AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
How Does a Language Model Use Different APIs?
In the current version of the tool form, we have like one limitation that we impose upon those APIs and that is they need to be text to text. And basically how those models can then call tools is they would first generate text as long as they want. And we allow them to generate one special token, one special kind of word that would signalize, now I want to call an API. We stop it from producing any further tokens. We do that API call. So we call our calculator or we look something up in our calendar, we do a search. We get the result from that, which is also a piece of text. And after that, we let it continue writing text