"It’s baldly incorrect to say that 'the use of language like "mistake" and "apologize" is commonly used in human-computer interactions.' I have been using computers for 40 years and literally no software has ever used such language in its interactions with me." #AI #ChatGPT http://www.briggeman.org/wordpress/posts/talking-chatgpt-with-chatgpt/
@JBriggeman The challenge/back-down pattern that you got there seems to be common in newer version chats. #AI #ChatGPT
The older chats were more entertaining because the LLMs would simply re-assert incorrect statements and sometimes appear to pick fights over them.
I wonder if the designers have updated training data, more heavily weighting sources with lots of apologetic backing down language.
Without an internal representation of reality, I'm not sure how LLMs can ever find a good balance between these two extremes.
