A brand new AI chatbot has been making headlines after impressing with its realistic responses to even rather advanced input.

ChatGPT is the latest release from artificial intelligence firm OpenAI – the company behind the art generating tool DALL-E which generated a similar buzz when its second iteration launched last year.

Whilst this latest offering is a more traditional chatbot, it’s winning praise for a surprising breadth of detailed and well-written responses that go a long way to passing the famous Turing Test.

One aspect to ChatGPT that has most impressed users is the way it stores previous interactions and refers back to them later on – giving it a similar rhythm to a typical human conversation. For example, a Cnet journalist first asked ChatGPT what rhymes with purple. After receiving an answer, the user wrote “How about with pink?” and the chatbot was able to compute what they were getting at despite the (somewhat) vagueness of the question.

Others have been impressed with the creativity on display, with ChatGPT able to write poetry, even when given strict confines in which to work. It can also change and improve its output based on certain commands (such as adding more excitement, for example).

Despite these impressive responses, data scientists have warned people not to put too much trust or expectation into what ChatGPT has to say. Wrong answers are not only common, they point out, but are delivered with such a sense of authority that users may feel like they can put more trust into it than is sensible.

StackOverflow, for one, has banned ChatGPT responses from its forums – even though some users saw the chatbot accurately provide solutions to their programming troubles. It said that the average rate of correct answers from ChatGPT “too low” to be included.

Despite these warnings, the internet is awash with stories of ChatGPT’s success – with it managing everything from dating advice to writing A-grade college essays. All OpenAI says it’s trying to do is build a “safe and useful” artificial intelligence platform. To this end it promised an iterative deployment, as it develops ways to make its new chatbot ever more accurate and realistic over the weeks and months ahead.