Dynamically choose how many recent messages to include for context, based on total token lengths. Instead of just taking the recent 10 messages like I did before, which meant sometimes I wasn't using as much context as I had available, and other times I was exceeding the limit of tokens which broke ChatGPT #marcbot