While there is potential for this to be a bad thing, I think on the whole it's positive. Most/many people don't interact with democracy at all or only "red vs blue" which means increasingly less the more local you get (IMHO). Instead, focusing on what they candidates say/have done and how that aligns with your own views is a far better system.
And maybe, just maybe, it will pull some of these dinosaurs out of the stone age and force them to put up a website with their views (crazy I'm saying this in 2026, it should have been table stakes decades ago). Local races are incredibly difficult to find information on. Your best bet is one of those "voter guides" that a local newspaper might put out and even those aren't great. Using an LLM to scour the internet to make a better-informed decision seems like a huge win over what most people do today.
Ultimate tool for propaganda and manipulation tailored to each person's taste. Highly scalable and can focus on the things you (don't) care about when presenting "both sides".
The stakes are too high for it to not be weaponized, the only question is who will control it.
> Local races are incredibly difficult to find information on
Big time. At best in more rural areas, you will find a facebook page for the candidate. Sometimes, there is nothing they put out and just maybe one of those "we interviewed the candidate and this is what they said" type sites. Sometimes, nothing at all and it is a real shame. It would solve a lot of problems if we knew more about our local school administrations, mayors, sheriffs, etc.
You have to disclose if a site is part of a PAC or candidate's committee due to FEC laws. This is not the case for LLM slop that often intermingles the two.
So yes, it's very very different.
Maybe these LLM companies should also disclose in every chat window on what political spending they have already done:
While this is true, there is a gaping hole in how foreign agents can amplify influencers on youtube and twitter. Probably even facebook, at this point.
I guess FEC laws don't protect you from the pravda network. However, LLM might be not so good at ranking false information, if it is already massively embedded in their training data (LLM grooming).
Still it seems that this is currently not the case [0] (yet).
I guess the risk atm is rather some crazy billionaires embedding their believes in post training and system prompts. But honestly I rather fear tiktok in this regard.
1) People at large seem to prefer large stacks of cash to base morality
2) Employees will be tasked with writing a generic system that affects results, but only a handful of suits will know what it's actually going to be used for
While we can throw around hypotheticals and conspiracy theories about influencing AI results, we know for a fact that search engines and social media are heavily influenced, because there are giant industries around SEO and viral marketing -- plus the huge amount of ads and promoted content, which are literally paying to influence results.
Nothing. And it will happen big time once the AI companies become enshittified. Anthropic and OpenAI won't be able to produce the results that would justify a trillion dollar IPO without doing shady stuff.
> focusing on what they candidates say/have done and how that aligns with your own views is a far better system.
How often have you looked into two candidates at any level above city council and decided that the "red vs blue" assumptions differ from who you would normally vote for? Add in a few specifying sub-labels (MAGA, DSA) and that number probably drops to insignificant.
> How often have you looked into two candidates at any level above city council and decided that the "red vs blue" assumptions differ from who you would normally vote for?
Above city council (or a few levels, at least up to state Rep) are the people I have less issue picking between because more information is available. It's the lower races or things like county coroner that I have much less info about. Some of those roles still have (R)/(D) suffixes on the candidates but I don't feel like that means as much as state senate/reps with a letter after their name.
who's going to cross check it? and even if everyone did, who's going to ensure that everything everything is mentioned, things can fact check just fine with some strategic omissions.
Cross checks will be useless because enough GEO/SEO will be pumped out using generative AI that even what the candidate said will be distorted or "interpreted" in ways that fit the agenda of the organization doing the SEO. "If you give me six lines written by the hand of the most honest of men, I will find something in them which will hang him." -- Cardinal Richelieu.
The best part is that the AI companies get paid for both causing immense societal problems and combating the problems their tools are causing. It's perfectly Kafkaesque and the best part it is going unremarked right out in the open.
you can have multiple agents run reviews and gut checks. open source models, etc. there are ways to go back and forth to get to some relative truth. regardless of complexity of said task, i would trust it a lot more than social media or search engines which are entirely corrupted by ad spend budgets.
Homogenizing opinion though a small number of private companies with their own incentives is a pretty terrible idea, actually.
The current system isn't perfect by any means, but at least there's a wide variety of source and experience. Diversity can provide a buffer against viral problems, be it literal living viruses or information.
