Hacker News new | past | comments | ask | show | jobs | submit login

I think that response is a hard coded filter and not a self generated assertion. I imagine it's a stop to make sure people don't project emotions or become attached to it. It responds similarly if you ask it questions regarding the tone/sentiment of the generated text. It responded similarly when I tried forcing it to classify its own personality, however when I asked questions about other fictional AI like Glados from portal, it had no problem answering. This disagreement only indicates that OpenAI spent a considerable amount of energy with adversarial prompts.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: