Post ATkHvrhyZL4sRWakLo by Cassidy@sleepy.cafe
 (DIR) More posts by Cassidy@sleepy.cafe
 (DIR) Post #ATkHdhN8GBM5AoBCXA by Inginsub@clubcyberia.co
       2023-03-18T16:28:00.273722Z
       
       1 likes, 0 repeats
       
       @Cassidy I'm not really sure about that but the GPT pay-per-token system counts both prompt and output tokens
       
 (DIR) Post #ATkHvrhyZL4sRWakLo by Cassidy@sleepy.cafe
       2023-03-18T16:31:21.590195Z
       
       1 likes, 0 repeats
       
       @Inginsub rip thread broken, what are you responding to? Also I saw people on reddit getting mad that they were "wasting their tokens" whenever chat-gpt hit them with the ethics response lol.
       
 (DIR) Post #ATkHz2wJGgb691oTXk by Inginsub@clubcyberia.co
       2023-03-18T16:31:54.375310Z
       
       1 likes, 0 repeats
       
       @Cassidy it was a direct tag
       
 (DIR) Post #ATkI2tgdbz4SbdJpnU by Cassidy@sleepy.cafe
       2023-03-18T16:32:38.276284Z
       
       1 likes, 0 repeats
       
       @Inginsub wait what are you not sure about
       
 (DIR) Post #ATkICIeg9IuceNt0BU by Inginsub@clubcyberia.co
       2023-03-18T16:34:15.221028Z
       
       1 likes, 0 repeats
       
       @Cassidy that it charges you for input too. It seems like that's the case, but the marketing material is not clear about that
       
 (DIR) Post #ATkIvGhwEi6HdqqyKu by Cassidy@sleepy.cafe
       2023-03-18T16:42:27.419183Z
       
       1 likes, 0 repeats
       
       @Inginsub oh right lol. From looking it up - it seems like it's both input and output they charge for yeah. I guess that makes the jailbreak expensive whenever it has to be repeated- can see why he thought it was just intentional money-wasting spam when he saw the wall of text that is the jailbreak.
       
 (DIR) Post #ATkIwaARN41oHhoNVY by Inginsub@clubcyberia.co
       2023-03-18T16:42:36.153072Z
       
       1 likes, 0 repeats
       
       @Cassidy that's a ripoff. As I understand it, chatgpt is a model on top of gpt-3, it's pretty basic and its operational costs are not as high. When the prompt contains blacklisted words, chatgpt spits out a generic no-no message, but charges you as if the base model received the input and replied
       
 (DIR) Post #ATkIzuyYSkW5gmqJDk by Cassidy@sleepy.cafe
       2023-03-18T16:43:18.180746Z
       
       1 likes, 0 repeats
       
       @Inginsub https://neoteric.eu/blog/how-much-does-it-cost-to-use-gpt-models-gpt-3-pricing-explained/this says it's both input and output and cites the pricing page as the source but it looks like they've redone the page since then, can't find the quote
       
 (DIR) Post #ATkJ59pEgBAhBzyxJg by Inginsub@clubcyberia.co
       2023-03-18T16:44:11.229974Z
       
       1 likes, 0 repeats
       
       @Cassidy unless they have really pajeeted this
       
 (DIR) Post #ATkJIASRdhRiv0WeKe by Cassidy@sleepy.cafe
       2023-03-18T16:46:35.701726Z
       
       1 likes, 0 repeats
       
       @Inginsub yeah I saw people getting mad about it, people who weren't even trying to actively fuck with it either. Like a dude wanting to write a detective novel but something about alcohol or smoking would be involved triggering the block and half of his prompts would be wasted as a result.
       
 (DIR) Post #ATkJJKLWMOglsNq8XI by Inginsub@clubcyberia.co
       2023-03-18T16:46:44.745475Z
       
       1 likes, 0 repeats
       
       @Cassidy yeah, that makes sense, because the to GPT both the input and the output are a continuous stream of text (or, rather, tokens), and it has to process both in pretty much the same way but in different directions