LLMs outperform humans at predicting the likely words for a given context. Their use isn’t wrong, especially, where there’s a language barrier (note Chinese authors). The true crime is when those with the relevant domain expertise don’t make value judgements about AI predictions.
Word Prediction: AI > Humans Domain Expertise: Humans > AI Judgement of Value: Humans > AI Judgement of Ethics: Humans > AI Judgement of Beauty: Humans > AI Judgement of ‘Interestingness’: Humans > AI
To be clear, BOTH the authors and the reviewers are the ones with the relevant domain expertise who need to take responsibility for making value judgements about AI predictions.
Originally posted on Twitter, March 2024.