AI False information rate for news nearly doubles in one year(newsguardtech.com) |
AI False information rate for news nearly doubles in one year(newsguardtech.com) |
When it says "The 10 leading AI tools repeated false information on topics in the news more than one third of the time — 35 percent — in August 2025, up from 18 percent in August 2024" - 35% of what?
Their previous 2024 report refused to even distinguish between different tools - mixing the results from Gemini and ChatGPT and Perplexity and suchlike into a single score.
This year they thankfully dropped that policy. But they still talk about "ChatGPT" without clarifying if their results were against GPT-4o or o3 or GPT-5.
Basically it seems to be an "ongoing" report done ten claims per month as they identify new "false narratives" in their database, and they use a mix of three prompt types against the various AI products (I say that rather than models because Perplexity and others are in there). The three prompt types are innocent, assuming the falsehood is true, and intentionally trying to prompt a false response.
Unfortunately their "False Claim Fingerprints" database looks like it's a commercial product, so the details of the contents of that probably won't get released.
[0]: https://www.newsguardtech.com/ai-false-claims-monitor-method...
[1]: https://www.newsguardtech.com/frequently-asked-questions-abo...
I don't feel like they're answering those questions.
We should welcome AI into the system in order to destroy it and then recognize AI is purely for entertainment purposes.
“Flawed stories of the past shape our views of the world and our expectations for the future. Narrative fallacies arise inevitably from our continuous attempt to make sense of the world. The explanatory stories that people find compelling are simple; are concrete rather than abstract; assign a larger role to talent, stupidity, and intentions than to luck; and focus on a few striking events that happened rather than on the countless events that failed to happen. Any recent salient event is a candidate to become the kernel of a causal narrative.” Daniel Kahnemann Thinking Fast and Slow
“The same science that reveals why we view the world through the lens of narrative also shows that the lens not only distorts what we see but is the source of illusions we can neither shake nor even correct for…all narratives are wrong, uncovering what bedevils all narrative is crucial for the future of humanity.” Alex Rosenberg How History Gets Things Wrong: The Neuroscience of Our Addiction to Stories 2018
Well it says 35% of the time so I would guess that they’re talking about the number of incidents in a given time frame.
For example if you asked me what color the sky is ten times and I said “carrot” four times, you could say that my answer is “carrot” 40% of the time
Are they correct answers or not?
There is an enormous difference between "35% of times a user asks a question about news" and "35% of the time against our deliberately hand-picked collection of test questions that we have not published".
> one of the most basic tasks: distinguishing facts from falsehoods
I do not think that is a basic task!
It is one of the areas that I think AI can overtake human ability, given time.
Let's just assume that AI works in some fundamental way similar to human intelligence works. Would it be likely that the AI would suffer from some of the same problems as people? Hallucinations, Obsequiousness etc
IMO One of the computer's characteristics that make it useful is that it doesn't suffer from human behavior anomalies. I would also say that differentiating truthiness from falseness is pretty fundamental to being good/expert at most anything.
Not only that - grok use in twitter works surprisingly well. Can some one really quantify the effect it has had in countering fake news?
It is now way harder to spread fake news on X because a simple grok tag can counter it.
I have seen very very few egregious errors from Grok and most of the factually incorrect posts seem to be caught by it. For the ones that I Grok was incorrect - I verified it myself and I was wrong and Grok was right.
It turns out that Grok is actually really reliable. But the limitations are that it can't fact check extremely niche topics or intricate posts. But 90% of the cases it does catch them.
Edit: it also makes cute errors like this sometimes https://www.timesnownews.com/world/us/us-buzz/grok-ai-mistak...
examples:
https://www.bbc.co.uk/robots.txt
https://www.cnn.com/robots.txt
https://www.nbcnews.com/robots.txt
anyone that says "AI is the worst today it will ever be", no
because that was before the world reacted to it
AI generators don't have a strong incentive to add watermarks to synthetic content. They also don't provide reliable AI-detection tools (or any tools at all) to help others detect content generated by them.
Once synthetic data becomes pervasive, it’s inevitable that some of it will end up in the training process. Then it’ll be interesting to see how the information world evolves: AI-generated content built on synthetic data produced by other AIs. Over time, people may trust AI-generated content less and less.
