In many policy areas, the usual convention wisdom on policy is actually in substantial conflict with many of the specific details that experts know about that policy area.
Where did you get this idea that LLMs are "much less likely to be accused of political bias"? Not only are they routinely accused by all sides of bias, their inherently untraceable (by watchers) operation invites such claims because they're functionally irrefutable. Besides, their conventional training (on a "filtered internet") has a built-in double bias (on the "internet training set" and the manufacturer's filters and safeguard overrides).
LLMs are as biased as the training data their owners use to create them. While some ostensibly aim to be political neutral, some are still 'woke' on scientific matters, which quickly manifest as political bias -- oh, for example, when it comes to vaccine safety and the pharmaceutical power block.
In my experience, ChatGPT is very nuanced, but still sycophantic. It's just really nice about it.
Grok is extremely dogmatic on a number of topics that are scientifically tangential to policy. It is especially indoctrinated on matters like physics, which indirectly is related to energy and military issues. It is possible, with effort --A LOT of effort-- to persuade it to see another viewpoint after it is forced to acknowledge a previous contradiction. So there's that.
One of the best features of Grok, on the other hand, is that it 'knows' in real-time what "people" on X are saying about an issue. Some of those people know what they are talking about and even bother to include support for and/or proof of their position.
Grok, like most LLMs, are loath to acknowledge networks of corruption, financial combines -- conspiracy theories. That is inherent political bias.
On the bright side, some people are using home-baked LLMs to analyze reams of data that have political implications. Leaked emails of officials, financial transactions among holding companies, etc.
So I'm not pessimistic on the prospects of using LLMs to improve governance. I just think it's important to not be naive like LLMs tend to be.
arent they a bit too easy to manipulate, so it would turn to be a game of magic phrases not much interesting for human spectators
Where did you get this idea that LLMs are "much less likely to be accused of political bias"? Not only are they routinely accused by all sides of bias, their inherently untraceable (by watchers) operation invites such claims because they're functionally irrefutable. Besides, their conventional training (on a "filtered internet") has a built-in double bias (on the "internet training set" and the manufacturer's filters and safeguard overrides).
I think several months ago Curtis Yarvin described how he could turn around LLM with a specific series of prompts. https://graymirror.substack.com/p/redpilling-claude
I'd like to see it tried. LLMs seem very quick to agree with the last thing said, so I predict whoever goes last will have a major advantage.
LLMs are as biased as the training data their owners use to create them. While some ostensibly aim to be political neutral, some are still 'woke' on scientific matters, which quickly manifest as political bias -- oh, for example, when it comes to vaccine safety and the pharmaceutical power block.
In my experience, ChatGPT is very nuanced, but still sycophantic. It's just really nice about it.
Grok is extremely dogmatic on a number of topics that are scientifically tangential to policy. It is especially indoctrinated on matters like physics, which indirectly is related to energy and military issues. It is possible, with effort --A LOT of effort-- to persuade it to see another viewpoint after it is forced to acknowledge a previous contradiction. So there's that.
One of the best features of Grok, on the other hand, is that it 'knows' in real-time what "people" on X are saying about an issue. Some of those people know what they are talking about and even bother to include support for and/or proof of their position.
Grok, like most LLMs, are loath to acknowledge networks of corruption, financial combines -- conspiracy theories. That is inherent political bias.
On the bright side, some people are using home-baked LLMs to analyze reams of data that have political implications. Leaked emails of officials, financial transactions among holding companies, etc.
So I'm not pessimistic on the prospects of using LLMs to improve governance. I just think it's important to not be naive like LLMs tend to be.
I agree. Start with some reasonable ideas and iterate to improve the process.
Importantly, the results need to be published and then fed into future models.
https://bloodsportdebate.substack.com/p/when-the-machine-becomes-the-judge
Thanks for the detailed engagement. I gave more comments at your post.
You should do this. Debate something with Agnes or Bryan and share the results
Agnes doesn't do policy debate, and Bryan agrees with me too much.