 ms·
There is no AI, there are LLM models, that people hardly understand. Modern solutions are to AI what Alchemy was to Chemistry. Sometimes they get some results b
by belter 3y ago
There is no AI, there are LLM models, that people hardly understand. Modern solutions are to AI what Alchemy was to Chemistry. Sometimes they get some results but nobody really sure why. The little that OpenAI mentioned about how they achieved GPT-4 performance, says it was by trying a bunch of things and dismissing others. They probably also themselves not sure how they managed to pull it of, meaning that, if for example an increase in parameter size or other settings, kick in negative effects that actually reduce the model performance, they will also be at a loss where to go next.