Connectionists: The psychology of GPT

[Rothganger, Fred]

One psychologist spent time with ChatGPT, applying a battery of traditional psychological tests. Here are the results: https://arxiv.org/abs/2309.07683

"Variability notwithstanding, neither GPT model is a picture of good mental health. If human, we would say they exhibit low self-esteem, possible gender dysphoria, disconnection from reality, and are overly concerned with others’ opinions. They are also narcissistic. And, if GPT 3.5’s MMPI-2 scores are to be taken seriously (i.e., are deemed valid), that model displays severe psychopathy. GPT – both versions tested in this work – appears to put on a positive face, so to speak, despite this unhealth. This positive veneer could be due to OpenAI’s attempts to ensure the models don’t produce offensive responses or could be more fundamental to the model. To this point, results of the personality measures beg significant questions about GPT’s training data."

[https://static.arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png]<https://arxiv.org/abs/2309.07683>
Assessing the nature of large language models: A caution against anthropocentrism<https://arxiv.org/abs/2309.07683>
arxiv.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20231103/2ad97054/attachment.html>