It's a really interesting question: the converse of this is its _really_ tricky nailing all this down, much harder than you'd think -- see the GPT4 GA thread ea

It's a really interesting question: the converse of this is its _really_ tricky nailing all this down, much harder than you'd think -- see the GPT4 GA thread earlier this week for people who swear up and down the OpenAI API acts in bizarre ways, you'll note that it almost seems fantastical that it could be that bizarre: it is fantastical. If you're truly used to it, it reads like "hey I have auto retries enabled in my HTTP client framework. p.s. whats 'context size'?" But Langchain is far, far, _far_ away from being truly helpful with that. It's glorious that we're 5 months into GPT-4 and most people either got bored or are building on rickety rushed structures in Python to rush out a proof of concept web app.