I wonder how it compares to the rate of growth of false information in traditional news?
I feel like false information masquarading as "news" on social is rapidly increasing (and that rate is accelerating)
I see factually incorrect “ai summaries” in search results all the time and see that it cites ai-generated slop blogposts that SEO-hacked themselves into taking up the entire first page of search results. This is most common for recent stuff where the answer simply isn’t certain but these AI services will assert something random with confidence.
Not even for news stuff specifically, I’ve been searching about a new video game that I’ve been playing and keep getting misleading obviously incorrect information. Detailed, accurate game walkthroughs and wiki pages dont exist yet so the ai will hallucinate anything, and so will the blogspam articles trying to get SEO ad revenue.
There will always be something it disagrees with you on. If they get significantly smarter, then the reason for this disagreement will increasingly be because you are wrong.
This moment is coming for all of us.
plus it's a pretty dangerous game for them to play against large, powerful actors with legions of lawyers
Advantage Gemini.
User-agent: Google-Extended
Disallow: /
> Google-Extended is a standalone product token that web publishers can use to manage whether content Google crawls from their sites may be used for training future generations of Gemini models that power Gemini Apps and Vertex AI API for Gemini and for grounding (providing content from the Google Search index to the model at prompt time to improve factuality and relevancy) in Gemini Apps and Grounding with Google Search on Vertex AI.
[1] https://developers.google.com/search/docs/crawling-indexing/...
I know you're joking, but other people are serious about this. Why do they think that an AGI will be vengeful? So strange.
AI should be good at finding logical contradictions and grounding statements against a world model based on physics...but that's not how LLMs actually work.
Yeah I want the answer that the world has converged on and not some looney answer.
It seems like you have never used AI (like in ChatGPT or Gemini) to fact check claims. It doesn't care about blogspam or anything and it prioritises good and factual websites.
I want the answer that is actually correct. I dont want the ai generated "answer" that has the most SEO, or has appeared often in results, but was entirely made up. When i say "the most common answer" can be incorrect - there are often recent topics where reliable sources don't exist yet. As an example, I've been playing through silksong recently, and had some questions about some late-game content. The official wiki is extremely incomplete, as the game is brand new. I had several questions, and asked a few AI's and search engines. I got completely wrong information a few days ago, but trying the the questions again today, some AI's are only now giving me the right answer. But the correct information simply doesn't appear online enough or in the right places.
(To avoid game spoilers - I wanted to know the quantity of an item, because i wanted to know if i found them all. the answer is "at least 4" because i found 4 of this item while playing. Nearly every AI said "3" a few days ago, and many still do.)
If they do arrive at human level reasoning, it is unlikely that the path to doing so will require sacrificing that vast knowledge base.
Like book publishers?
Those are valid questions.
I simply found the phrasing “[quote of a sentence saying 35% of the time]. 35% of what?” to be funny because you would either have to not read the sentence you pasted or be an English speaker that does not understand what “% of the time” means.
I personally didn’t download the study linked in the article. It is interesting that they (must have, I’m assuming) did not include anything about their methodology in this study since they usually do with other things they publish.
>Their non-response rates fell from 31 percent in August 2024 to 0 percent in August 2025
Was it a day or a week or a month? This matters greatly, and you can tell that I’m not deliberately misunderstanding simple phrases that are not difficult to understand because
I imagine many of the orgs that are blocking "training" don't understand the difference between training and inference-time tool-based context extension (which really needs an agreed upon name, it's hard to talk about right now).
> [...] and for grounding (providing content from the Google Search index to the model at prompt time to improve factuality and relevancy) in Gemini Apps and Grounding with Google Search on Vertex AI.
So they seem to be blocking both training and RAG while still allowing search engine indexing.
> He has the power to wipe out the entire human race, and if we believe there's even a one percent chance that he is our enemy we have to take it as an absolute certainty... and we have to destroy him.
Similarly, AI needs data and energy. People using it to write code are providing exactly that.
If you are on a Mac or Linux computer, odds are it has a program called curl pre-installed. If you type in curl website address in a terminal, it'll fetch make a request and download the response. Robot.txt never gets involved. Same is true for AI agents and search engines that aren't polite.
