And how can I reduce the chance for ChatGPT to have one?
#why does ChatGPT have "hallucinations"?
lethal dock
tulip domeBOT
Hi @lethal dock. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
ChatGPT, like other language models, can sometimes generate outputs that are not based on the input provided or the data it was trained on. These outputs, often referred to as "hallucinations", can include made-up facts, quotes, or citations. This behavior is a known limitation of the model and can occur due to various reasons, such as the model's inability to access real-time information from the internet or its inability to verify facts[^1^].
To reduce the chance of hallucinations, you can manage the conversation history effectively. The AI model itself doesn’t store a history of what you talked about. Individual AI model calls are standalone, and must be told again what you were talking about recently[^2^]. You can enhance the illusion of memory and reduce the token usage of particular turns by replaying to the AI just the user input and not the AI response, summarizing, or using a semantic retrieval of older conversation turns[^2^].
However, it's important to note that the backend for conversation management in ChatGPT, which decides what past conversation is most informative and should be given to the AI model for the most recent question, is not fully documented[^2^]. Therefore, while these strategies can help, they may not completely eliminate the occurrence of hallucinations.