🤯 This is a wild tale of how you can use LangChain to interface (Chat)GPT with other tools such as wikipedia or calculators. This article was written before they announced ChatGPT plugins, but it feels more powerful in a way, because you can now interface it with whatever you decide to program (internal tools, preferred data providers, etc.)
While I don't know yet what to make of the final remarks (Will there be a GPT-N that can build its a GPT-N+1?), the research into this is just beginning. Some people compare the current large language models (LLM) to compilers; now we can also compare them to fuzzy API connectors. Besides thinking about REST for your APIs, you might also start to consider how to interface your tools with language models!
(Matt Webb now likes to think of himself as an AI sommelier 😁)
(Also posted on my LinkedIn feed)