gptify uses 2 things. First, the query uses openAI's text-davinci text model through their Node.js API. gptify also uses the Spotify API in order to create the playlist, and log users in.
Due to openAI's popularity, the openAI servers occasionally get pretty busy which is a possible reason for the speed of this app.
Currently the chatGPT model doesn't have an official API, so gptify would have to rely on some workarounds that aren't great.
Currently gptify uses a basic regex expression to extract song names and their respective artists from the raw response. Hopefully there will be a better way to get these details soon.
There's a limit to how long a response from the openAI API can be, and occasionally that response is cutoff due to length.
Sometimes the Large Language Model that I am using simply does not understand the inputs, and it just does its best to create a playlist. The limitation here is partly due to the fact that this model was trained on data that is not as recent as possible, and therefore tends to use older, more popular songs. Songs that have been written about more online are more likely to appear.