Tag
ETH Zurich researchers show that fine-tuned RoBERTa models can infer users’ Big-Five personality traits from ChatGPT chat logs with up to 44 % above-random accuracy, highlighting privacy risks of conversational AI.