We’d like to use ChatGPT to search for real estate listings from free text typed in by the user. The text will include properties such as price, location, amenities and could also include geographical conditions such as “less than a 5 minute walk from a public transport hub”. So some data would be searched from our DB and other data from publicly available mapping information. What’s the best approach for this?
Hi @miguelgarvie, and welcome to the community!
I think you just described a solution that benefits greatly from the recently announced ChatGPT Plugins.
This is potentially problematic. Bear in mind that LLMs are trained with content to a specific date. If something important changed on a map after the LLM was trained, it would have no knowledge of this information. This is why the implementation of your plugin must factor these and other requirements.
Have you had some success with this? I’ve been working on a similar concept and I’ve made a little headway, but not too much.?
Hi yes some success using the gpt-3.5-turbo engine. Feeding real estate entries in the first question and then interactive from then on. It occasionally makes use of knowledge of the world such as public transport stops, though sometimes just says it doesn’t have real time information. How have you been getting on?