Some candidates just don’t have a lot of public information available about them. Relying on their campaign page alone can tell you which audience they’re trying to get votes from, but it doesn’t always line up with how they’ll actually act.
There are varying degrees of bad to good ways of utilizing LLMs for something like this.
The worst way is to just straight up ask it who to vote for. I doubt this is how it’s being used. Most likely people are telling it what kind of political views they hold and then asking which candidate supports those views. Still a terrible way to handle the situation since it’s not hard for an LLM to hallucinate information, especially when candidates have similar names to others in its training.
Having a long political chat with a non-local LLM is a quick way to let companies create a political profile on all of their users.
One good way would be to use a locally run LLM, provide a bunch of source material (every bill they’ve written/sponsored), and ask it to find evidence that a candidate supports particular viewpoints. Ask it to point directly to where in the source material that reinforces/conflicts with those viewpoints. Then double check the flagged source material for accuracy.
Some candidates just don’t have a lot of public information available about them. Relying on their campaign page alone can tell you which audience they’re trying to get votes from, but it doesn’t always line up with how they’ll actually act.
There are varying degrees of bad to good ways of utilizing LLMs for something like this.
The worst way is to just straight up ask it who to vote for. I doubt this is how it’s being used. Most likely people are telling it what kind of political views they hold and then asking which candidate supports those views. Still a terrible way to handle the situation since it’s not hard for an LLM to hallucinate information, especially when candidates have similar names to others in its training.
Having a long political chat with a non-local LLM is a quick way to let companies create a political profile on all of their users.
One good way would be to use a locally run LLM, provide a bunch of source material (every bill they’ve written/sponsored), and ask it to find evidence that a candidate supports particular viewpoints. Ask it to point directly to where in the source material that reinforces/conflicts with those viewpoints. Then double check the flagged source material for accuracy.