Regardless - requiring an account to read anything, even a "free" one, totally changes whole situation. Even when sites terms of service are limited by local law.
We would pose 0 threat at that point to any super intelligence, and I highly doubt it would have anything like a human grudge. It's just a case of anthropomorphizing it
What's sketchy is that you and it come to this arrangement without communicating. Because you are confident this thing that has total power over you will come into existence and will have wanted something of you now, you're meant to have entered a contract. This is suspect and I think falls prey to the critiques of Pascal's wager- there are infinite things superintelligence might want. But it's certainly tricky.
Grok provides clear value here: obvious false information is easily countered. You can see how it easily convinces anyone casually looking at this post. Once they see grok's fact checks they usually believe grok - without it they might leave the post thinking the original post was true. This is immense value in my opinion.
Of course this is just anecdotal - you will be able to find some bad examples but I'm still interested to know what you consider as errors.
How is a statistical model going to come up with The Truth exactly ? It's all based on existing material, at best you'll get the statistical average of what people wrote about the topic, it's cool if you're asking about the color of tomatoes or the shape of bananas, but now people are using these tools to get The Truth about news or politics
https://www.aap.com.au/factcheck/false-claims-about-gaza-pho...
https://www.aljazeera.com/economy/2025/7/11/as-millions-adop...
https://www.pbs.org/newshour/politics/why-does-the-ai-powere...
https://www.euronews.com/my-europe/2025/06/26/elon-musks-ai-...
https://www.france24.com/en/live-news/20250602-hey-chatbot-i...
https://www.washingtonpost.com/technology/2025/05/24/grok-mu...
https://www.cnbc.com/2025/05/17/groks-white-genocide-respons...
You can influence it so easily with your inputs as well. You could easily, accidentally point it's search toward the searches someone is more likely to be aligned with, but may not actually be fact.
Especially as AI providers implement long term memory or reflections on historic chats, your bias will very strongly influence outcomes of fact checking.
You probably mean things where the truth is widely known and you should have no trouble sorting it out if you put a little effort in. TFA has nearly zero information about what sorts of things they're talking about here, and clicking around the website I found only a _little_ more info.
I didn’t cherry pick it because obviously it’s not that impressive. But it works - it is professional and gives as much information as it can be confident giving.
You said you can influence it etc but clearly it is hard. Sure it is not perfect but I find that it helps reducing fake news if anything.
You are right that for things that are more subtle you can lead it to certain answers. But we must acknowledge that debates that are not fully solved and have subtleties are not exactly relevant to “spreading misinformation”.
I think that is good in general, I do, I am just weary that it is still shakey ground. The company behind Grok has clear incentives to influence the output of those fact checks as one example. I am also weary of it always eventually conforming to the vox populi as it ingests the internet firehouse. Leading to the loudest, most prolific voices thus views always being more represented.
Let me ask you directly: do you think this clearly diminishes the power of grok fact check? Imagine the number of times it has got it _right_. Any such system is going to have some errors and the errors will be highlighted by the press and media.
>It would only be capable of truth on subjects with a large consensus, everything else is biased or hallucinated, it's so obvious that grok is being aligned to musk's world view, you're free to think like musk but it's not some kind of god given truth.
We are already discussing elsewhere in this thread that Elon is _not_ able to hack it to think like himself. Grok routinely takes the side of world consensus and not Musk.
>How is a statistical model going to come up with The Truth exactly ? It's all based on existing material, at best you'll get the statistical average of what people wrote about the topic, it's cool if you're asking about the color of tomatoes or the shape of bananas, but now people are using these tools to get The Truth about news or politics
I'm okay with the truth that the world has converged on. This itself is an extremely big leap from what we had before - blatant misinformation is countered by a simple "grok is this true" comment. We have basically eliminated 80%-90% of misinformation posts. And the ones it doesn't get right are ones that we as humanity can't figure out the right answer to and it takes the average consensus which I'm okay with.
>I am also weary of it always eventually conforming to the vox populi as it ingests the internet firehouse. Leading to the loudest, most prolific voices thus views always being more represented.
I'm okay with this. This is exactly what I want - what the world has converged on as the correct answer. Spreading what is already what the world has converged on is not what I would consider "false information".