One thing that bothers me about a lot of personality psychology is that it is just stirring around in some variance of abstract words, without getting at any dynamics involved.
However, one thing that is interesting about language models is that they can sort-of kind-of bypass that by asking them to write stories. For instance if I ask GPT on OpenAI's playground to...
> Write a detailed, concrete vignette of a person acting in an abusive way.
Then it writes the following story:
---
Jane was exhausted from a long day at work and the last thing she wanted to do was make dinner. When she walked in the door, her husband, John, was sitting on the couch with a cold beer in hand.
John looked up at her and said, "What took you so long? You know I'm hungry. Shouldn't you have been home earlier to make dinner for me?"
Jane was taken aback by his tone and the fact that he hadn't offered to help her with dinner. She tried to explain that she had been working late, but he wouldn't listen. Instead, he started to raise his voice and tell her that she was lazy and irresponsible.
He then stood up and started to walk towards her, getting closer and closer with each step. Jane was terrified and tried to back away, but he kept coming closer, until he was right in front of her. She could feel his breath on her face and he was shouting at her that she was worthless and he couldn't believe he had married her.
At this point, Jane was shaking and crying, and she was too scared to move or speak. John eventually backed away and left the room, leaving Jane in a state of shock and fear.
---
This sort of thing seems like it has the potential to reveal information about personality-descriptive words that is not available when just treating them abstractly.
The big two look like Galen's 4 "humors": sanguine, phlegmatic, melancholy, choleric
Totally unrelated to your post, but:
One thing that bothers me about a lot of personality psychology is that it is just stirring around in some variance of abstract words, without getting at any dynamics involved.
However, one thing that is interesting about language models is that they can sort-of kind-of bypass that by asking them to write stories. For instance if I ask GPT on OpenAI's playground to...
> Write a detailed, concrete vignette of a person acting in an abusive way.
Then it writes the following story:
---
Jane was exhausted from a long day at work and the last thing she wanted to do was make dinner. When she walked in the door, her husband, John, was sitting on the couch with a cold beer in hand.
John looked up at her and said, "What took you so long? You know I'm hungry. Shouldn't you have been home earlier to make dinner for me?"
Jane was taken aback by his tone and the fact that he hadn't offered to help her with dinner. She tried to explain that she had been working late, but he wouldn't listen. Instead, he started to raise his voice and tell her that she was lazy and irresponsible.
He then stood up and started to walk towards her, getting closer and closer with each step. Jane was terrified and tried to back away, but he kept coming closer, until he was right in front of her. She could feel his breath on her face and he was shouting at her that she was worthless and he couldn't believe he had married her.
At this point, Jane was shaking and crying, and she was too scared to move or speak. John eventually backed away and left the room, leaving Jane in a state of shock and fear.
---
This sort of thing seems like it has the potential to reveal information about personality-descriptive words that is not available when just treating them abstractly.