Post ATmTHAJ66O9w1FEZP6 by AFoucart@social.sciences.re
 (DIR) More posts by AFoucart@social.sciences.re
 (DIR) Post #ATmRsxBV8jB4PrZWzY by tante@tldr.nettime.org
       2023-03-19T17:32:07Z
       
       0 likes, 0 repeats
       
       More than the bullshit generator the manual rule-based filters #chatGPT uses are interesting. They tell you a lot about how heavy-handedly OpenAI intervenes to make their network look acceptable.
       
 (DIR) Post #ATmSdRkvo9x6YKJWL2 by kfort@social.sciences.re
       2023-03-19T17:40:33Z
       
       0 likes, 0 repeats
       
       @tante Did you find information about them? How they handle this?
       
 (DIR) Post #ATmSr75nQmumWJadxA by AFoucart@social.sciences.re
       2023-03-19T17:43:01Z
       
       0 likes, 0 repeats
       
       @tante is there any actual information about those ? It seems obvious to me as well that at least some of the behaviour is hard-coded rather than "learned" but I haven't seen anything on what they've actually donne.
       
 (DIR) Post #ATmSuaitimwRHqXPFY by tante@tldr.nettime.org
       2023-03-19T17:43:15Z
       
       0 likes, 0 repeats
       
       @kfort You can only stumble on them. Asking GPT-4 to access a URL triggers a filter. Certain asks do.
       
 (DIR) Post #ATmSxy0raSUqO5YClk by tante@tldr.nettime.org
       2023-03-19T17:43:36Z
       
       0 likes, 0 repeats
       
       @AFoucart you can only explore them by trying.
       
 (DIR) Post #ATmTHAJ66O9w1FEZP6 by AFoucart@social.sciences.re
       2023-03-19T17:47:45Z
       
       0 likes, 0 repeats
       
       @tante yeah that's what I thought. Obfuscation is the order of the day at openai...
       
 (DIR) Post #ATmUPcFEDMSAw919ns by Soyweiser@masto.ai
       2023-03-19T18:00:26Z
       
       0 likes, 0 repeats
       
       @tante I'd assume if you try to get it to say certain things which are banned in Germany you would run into those filters very fast. Esp after the microsoft Tay chatbot debacle.