Perplexity Will Show Live US Election Results Despite AI Accuracy Warnings (arstechnica.com) 18
An anonymous reader quotes a report from Ars Technica: On Friday, Perplexity launched an election information hub that relies on data from The Associated Press and Democracy Works to provide live updates and information about the 2024 US general election, which takes place on Tuesday, November 5. "Starting Tuesday, we'll be offering live updates on elections using data from The Associated Press so you can stay informed on presidential, senate, and house races at both a state and national level," Perplexity wrote in a blog post. The site will pull data from special data sources (called APIs) hosted by the two organizations. As of Monday, Perplexity's hub currently provides interactive information on voting requirements, poll times, and summaries about ballot measures, candidates, policy positions, and endorsements. Users can ask questions about the information similar to using a chatbot like ChatGPT.
Perplexity's embrace of providing election information is an exception in the AI field. Wary about accidentally providing misinformation, competitor AI assistants from OpenAI, Google, and Anthropic currently direct users elsewhere or decline to answer election questions. OpenAI's ChatGPT Search directs election result queries to The Associated Press and Reuters. Perplexity describes its new elections hub as "an entry point for understanding key issues." But like other AI models, Perplexity can produce confabulations (plausible incorrect information) when generating responses. That could present an accuracy problem because the site's Voter Guide service uses AI language models to summarize and interpret information pulled from the web. Here's what Ars Technica advises: "Take what you see on Perplexity's site with a huge grain of salt -- do not rely on it without verifying the information with a trustworthy external source."
Perplexity's embrace of providing election information is an exception in the AI field. Wary about accidentally providing misinformation, competitor AI assistants from OpenAI, Google, and Anthropic currently direct users elsewhere or decline to answer election questions. OpenAI's ChatGPT Search directs election result queries to The Associated Press and Reuters. Perplexity describes its new elections hub as "an entry point for understanding key issues." But like other AI models, Perplexity can produce confabulations (plausible incorrect information) when generating responses. That could present an accuracy problem because the site's Voter Guide service uses AI language models to summarize and interpret information pulled from the web. Here's what Ars Technica advises: "Take what you see on Perplexity's site with a huge grain of salt -- do not rely on it without verifying the information with a trustworthy external source."
Why not? (Score:3)
From their perspective, I mean. If they get things wrong, nobody will remember. It's a new technology after all, and you're asking it to predict the future, so you chalk it up to the impossibility of the task with present levels of technology. But if they get it *right*, they'll have the media and the nattering classes eating out of their hand for years to come.
This is much like polling. Polling is inherently imprecise. Forget the "margin of error", that's the sampling error if they do everything else perfectly, which they never do. If a poll has a "2% margin of error" it's doing well to get things within 4%. The wonder is how accurately polls actually get things. So within the relatively wide pragmatic confidence interval of polling results, and with diversity of models and methods used by pollsters, *somebody* always gets it right by sheer luck. And in particularly if that person has a contrarian position, they will for the next decade or so be treated as the genius who nailed the election of year X.
Re:Why not? (Score:4, Interesting)
So within the relatively wide pragmatic confidence interval of polling results, and with diversity of models and methods used by pollsters, *somebody* always gets it right by sheer luck.
Which is what makes the polls this year so suspicious. Nate Silver called it "herding" - all the polls are showing a race tied within the margin of error. Almost every poll reports a confidence interval of 95%, so that means you'd expect 5% of polls - if they were truly random samples - to be outside that range.
But they aren't. Because every pollster is absolutely terrified that they'll be declared "wrong" if the election doesn't go the way they polled. And a pollster who is "wrong" won't get paid to run polls in future elections. So they all do some "modelling" to "correct" whatever raw data they got. (It's also to deal with the fact that getting a random sample is, literally, impossible. There are just too many people who simply won't answer a poll, which makes any sample that doesn't include non-responders non-random, so pollsters do a whole bunch of things to try and deal account for that. If they did include non-responses you'd get polls that show something like 5% for each candidate and 90% "didn't answer.")
You can't be accused of being wrong if everyone else in the herd was too.
Re: (Score:3)
The way this is supposed to work is that you develop a methodology to weight stuff and you follow it where it leads, and only *then* do you update your models for things like voter turnout and response, after you've taken your lumps. Altering your method to fix results you don't want to see should technically be regarded as professional malpractice.
Re: (Score:2)
Computers [historyofinformation.com] vs humans [enemyinmirror.com]. It could go either way.
Re: (Score:2)
If they get things wrong, nobody will remember.
Which is why it is the duty of the legitimate media to ring this bell loudly and often.
And with .1% of the vote in... (Score:4, Funny)
Perplexity has called the race for US President! Congratulations, President Pat Buchanan!
Who? (Score:2)
Re:Who? (Score:5, Interesting)
They say exactly what you're asking for as the first paragraph of the summary.
Did you want them to put it in the title so you could ignore it there too?
You'd just complain that the title was too long to read.
Why didn't they pre-plant that information directly into my brain before showing me the headline? Now I'll have to highlight the word and click search.
Re: (Score:1)
I'm sure they would never ever call it too early for their preferred candidate in an ultra close election !!!
You know that's not how we actually decide who won right?
Well unless you're WayMoreGullibleThanMost. Then anything you see on Fox is the gospel truth.
Can't wait for RFK Jr. to win (Score:2)
Idiots. (Score:2)
Might as well throw another gasoline soaked log on the blazing fire of things to point to when denying the election.
Doesn't matter. The US has given up on democracy. The perception persists... but if the vote is only valid if your candidate wins, it's a fraud. If the view was only held by a few folks maybe you could claim the abandonment of democracy is an overstatement... but it's not.
Honestly ignore any results (Score:3)
In early voting voter turnout among youth has been pretty insane. Georgia is up 30% equating to around 80,000 additional votes, Pennsylvania is up 100% and Michigan is up 220%.
If that is somehow sustained through election day then you are going to see a blowout for the Democrats, even more so if the current 55 / 45% rates favoring women hold true.
On the other hand if that evens out during tomorrow's voting then we won't know until probably mid-morning on Wednesday how the election actually went because it'll be too close to call. The only reason the youth and woman votes would allow the election to be called is they've been pretty consistently 75% for Harris so extremely high voter turnout in those groups carrying through to election day wood allow for the race to be called but anything short of that and it's going to be down to the wire and a few thousand votes either way.
If you haven't already voted basically every place in the country is up for grabs. Even traditionally red states have the potential to flip blue. So if you didn't vote early you probably want to get your ass to the polls tomorrow
Can you explain the Singapore thing? (Score:3)
Big data knows (Score:2)
Poll-takers have to account for the bias in the answers they get: Over the last few elections, that's been very difficult. In the last week, the poll-takers have admitted the factors of bias have changed and their corrections are probably over-corrections.
There's been a lot of talk about the woman-vote, youth-vote and black-men-vote being vastly mis-represented: That is, the true swing is much greater than poll-takers are reporting. If true, it won't be a close election.
Big data has been calculating t