One big harm is that GPT4 can trivially cheat essays in most contexts. OpenAI planned to address this with a classifier (https://openai.com/index/new-ai-classif

One big harm is that GPT4 can trivially cheat essays in most contexts. OpenAI planned to address this with a classifier (https://openai.com/index/new-ai-classifier-for-indicating-ai-written-text/ https://openai.com/index/new-ai-classifier-for-indicating-ai...), but it didn't work and they withdrew it without offering any better ideas - so now evaluations of unmonitored writing are indefinitely and likely permanently broken for everyone.