ChatGPT is a conversational AI system: a user writes a natural-language prompt, and it generates a response. It can carry on a coherent exchange, draft content, and adapt its wording to a requested communication style. Its accessible interface lets people use generative AI without specialized technical knowledge. section 5.0 section 1.2
ChatGPT belongs to a line of chatbots that began with ELIZA, a 1966 program that responded using rules triggered by words in a user’s message. Modern chatbots such as ChatGPT instead use large language models trained on vast amounts of text and other data. Fluent responses are a substantial advance over ELIZA’s simple exchanges, but fluency alone does not establish understanding. section 1.4 section 5.6
Its usefulness depends on the task and the cost of an error:
- For drafting and everyday assistance, ChatGPT can quickly produce plausible text, including material for content creation and customer-support applications. section 5.0
- For factual research, a confident answer may be fabricated. Repeating a question can produce different claims rather than a dependable correction. section 5.3.2
- For consequential decisions, check claims and citations independently before relying on them. A legal brief included fictitious cases supplied by ChatGPT, even after the chatbot affirmed that a questioned case was real. section 5.10
One question, several answers
Asked the same biographical question in separate sessions, ChatGPT gave multiple incorrect accounts before producing an accurate one on the fifth attempt. The example shows why a polished answer should not be treated as verification. section 5.3.2
Figure 5.1 ChatGPT prompt, which resulted in incorrectly identifying the author as the founder and CEO of a company he was not affiliated with.
The prompt-and-response figure shows an incorrect identification presented as a factual answer. section 5.3.2
Use ChatGPT to help generate and shape language, not as an authority that has checked its own claims. Its apparent command of words can coexist with errors about facts and real-world context. section 5.3.2 section 5.6
