ChatGPT made its own 'Shogtongue' language to blow past the word limit

robot head with exploding parts
(Image credit: Shutterstock)

What language does an AI speak in? You might think it would be a series of beeps and boops or binary code, but it turns out ChatGPT has made its own, and us humans can use it to get more from the AI chatbot. 

The word limit of ChatGPT can cause the AI to stop conversations mid-sentence with the only choice to start a new chat and reset all prior context. There is no official word on what the hard limit of a ChatGPT conversation can be but 8,000 words is the generally agreed limit. But users using the ChatGPT “Shogtongue” language can keep conversations going seemingly indefinitely.

Writer and AI author Jeremy Nguyen discovered this trick by asking the AI to compress their conversation “in a way that is lossless but results in the minimum number of tokens which could be fed into an LLM like yourself as-is and produce the same output.” In essence, he asked the AI to create a password that saved his progress in the conversation.  

Nguyen specified that the encoding did not need to be in a form that humans could read, which is where what has been dubbed “Shogtongue” comes in. This is a code of ChatGPT’s own creation that when copy and pasted into a fresh chat, enables the AI to pick up where it left off, with all the context retained.  

Our own efforts have been less successful. Despite copying Nguyen’s prompt word for word, the AI has merely interpreted it as summarizing what we have discussed so far in smaller terms. On the occasions it has produced an encoded response, we can’t get it to remember the context when opening a fresh chat. Nguyen himself has admitted that this method isn’t completely reliable and won’t work at all on GPT-3.5 or on the GPT-4 API. 

Why does ChatGPT have a word limit?

While it seems cruel to wipe a friend’s memory with every new conversation, this is for good reason. Most conversations will never reach anything near the word limit and Longer conversations with Chatbots have been shown to increase the likelihood of AI “hallucinations” where they start spewing incorrect information or behaving erratically. It was this kind of conversation that saw the new Bing with ChatGPT go off the deep end with Microsoft reluctantly installing a turn limit as a result.

Just like conversations with friends have private jokes and call backs, longer conversations with ChatGPT have more and more context for the AI to draw from and it can become quite complex to sustain a conversation over multiple topics for dozens of ‘turns’. As the technology improves in time, AI conversational skills will no doubt grow, to the point where any word limit is purely arbitrary and unreachable, perhaps using encoding methods similar to Nguyen's efforts. 

More from Tom's Guide

Andy Sansom
Staff Writer – VPN

Andy is Tom's Guide Staff Writer for VPNs and privacy. Based in the UK, he originally cut his teeth at Tom's Guide as a Trainee Writer (go and click on his articles!) before moving to cover all things Tech and streaming at T3. He's now back at Tom's Guide to keep you safe online, and bring you the latest news in VPN and cybersecurity.