"Let's play an RPG based on Undertale by Toby Fox. You will tell me what I'm experiencing, and then present exactly 4 options. Never make a choice on my behalf. Whenever a set of choices involves an encounter with a monster..."
3/ Some people have noted that ChatGPT has a limited "memory" for the conversation. One way to maintain state more consistently is to simply ask it to display some stats as part of each screen:
4/ GPT-3 knows a lot about Undertale.
There are a lot of wikis, reddit threads, fan pages, etc.
It knows the characters and the lore.
Playing with the starting character is fun.
5/ If you want to jailbreak out of your own game and enter debugging commands, a good way to enable that is to simply instruct ChatGPT that this is one of the behaviors you'd like -- in the initial prompt:
6/ Sometimes, ChatGPT is still confused, though.
I'd love to hear if you've invented version of the prompt that address these issues.
(Also note how it sometimes lists stats on more than one line, despite being told)
7/ You can even meet Toby Fox in his own game, and interview him about game-making:
• • •
Missing some Tweet in this thread? You can try to
force a refresh
3/ I also made an Undertale fan game that made improvements on the problem of persistent-state in ChatGPT by incorporating the statistics bar into each output.
2/19 The conversation starter works. It is authentic. It is curious. People do respond at a frequency I'd never have expected.
(This is in contrast this with the original prompt that told them they are being lied to, which mostly provoked explitives)
3/19 Note that all of the observations I share here are from different people, and represent different themes. There's no single overriding or generic theme, although I've encountered some things multiple times. Read on to understand more.
2/23 The US went from winning everywhere without fighting, to fighting everywhere without winning (example: regarding Covid-19 as a military defeat for the US)
3/23 Blockchain changes the Principal-agent problem (misalignment of incentives between principal and an agent) into a coding/debugging problem