Funneling information through a handful of sources is akin to Sinclair Broadcast Group buying up rural news stations. Awful for democracy and diversity. Can put us a single virus away from destruction.
Yea but that’s not happening. The question is how this compares to the status quo. It could easily be much worse, could easily be a fair bit better too imo. Most people don’t know a thing about who they vote for
No problem with two companies owned by sociopaths who largely have the same political ideas, and are deeply tied to the current administration, controlling the tool used by the mass to decide who to vote for? Said tool being trained on reddit, twitter and other trash tier platforms where minorities of radicalised people on both side of the spectrum "discuss" political topics 90% of people do not care about.
not just two companies, by the way. remember grok and its views?
to me it's more alarming not just the cesspool that these are trained on, but subtle bias that can be added by people in charge or their friends... how naive to expect this won't happen
Grok is a good example of how bad the labs still are at any sort of intentional, subtle bias. I'm sure they'll get there eventually, but the best attempts today remain pretty hamfisted.
1) believe or not even if it's obvious to you it is enough to mislead plenty of people. you know how many people fall for obviously scammy ads? Yeah, good news, those people vote
2) it's an example of how biased models are. how bad or good they are now is irrelevant because it'll be different in a year or month. remember if you don't see bias it doesn't mean it's not there.
3) have you checked how much money is going into this, do you think they don't have resources to get extremely good at it. just follow the money.
>> Using an LLM to scour the internet to make a better-informed decision seems like a huge win over what most people do today.
A spin on that I used for my local elections last year: My prompt stated the issues that were important to me and why. I then copied in any election guide I could find from newspapers, blogs, newsletters, endorsements, etc.
Then I asked for 1) which best fits my priorities, and 2) risks for each (eg "says they will support homelessness programs, has no direct experience in an elected position")
> Chatting through all the races and the different candidates, he worked to figure out who best represented his values.
This sounds like it can be a really good thing. In an ideal world it’s a perfect use of LLMs. It helps make the ballot more approachable to people, and it can encourage more people to vote or help people vote less blindly.
At the same time, it also relies on these tools being unbiased adjudicators. And over time I think there’s a greater risk of this straying from the truth.
And so it begins: the outsourcing of research, thinking, and reasoning to AI in lieu of using their own brains, an opaque AI in the hands of trillion dollar corporations with their own agendas and shareholders to please.
In the future, maybe we won't need election analyses to predict political victories, we can just ask all of the major brand of AIs who a voter should choose and then multiply by the market share of each AI, dohohoho.
Nonsense. There's no "using your own brains" in voting in the midterms. You get no information about the vast majority of the candidates, you're just expected to vote as the party tells you to, and the rest of the information you get is from PR firms in proportion to the millions spent to market to you and distort all the information that you could possibly find.
Even if you work your hardest and get a reasonable overview, you've dedicated far more of your time to this than anyone could be expected to (a sign of the leisure and access you have as a privileged middle-class person), and the vote will be necessarily decided by the vast majority who didn't.
People using LLMs for the midterms will be the best informed voters we've had in a long time, even if they use Grok (which will desperately try to lie to them, and won't be the only one.) They should also use LLMs to help them with strategies to get to an actual voting booth, because the parties like to sabotage voting in places they feel they'll be weak.
You seem to think everybody (but you) is a dummy and will just do what the LLM says to do, but were totally comfortable for them just doing what the party says to do. I might use an LLM myself to figure out what's going on with judges, if I can find one with access to the proper databases.
In fact, if you want to do good for the world, get access to good databases with good search engines, hook deepseek up to them, and open it to the public. People need to find out how judges decide, and if there are any complaints against them, etc.
Is there a sci-fi novel yet, where everyone teaches their own AI how to correctly predict how they answer moral questions, and how they prioritize their principles? And then their AI actually represents them in Congress? Representative AI Democracy?
In general I think it's a good thing that people are trying to get more informed before voting, but I worry that if this becomes popular enough, there will be a lot of pressure on the labs to influence what the LLMs say about political questions.
No need. Just create (using an llm) a corpus of websites of "facts" about candidates, made up to your liking, that the LLMs will happily slup up and spit out.
