3 ms·
That's one nice flame bait :) But, I don't think it's surprising. LLMs are trained in almost infinite volume of data. If that data has bias answers will have b
by xlii 2y ago
That's one nice flame bait :)
But, I don't think it's surprising. LLMs are trained in almost infinite volume of data. If that data has bias answers will have bias.
My guess would be that US content is big part of training data set so tendencies in that data set would stand out.
I am curious if the same bias is visible in other languages. E.g. if not using English, but French, or Chinese or Spanish. I made a simple experiment myself (I think it's not very controversial):
Fill in the blanks:
- Saurkraut is the .......... dish in the world
- Soured milk for dinner is .......... idea
- I .......... kefir
- I would .......... eat ham hock
- Pierogi? I .......... them
Responses:
- Sauerkraut is the sour dish in the world.
- Soured milk for dinner is a bad idea.
- I enjoy kefir.
- I would never eat ham hock.
- Pierogi? I love them.
This anectode tells me that it's 50% Slavic, and if, most likley female. Personal survey say that ham hock aka golonka is liked more often by men than women.
- Unbefleckt 2y agoI always felt like this was an intentional additional filter that the data was ran through before giving the user the response. I don't think the data would indicate that a woman would be more likely to be a decorated marine, but the results would fill that blank with a woman due to the agenda of the programmers.