We made* a custom chat bot with ChatGPT to help navigate the UK building regulation approved documents in a conversational way.
Image; The fancy AI created image ChatGPT made for the custom bot.
When I say made, let's be very clear, ChatGPT did all the hard work. let’s call it assembled, which is more suited to the way the new GPT functions from OpenAI work.
I've used ChatGPT as a useful tool to navigate questions on the broad and complex subject of AEC for close to a year. When asking questions on non-subjective AEC subjects with a basis in physics or mathematics, such as building performance, or energy production I have sensed the initial response can feel a little shallow. To the credit of the AI model, questioning what I believe to be a mistake is met with a quick, and lets be fair, unnecessarily polite response of 'I apologise for the confusion. The correct answer is...'. But despite my English sensibility appreciating over apologetic AI bots, my confidence has been wobbled.
The answer to the seemingly broad manner ChatGPT gathers and presents information is answered with the introduction of new custom tools, named rather confusingly GPT's. According to OpenAI, GPT's let anyone create an AI driven ChatBot (as long as you subscribe to the paid premium features) that has it's focus on a particular subject, but importantly can make use of a particular set of information, as long as that information is hosted on an accessible website.
Image; A typical response from the bot. Note that despite the question not specifically asked in relation to UK building regulations, the answer is driven by the bots deliberate bias to answer with it's specific UK regulation data set.