Is there any evidence of those competitors pushing political biases? (Actual biases, not facts like Trump losing the 2020 election and staging an attempted coup)
It's interesting; when I clicked I thought the angle would be "it's dangerous that we let chatbots influence people" but the article seemed to care more about the sycophancy angle: some form of "maybe it doesn't influence them enough"
Depends on how they’re being informed I guess. Homogeneously informed can be okay to pretty bad. From mid-century American television and print media all the way down to North Korea and Nazi Germany.
While the knee-jerk reaction is to roll one's eyes, if it gets people that would otherwise not vote to be a bit educated and go out and make their voices heard, I am all for it. Yes, this has the potential for abuse and needs to be watched closely, but is this worse than the classic "I asked my coworker and..." type of non-involvement many already have? Not really and my hope is that people will only use it as a stepping stone to more research.
Are they becoming educated? Or are they just adopting the bias of the AI? (Which is the bias of its training data and/or trainers.)
Is it worse than "I asked my coworker"? Yes, because maybe six coworkers total are who the entire country is asking. That's too much influence for a program.
The best way for these hosted LLMs to handle this is to follow the example of Tolkien's elves:
> “The choice is yours: to go or wait.”
> “And it is also said,” answered Frodo: “Go not to the Elves for counsel, for they will say both no and yes.”
> “Is it indeed?” laughed Gildor. “Elves seldom give unguarded advice, for advice is a dangerous gift, even from the wise to the wise, and all courses may run ill.”
So much worse than just voting the party slate like they're told, and operating from television commercials.
Everybody should use ChatGPT to help them decide how to vote in the midterms. They should ask for specific references to platforms and tweets, with links that they can follow.
Could the information you get be horribly distorted? Yes. But no more than the information you're receiving from every other source - and with LLMs you can drill down until it runs out of alignment. The TV doesn't take questions.
This can only be illegal.
Law, and all human rights and responsibilitys are built on the idea of humans making indipendent Judgments, free from any coersion or missleading information,, and so to allow giving voting advice from a machine with known, programed in, biases and inhibitory blocks is impossible, and could very quickly lead to total AI bans in many juristictions as there may be no reasonable way to monitor and certify such devices.
IMO for an involved voter, this is actually a good thing. AI can both help you clearly define your priorities and policy preferences via an "interview", and can subsequently find candidates who's publicly stated positions, and past record, align most closely with your priorities.
As always, AI can be used to completely outsource critical thinking, and doing so for elections is extremely concerning. However I would hope that more people use it to make better, more informed decisions, than to blindly choose candidates
To be fair, Americans have previously asked ouija boards, psychics, and the Magic 8-Ball the same question in the past… at least they’re not asking Grok.
I don't believe the American economy ever hinged on the success of the 8-ball, nor the CEO of Hasbro (owners of the Ouija trademark) called before Congress to question whether the boards could build bioweapons.
While there is potential for this to be a bad thing, I think on the whole it's positive. Most/many people don't interact with democracy at all or only "red vs blue" which means increasingly less the more local you get (IMHO). Instead, focusing on what they candidates say/have done and how that aligns with your own views is a far better system.
And maybe, just maybe, it will pull some of these dinosaurs out of the stone age and force them to put up a website with their views (crazy I'm saying this in 2026, it should have been table stakes decades ago). Local races are incredibly difficult to find information on. Your best bet is one of those "voter guides" that a local newspaper might put out and even those aren't great. Using an LLM to scour the internet to make a better-informed decision seems like a huge win over what most people do today.
Ultimate tool for propaganda and manipulation tailored to each person's taste. Highly scalable and can focus on the things you (don't) care about when presenting "both sides".
The stakes are too high for it to not be weaponized, the only question is who will control it.
who do you think controls it? it's not a secret, they trade on the stock market
> Local races are incredibly difficult to find information on
Big time. At best in more rural areas, you will find a facebook page for the candidate. Sometimes, there is nothing they put out and just maybe one of those "we interviewed the candidate and this is what they said" type sites. Sometimes, nothing at all and it is a real shame. It would solve a lot of problems if we knew more about our local school administrations, mayors, sheriffs, etc.
What's stopping someone from paying to influence AI results?
How is that a different question than search results and which voter guides or topic editorials get highlighted first?
