How do Large Language Models compare to NLP toolkits for NLP tasks?

I need to do some NLP on text in a number of different languages (English, Spanish, Russian etc). I've experimented using spaCy, stanza and NLTK, as well as some LLMs like ChatGPT, Bard, LLaMa 2 and GPT-4, to do things like lemmatization and POS tagging.

In my experimentation, GPT-4 with adequate prompting outperformed everything else in every language. I wasn't able to spot any errors.

The other LLMs were more or less on par with NLP toolkits: LLMs were a …

bard chatgpt english etc experimentation gpt gpt-4 language language models languages languagetechnology large language large language models lemmatization llama llama 2 llms nlp nltk spacy spanish tagging tasks text