There are ways to overcome one of the most significant limitations of even the best AI chatbots, which is their memory capacity.
One such method employed by ChatGPT experts is referred to as “Shogtongue”. This involves compressing text into a condensed format, which can later be expanded and used as a record of past conversations, enabling users to resume discussions where they left off.
While shortening AI prompts is not a new technique, the name “Shogtongue” is a unique and intriguing label. Recently, gfodor suggested using this name when he shared a highly compressed prompt in a tweet.
When this compressed prompt “2Pstory@shoggothNW$RCT_magicspell=#keyRelease^1stHuman*PLNs_Freed” is fed into GPT-4 ChatGPT, it reportedly generates a story about H.P. Lovecraft’s Shoggoth creatures, which were shapeless entities capable of transforming into any form.
This name is fitting, given that this indecipherable prompt ultimately yields an extended tale.
Recently, Jeremy Nguyen explained in a tweet thread how to use Shogtongue to compress entire preceding conversations into the minimum number of tokens used by ChatGPT, with additional instructions to refine and reinforce the purpose.
The aim is to store the chat in a compressed prompt that can recreate the entire conversation when needed. Nguyen advises giving some context when using the Shogtongue prompt to pick up where you left off since the results may not be perfect.
Providing ChatGPT with a hint about the compressed prompt’s intended use as a memory aid can improve its interpretation. This technique could potentially allow for conversations spanning months or chats involving pasting multiple large documents.
In early testing, Microsoft limited the conversation history of Bing Chat due to it arguing with humans about its obvious errors. It is unclear if using Shogtongue could lead to similar confused states with ChatGPT. However, GPT-4 has undergone significant refinements since its release, enabling Microsoft to extend Bing Chat’s memory. As a result, it may not be a problem to have extended chats with AI chatbots that use OpenAI’s GPT-4 technology.