You have to disclose if a site is part of a PAC or candidate's committee due to FEC laws. This is not the case for LLM slop that often intermingles the two.
So yes, it's very very different.
Maybe these LLM companies should also disclose in every chat window on what political spending they have already done:
https://influence.citationneeded.news/
While this is true, there is a gaping hole in how foreign agents can amplify influencers on youtube and twitter. Probably even facebook, at this point.
An excellent reason not to use those sites, and certainly not for political decision making...
I guess FEC laws don't protect you from the pravda network. However, LLM might be not so good at ranking false information, if it is already massively embedded in their training data (LLM grooming).
Still it seems that this is currently not the case [0] (yet).
I guess the risk atm is rather some crazy billionaires embedding their believes in post training and system prompts. But honestly I rather fear tiktok in this regard.
[0] https://misinforeview.hks.harvard.edu/article/llms-grooming-...
The employees of the AI labs. They're a pretty vocal bunch, and would be very likely to quit, whistleblow, or both.
(xAI may be the exception here.)
Some have been vocal, a few have quit.
How has that changed how anyone interacts with AI?
Oh sure. The only whistleblower, Suchir Balaji. was murdered and the other employees just want to get rich (Brockman even is on record saying so).
A "vocal bunch"? They never depart from the party line.
1) People at large seem to prefer large stacks of cash to base morality
2) Employees will be tasked with writing a generic system that affects results, but only a handful of suits will know what it's actually going to be used for
Multiple high profile employees have quit... 99.99% have not. What has changed as a result?
We heard their concerns. My argument here is that any attempts within an AI lab to bias political answer would come to light.
Based on what? How could you possibly be sure?
Musk for example, has claimed in the open to push for fine-tuning to Grok that benefits him personally.
What’s stopping people from putting out biased voting guides that have nothing to do with AI?
While we can throw around hypotheticals and conspiracy theories about influencing AI results, we know for a fact that search engines and social media are heavily influenced, because there are giant industries around SEO and viral marketing -- plus the huge amount of ads and promoted content, which are literally paying to influence results.
Nothing. And it will happen big time once the AI companies become enshittified. Anthropic and OpenAI won't be able to produce the results that would justify a trillion dollar IPO without doing shady stuff.
> focusing on what they candidates say/have done and how that aligns with your own views is a far better system.
How often have you looked into two candidates at any level above city council and decided that the "red vs blue" assumptions differ from who you would normally vote for? Add in a few specifying sub-labels (MAGA, DSA) and that number probably drops to insignificant.
> How often have you looked into two candidates at any level above city council and decided that the "red vs blue" assumptions differ from who you would normally vote for?
Above city council (or a few levels, at least up to state Rep) are the people I have less issue picking between because more information is available. It's the lower races or things like county coroner that I have much less info about. Some of those roles still have (R)/(D) suffixes on the candidates but I don't feel like that means as much as state senate/reps with a letter after their name.
We have a local far-left newspaper I've gotten good info from by asking AI to review their endorsement to see what I should know before voting.
Turns out "my guy" was a domestic violence abuser, so the left-wing wacko got my vote.
> focusing on what they candidates say/have done
who's going to cross check it? and even if everyone did, who's going to ensure that everything everything is mentioned, things can fact check just fine with some strategic omissions.
Cross checks will be useless because enough GEO/SEO will be pumped out using generative AI that even what the candidate said will be distorted or "interpreted" in ways that fit the agenda of the organization doing the SEO. "If you give me six lines written by the hand of the most honest of men, I will find something in them which will hang him." -- Cardinal Richelieu.
The best part is that the AI companies get paid for both causing immense societal problems and combating the problems their tools are causing. It's perfectly Kafkaesque and the best part it is going unremarked right out in the open.
you can have multiple agents run reviews and gut checks. open source models, etc. there are ways to go back and forth to get to some relative truth. regardless of complexity of said task, i would trust it a lot more than social media or search engines which are entirely corrupted by ad spend budgets.
Literally no one is or will be doing this (agents, open source models etc). People who do are rounding errors population scale
Also are you saying LLMs are and will be immune to ads?
Homogenizing opinion though a small number of private companies with their own incentives is a pretty terrible idea, actually.
