Remix.run Logo
modriano 3 days ago

I just start new chats whenever the chat gets long (in terms of number of tokens). It's kind of a pain to have to form a prompt that encapsulates enough context, but it has prevented me from hitting the Pro limit. Also, I include more questions and detail in each prompt.

Why does that work? Claude includes the entire chat with each new prompt you submit [0], and the limit is based on the number of tokens you've submitted. After not too many prompts, there can be 10k+ tokens in the chat (which are all submitted in each new prompt, quickly advancing towards the limit).

(I also have a chatGPT sub and I use that for many questions, especially now that it includes web search capabilities)

[0] https://support.anthropic.com/en/articles/8324991-about-clau...

greenie_beans 3 days ago | parent [-]

> It's kind of a pain to have to form a prompt that encapsulates enough context, but it has prevented me from hitting the Pro limit. Also, I include more questions and detail in each prompt.

i get it to provide a prompt to start the new chat. i sometimes wish there was a button for it bc it's such a big part of my workflow

greenie_beans 3 days ago | parent [-]

also, do any data engineers know how context works on the backend? seems like you could get an llm to summarize a long context and that would shorten it? also seems like i don't know what i'm talking about.

could the manual ux that i've come up happen behind the scenes?