ChatGPT and other AI chatbots sometimes say things that don't make sense because they are guessing what to say next based on what they have learned. They use a lot of probabilities, which are like chances or odds, to decide what to say. When they talk too much without getting feedback, they can start saying weird things that don't match the conversation. This is called hallucinating. It happens because the chatbots don't know everything and sometimes they make mistakes when they try to answer questions that are new or different from what they have seen before. Read from source...

1. The title is misleading and sensationalist. It suggests that all AI chatbots hallucinate frequently or severely, which is not true. Hallucination is a rare occurrence in most cases, and it depends on the model architecture, training data, and input context. The article should acknowledge this nuance and provide more balanced information about the phenomenon.
2. The author relies heavily on Yann LeCun's explanation, without questioning his credentials or the validity of his claims. LeCun is a prominent figure in AI research, but he also has conflicts of interest with Meta Platforms and its ambitions to develop advanced AI models. His perspective might be influenced by his own interests and goals, and readers should be aware of this potential bias.
3. The article does not provide enough technical details or examples to help readers understand what hallucination means in the context of AI chatbots. It uses vague terms like "nonsensical", "irrelevant", and "disconnected" without explaining how these qualities are measured or identified. The article should include some concrete examples of hallucinated responses or outputs, and compare them with expected or correct answers, to illustrate the phenomenon more clearly.
4. The article does not explore the causes or consequences of hallucination in AI chatbots. It briefly mentions that hallucination happens because of autoregressive prediction and the accumulation of errors, but it does not explain how these mechanisms work or why they lead to hallucination. It also does not discuss the impact of hallucination on the performance, reliability, or trustworthiness of AI chatbots, nor the possible solutions or mitigations to reduce or prevent hallucination.
5. The article ends with a vague and irrelevant question about user safety, which seems out of place and sensationalist. It does not explain how or why AI chatbots might endanger users, or provide any evidence or examples of such cases. The article should either address this issue more seriously and thoroughly, or remove it altogether.