The current system isn't perfect by any means, but at least there's a wide variety of source and experience. Diversity can provide a buffer against viral problems, be it literal living viruses or information.
Funneling information through a handful of sources is akin to Sinclair Broadcast Group buying up rural news stations. Awful for democracy and diversity. Can put us a single virus away from destruction.
I don't think it's a good thing. The good thing would be to have a good education, and local groups with whom to discuss.
Blindly trusting a model that's easy to manipulate is not a good thing
Yea but that’s not happening. The question is how this compares to the status quo. It could easily be much worse, could easily be a fair bit better too imo. Most people don’t know a thing about who they vote for
No problem with two companies owned by sociopaths who largely have the same political ideas, and are deeply tied to the current administration, controlling the tool used by the mass to decide who to vote for? Said tool being trained on reddit, twitter and other trash tier platforms where minorities of radicalised people on both side of the spectrum "discuss" political topics 90% of people do not care about.
Yeah, we're cooked chat, as they say
not just two companies, by the way. remember grok and its views?
to me it's more alarming not just the cesspool that these are trained on, but subtle bias that can be added by people in charge or their friends... how naive to expect this won't happen
Grok is a good example of how bad the labs still are at any sort of intentional, subtle bias. I'm sure they'll get there eventually, but the best attempts today remain pretty hamfisted.
1) believe or not even if it's obvious to you it is enough to mislead plenty of people. you know how many people fall for obviously scammy ads? Yeah, good news, those people vote
2) it's an example of how biased models are. how bad or good they are now is irrelevant because it'll be different in a year or month. remember if you don't see bias it doesn't mean it's not there.
3) have you checked how much money is going into this, do you think they don't have resources to get extremely good at it. just follow the money.
>> Using an LLM to scour the internet to make a better-informed decision seems like a huge win over what most people do today.
A spin on that I used for my local elections last year: My prompt stated the issues that were important to me and why. I then copied in any election guide I could find from newspapers, blogs, newsletters, endorsements, etc.
Then I asked for 1) which best fits my priorities, and 2) risks for each (eg "says they will support homelessness programs, has no direct experience in an elected position")
> Chatting through all the races and the different candidates, he worked to figure out who best represented his values.
This sounds like it can be a really good thing. In an ideal world it’s a perfect use of LLMs. It helps make the ballot more approachable to people, and it can encourage more people to vote or help people vote less blindly.
At the same time, it also relies on these tools being unbiased adjudicators. And over time I think there’s a greater risk of this straying from the truth.
If you’re too lazy to think for yourself then it follows that you’d be too lazy to think about elections/voting.
https://en.wikipedia.org/wiki/Franchise_(short_story) So it begins...
More like the year of the Whopper imo.
And so it begins: the outsourcing of research, thinking, and reasoning to AI in lieu of using their own brains, an opaque AI in the hands of trillion dollar corporations with their own agendas and shareholders to please.
In the future, maybe we won't need election analyses to predict political victories, we can just ask all of the major brand of AIs who a voter should choose and then multiply by the market share of each AI, dohohoho.
> using their own brains
Nonsense. There's no "using your own brains" in voting in the midterms. You get no information about the vast majority of the candidates, you're just expected to vote as the party tells you to, and the rest of the information you get is from PR firms in proportion to the millions spent to market to you and distort all the information that you could possibly find.
Even if you work your hardest and get a reasonable overview, you've dedicated far more of your time to this than anyone could be expected to (a sign of the leisure and access you have as a privileged middle-class person), and the vote will be necessarily decided by the vast majority who didn't.
People using LLMs for the midterms will be the best informed voters we've had in a long time, even if they use Grok (which will desperately try to lie to them, and won't be the only one.) They should also use LLMs to help them with strategies to get to an actual voting booth, because the parties like to sabotage voting in places they feel they'll be weak.
You seem to think everybody (but you) is a dummy and will just do what the LLM says to do, but were totally comfortable for them just doing what the party says to do. I might use an LLM myself to figure out what's going on with judges, if I can find one with access to the proper databases.
In fact, if you want to do good for the world, get access to good databases with good search engines, hook deepseek up to them, and open it to the public. People need to find out how judges decide, and if there are any complaints against them, etc.
