Frequent AI-assistant hallucinations, paid experience hard to keep stable

Ask the same question and an AI assistant admits its mistake one moment, then trots out boilerplate to rationalise itself the next, making it hard to trust. Local users note that mainstream AI chat tools commonly hallucinate, frequently getting it wrong when calculating numbers or looking up shop or product information, with some inclined to agree with the questioner rather than point out an error.

Some say the free tier has been heavily stripped back and constantly prompts an upgrade to paid, while the paid experience is not necessarily stable and model quality fluctuates with each version; functions like image generation and fact-checking often refuse to respond because of safety restrictions. For Hong Kong users willing to pay, the lack of a reliable, transparent and consistent set of criteria for choosing an AI assistant is a clear painpoint in emerging-technology consumption.