With all the excitement around LLMs, are people still labeling data manually for NLP tasks? Or has this been largely replaced with prompting, few-shot models, and/or LLM-based labeling?

Would be curious to hear what other people's experiences have been like trying LLMs in these contexts.