LLMs are a failure. A new AI winter is coming.

Despite initial excitement over transformer-based AI models like ChatGPT, these Large Language Models have a fundamental flaw that makes them unsuitable for most practical applications. Transformers generate plausible-sounding text by predicting the next most likely word, which means they inevitably produce convincing but incorrect "hallucinations" 5-40% of the time without any way to distinguish right from wrong answers. This reliability problem, combined with unrealistic corporate expectations, has led to approximately 95% of business AI projects failing, setting the stage for a market crash similar to the dot-com bubble.