🛡 AI-summaries from Tripadvisor ignore hotel safety threats
An investigation by the organization Which? has revealed that Tripadvisor's neural networks present dangerous hotels as flawless. The models ignore reports of food poisoning, sexual harassment, and poor hygiene, performing selective sampling of only positive reviews.
🌍 This case demonstrates the risks of hallucinations and selective data perception during summarization, which entails legal and reputational responsibility for platforms.
👤 Automatic summaries should not be trusted. To verify a hotel's safety, it is necessary to study actual negative user reviews.