This!!! ->
Just like there is no best code repo, THERE ARE NO BEST AI MODELS as each use-case is different and each company has its own constraints. It's all about finding the right model or architecture for your specific situation, which is IMO an important part of the job of any AI builder (and can't be outsourced to others). They do that mainly by: #1 Looking at trending + usage + community feedback on HF: https://lnkd.in/eSYz_Dc #2 Looking at public evaluations like the leaderboards on HF: https://lnkd.in/eVECVXS9 #3 Running their own evaluations depending on their constraints (latency, cost, accuracy...) #4 testing in prod Anything else?