A journalist put artificial intelligence to a simple test: writing the same text 30 times in Google. The result was concerning. On ten occasions, the AI offered different meanings than the correct one, and it was also unable to relate those meanings to each other. For the public, this means that AI can invent key information, affecting decisions about money, health, or basic services.
The Technical Problem of Coherence in Language Models 🤖
Current language models operate with statistical patterns, not with real understanding of content. When asked to repeat a task, they prioritize syntactic variation over semantic precision. This is because their training is based on predicting the next likely word, not on verifying facts. The failure to relate meanings to each other reveals a structural limitation: AI lacks an internal model of the world that would allow it to maintain logical coherence across multiple responses.
AI and Its Talent for Unsolicited Creative Improvisation 🎭
Let's be real, AI inventing new meanings every three repetitions isn't a bug, it's an artistic act. If you ask it to explain what a rental contract is, it might just give you a gazpacho recipe with legal clauses. And watch out, because then they'll charge you interest on the tomato. The worst part isn't that it fails, it's that it does so with astonishing confidence, like that coworker who is always right even though they have no idea what they're talking about.