3 ms·
OpenAI GPT-3 medical chatbot told a mock patient to kill themselves
- uberman 6y agoGPT does not have a commonsense understanding of the inputs given to it or the text it generates. It just a statistical model of word co-occurrences. At best it is a tool to marvel at from an entertainment perspective. At worst, some foolhardy researchers confuse it for something that might have the wisdom and sensitivity to council the sick. Come on, of course it eventually told someone to kill themselves. The topic of suicide (assisted or not) should is not come as unexpected from a statistical model of word co occurrences related to patient care.