Zero shot text classification is the most underrated and useful task for various NLP tasks 
These models can predict what a given text is about without explicitly being trained on the classes 
You can use these models off the shelf for topic classification or sentiment analysis, or even use it for labelling! 
Learn more here 
https://huggingface.co/tasks/zero-shot-classification
@mervenoyan cross encoders look interesting, too
https://huggingface.co/cross-encoder/nli-distilroberta-base
huggingface.cocross-encoder/nli-distilroberta-base · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.
@mervenoyan gotta check this out. this sounds brilliant. they also use textual entailment, fine-tuned to predict entailment relation of 3 million wikipedia articles with their categories.
cogcomp.seas.upenn.eduCognitive Computation Group{{data.basic.author}},{{data.basic.title}},{{data.basic.journal}},{{data.basic.booktitle}},{{data.basic.year}}