ChatGPT chatting with itself

Max. Tokens:

Just for fun: this implements two instances of ChatGPT chatting with itself. You can give two different system messages - that is messages saying ChatGPT how it should behave in general, and then press start to see it jabbering away with itself, until the number of turns is reached. If you get something interesting, please please send me a copy! You can use the "Clipboard" button to coy the chat to the clipboard.

You can either type out your own system messages, or use one of the examples to get some ideas. There are also some advanced options: The maxtokens parameter is a hard limit of the number of tokens that ChatGPT is allowed to generate, which is usually something like half to 3/4 of a word. For special experiments, you can also set the temperature parameter (between 0 and 1), which controls how "creative" ChatGPT is allowed to be. The higher the temperature, the more creative, but also the more nonsensical. It's possible to switch the system message to a user message, which seems to be stronger in some cases. Normally we transmit a "Hi!" (invisibly) as first message, which seems useful to initiate an actual chat. You can omit that with the "Omit Hi" checkbox if that is a problem for your use case.

You will need a OpenAI API Key to use this. If you don't have one: login there with your OpenAI ChatGPT account and go from the account menu into the "View API keys" to create one. Don't worry - this app just stores that in your browser and transmits it to the ChatGPT api, it doesn't send it to me or something. :-)

Have fun! :-)