When chatting with @bot, be careful with interruptions. As people, we collaborate by chipping at the problem, throwing in ideas, and proposing corrections. On the surface, it seems to work really well wth @bot too. However, every time you interrupt, @bot has to stop, go back, and reread the whole conversation. Out goes a large chunk of your token limit.
So think hard about your prompt. Ideally, you preempt all those corrections. If this is not directly erroneous and must be stopped, collect your notes and wait for @bot to complete the activity. Then correct/redirect/improve in one prompt
Everyone seems to post those prompts that somehow will make @bot work more efficiently. You put that prompt in; it goes and burns some tokens that you don't have, and comes up with a very plausible explanation and a fix. Then it happily goes back to wasting your tokens. No matter how many times you ask. If you are coding and running out of your weekly allowance in 2 days, just go back to the old routine and run your chat in Cursor. Yes, I know it sucks, but there you can manage models, select lesser effort level, context window, start new chats, and compact your conversation. Everything that Grok Bot does, but in a very expensive way