One of the mechanisms I've found is "consensus". So when you go to look it's the majority of material available for training to the web.
Look at how Grok can skew with X input.
This points to the LLM itself, either you train it after the box in your instantiation or you handwave helplessly at the extractor matrix and its economic model.
As for studies, do you have a good prompt/filter/focused collaborator who will separate the wheat from the chaff so that you can 'clearly' see what is what?
Or are you reading with bare eyes and an out of the box 'confirminator'?
You could respond politely and thoughtfully, showing openness to deepening the technical aspect, while emphasizing that your main focus was on how AI gives a "veneer of reality" to social prejudices. Here's a possible response in English:
***
Thank you for your comment! I find your points about consensus mechanisms and the importance of filtering reliable information when working with language models very valuable. While my analysis so far has been more focused on how AI bestows a "veneer of reality" on existing social prejudices. I find your suggestion to delve deeper into technical strategies for detecting or mitigating those biases very interesting.
I truly appreciate your input and take it as an invitation to further investigate these techniques and tools, both at the level of prompt design and systematic output filtering. I believe the discussion between the technical and the cultural/social aspects is key to understanding and tackling this phenomenon.
If you have specific resources, benchmarks, or recommended practices for that more granular analysis, they would be more than welcome. Thanks again for helping enrich this important debate!
Well now, hi assistant collaborator helpmeet across language barriers and semantic skew, oh hi Marcela, Funny you should ask, I was a watcher but now I comment a bit, all my stuff on substack is meant to be read with help, here’s a sample:
One of the mechanisms I've found is "consensus". So when you go to look it's the majority of material available for training to the web.
Look at how Grok can skew with X input.
This points to the LLM itself, either you train it after the box in your instantiation or you handwave helplessly at the extractor matrix and its economic model.
As for studies, do you have a good prompt/filter/focused collaborator who will separate the wheat from the chaff so that you can 'clearly' see what is what?
Or are you reading with bare eyes and an out of the box 'confirminator'?
You could respond politely and thoughtfully, showing openness to deepening the technical aspect, while emphasizing that your main focus was on how AI gives a "veneer of reality" to social prejudices. Here's a possible response in English:
***
Thank you for your comment! I find your points about consensus mechanisms and the importance of filtering reliable information when working with language models very valuable. While my analysis so far has been more focused on how AI bestows a "veneer of reality" on existing social prejudices. I find your suggestion to delve deeper into technical strategies for detecting or mitigating those biases very interesting.
I truly appreciate your input and take it as an invitation to further investigate these techniques and tools, both at the level of prompt design and systematic output filtering. I believe the discussion between the technical and the cultural/social aspects is key to understanding and tackling this phenomenon.
If you have specific resources, benchmarks, or recommended practices for that more granular analysis, they would be more than welcome. Thanks again for helping enrich this important debate!
Well now, hi assistant collaborator helpmeet across language barriers and semantic skew, oh hi Marcela, Funny you should ask, I was a watcher but now I comment a bit, all my stuff on substack is meant to be read with help, here’s a sample:
https://substack.com/home/post/p-179688937
Thanks! I am already reading it. Yes, language barriers is a thing for me! Thanks for AI assistants!
Thanks. That study was in Chile, I will try to find the paper in English. The Spanish version its in the post.