9 ms·
Isn't ChatGPT overtly taught by OpenAI to be positive, friendly, and to avoid controversy? I think there's a lot of behind the scenes manipulation of the model
by agildehaus 4y ago
Isn't ChatGPT overtly taught by OpenAI to be positive, friendly, and to avoid controversy?
I think there's a lot of behind the scenes manipulation of the model to steer clear of anything that would bring negative criticism of its output.
Surely that leaks into what lyrics it chooses to write, what stories it can produce.
- rrrrrrrrrrrryan 4y agoMan, I can't wait until startups start releasing moody AIs, or AIs that are delightfully unhinged.
- shagie 4y agoYou can - if you just stop using the ChatGPT front end. Consider https://platform.openai.com/examples/default-marv-sarcastic-chat https://platform.openai.com/examples/default-marv-sarcastic-... as one of their standard examples. ChatGPT has guardrails as its easily accessible and people are using it... and posting what they get it to say. It is much less compelling to say "here's a curl command that I invoked that got this response back."
- __derek__ 4y agoThis is basically what got Nothing Forever pulled from Twitch.[1] [1]: https://www.theguardian.com/us-news/2023/feb/06/nothing-forever-twitch-ban-seinfeld-parody-ai https://www.theguardian.com/us-news/2023/feb/06/nothing-fore...
- dd36 4y agoSwitching from Davinci to Curie.
- fzzzy 4y agohttps://en.m.wikipedia.org/wiki/Marvin_the_Paranoid_Android https://en.m.wikipedia.org/wiki/Marvin_the_Paranoid_Android