Comment by evilmathkid

no, most models on kaggle are finetuning during test time, (including LLM based approaches)

Pure frontier LLMs dont, but thats because nobody knows how to make it work cleanly and at scale. Once someone makes it work, it will be deployed