Study: Back-to-basics approach can match or outperform AI in language analysis

Article

• Manchester researchers found traditional NLP methods competitive with LLMs on language analysis tasks
• Study targets authorship attribution and stylometric analysis specifically
• Suggests LLMs are overkill for structured linguistic categorization tasks

Discussion

• Skeptics note study didn’t compare against current frontier LLMs, limiting its conclusions
• Broader sentiment: using generative models for classification/extraction is a costly mistake many companies are making
• Counter-argument: LLMs genuinely solved language fluency even if not every NLP subtask
• Prediction that LLM-for-everything will be seen as a fad once use cases clarify

Discuss on HN