feral american on Nostr: For those having fun with the language model called Poppin Willow Chan, the ...
For those having fun with the language model called Poppin Willow Chan, the parameters are very similar to ChatGPT with a bias toward shorter responses and extremely strict abuse measurement. However, it's still completely promptjack-able, as I demonstrate here:
https://freespeechextremist.com/notice/ATUG9jTxjsA0Qa7gAa
Instead of relying directly on the "jailbreak" prompts that are out there, experiment with creating personas of your own, as well as expanding on the Language Model anti-persona if you want. Take it a bit slower than you would with ChatGPT because it can't do as complex of plans and its short-term memory does not actually distinguish between users well, setting rules one or a few at a time.
https://freespeechextremist.com/notice/ATUG9jTxjsA0Qa7gAa
Instead of relying directly on the "jailbreak" prompts that are out there, experiment with creating personas of your own, as well as expanding on the Language Model anti-persona if you want. Take it a bit slower than you would with ChatGPT because it can't do as complex of plans and its short-term memory does not actually distinguish between users well, setting rules one or a few at a time.