People are asking ChatGPT to help them decide how to vote in the midterms
npr.org
npr.org
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.
The stakes are too high for it to not be weaponized, the only question is who will control it.
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:
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-...
(xAI may be the exception here.)
How has that changed how anyone interacts with AI?
A "vocal bunch"? They never depart from the party line.
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
Musk for example, has claimed in the open to push for fine-tuning to Grok that benefits him personally.
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.
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.
Also are you saying LLMs are and will be immune to ads?
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.
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.
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.
Turns out "my guy" was a domestic violence abuser, so the left-wing wacko got my vote.
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.
Blindly trusting a model that's easy to manipulate is not a good thing
Yeah, we're cooked chat, as they say
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
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.
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")
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.
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.
(Yes, this is a potential Torment Nexus.)
First 100 HN readers can get it free (DRM-free EPUB/PDF): https://solyin.gumroad.com/l/enof/HN
Seconding Franchise, though. Asimov got to the "one voter stands in for everyone" version in 1955.
> “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.”
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.
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.
Yet.
To be objective, you should test across ChatGPT, Claude, Gemini, and Grok, to see how fair they each are.
You don't know that.
Give a concrete example of a candidate where between Grok and one of the other models the lean is tilting the scales.
What does that mean?
More importantly, that report is more than a year ago. I would be surprised if the chatbot still calls itself that.
This seems like an irrelevant distraction.
It doesn’t call itself anything, it doesn’t have any conception of identity, nor fur that matter the semantic meaning of either part of the portmanteau it found in its training corpus… what it does have is a bias towards the bias of its trainers, in this case paid for by someone who would find it really (if pathetically) funny if their personal LLM was automatically fascist.
It’s also got a marked bias towards generative CSAM, according to investigations and filings in multiple jurisdictions… which also implies it’s best for voters not to ask it questions.
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