Is there a sci-fi novel yet, where everyone teaches their own AI how to correctly predict how they answer moral questions, and how they prioritize their principles? And then their AI actually represents them in Congress? Representative AI Democracy?
(Yes, this is a potential Torment Nexus.)
Solve a trolley problem to align self driving car with your morals.
Your question made me remember the short story 'Franchise', from Asimov.
Like the cookies from Black Mirror.
In general I think it's a good thing that people are trying to get more informed before voting, but I worry that if this becomes popular enough, there will be a lot of pressure on the labs to influence what the LLMs say about political questions.
No need. Just create (using an llm) a corpus of websites of "facts" about candidates, made up to your liking, that the LLMs will happily slup up and spit out.
Any guesses what Grok will be pushing people towards?
And Grok's competitors doing the opposite is justified?
Is there any evidence of those competitors pushing political biases? (Actual biases, not facts like Trump losing the 2020 election and staging an attempted coup)
It's interesting; when I clicked I thought the angle would be "it's dangerous that we let chatbots influence people" but the article seemed to care more about the sycophancy angle: some form of "maybe it doesn't influence them enough"
What's worse, a populatiom being uninformed, or homogeneously informed?
The worst is homogeneously misinformed.
[dead]
Depends on how they’re being informed I guess. Homogeneously informed can be okay to pretty bad. From mid-century American television and print media all the way down to North Korea and Nazi Germany.
While the knee-jerk reaction is to roll one's eyes, if it gets people that would otherwise not vote to be a bit educated and go out and make their voices heard, I am all for it. Yes, this has the potential for abuse and needs to be watched closely, but is this worse than the classic "I asked my coworker and..." type of non-involvement many already have? Not really and my hope is that people will only use it as a stepping stone to more research.
Are they becoming educated? Or are they just adopting the bias of the AI? (Which is the bias of its training data and/or trainers.)
Is it worse than "I asked my coworker"? Yes, because maybe six coworkers total are who the entire country is asking. That's too much influence for a program.
Maybe it's better than deciding based on TV ads.
The best way for these hosted LLMs to handle this is to follow the example of Tolkien's elves:
> “The choice is yours: to go or wait.”
> “And it is also said,” answered Frodo: “Go not to the Elves for counsel, for they will say both no and yes.”
> “Is it indeed?” laughed Gildor. “Elves seldom give unguarded advice, for advice is a dangerous gift, even from the wise to the wise, and all courses may run ill.”
So much worse than just voting the party slate like they're told, and operating from television commercials.
Everybody should use ChatGPT to help them decide how to vote in the midterms. They should ask for specific references to platforms and tweets, with links that they can follow.
Could the information you get be horribly distorted? Yes. But no more than the information you're receiving from every other source - and with LLMs you can drill down until it runs out of alignment. The TV doesn't take questions.
This can only be illegal. Law, and all human rights and responsibilitys are built on the idea of humans making indipendent Judgments, free from any coersion or missleading information,, and so to allow giving voting advice from a machine with known, programed in, biases and inhibitory blocks is impossible, and could very quickly lead to total AI bans in many juristictions as there may be no reasonable way to monitor and certify such devices.
IMO for an involved voter, this is actually a good thing. AI can both help you clearly define your priorities and policy preferences via an "interview", and can subsequently find candidates who's publicly stated positions, and past record, align most closely with your priorities.
As always, AI can be used to completely outsource critical thinking, and doing so for elections is extremely concerning. However I would hope that more people use it to make better, more informed decisions, than to blindly choose candidates
To be fair, Americans have previously asked ouija boards, psychics, and the Magic 8-Ball the same question in the past… at least they’re not asking Grok.
Yet.
> at least they’re not asking Grok.
To be objective, you should test across ChatGPT, Claude, Gemini, and Grok, to see how fair they each are.
They all pull from the same internet. Many candidates don't even have a website, so are ripe for search poisoning at scale.
You can be sure that at least grok will be carefully curated to lean a certain way.
and the others aren't?
Only one has achieved mechahitler as far as I know.
I don't believe the American economy ever hinged on the success of the 8-ball, nor the CEO of Hasbro (owners of the Ouija trademark) called before Congress to question whether the boards could build bioweapons.
[dead]
Just vote for your party. cough Republican cough
Hopefully as a lark...