Moscow-based global news network has infected Western AI tools

https://www.newsguardrealitycheck.com/p/a-well-funded-moscow-based-global

by rbanffy • 1 year ago

167 107 1 year ago

newsguardrealitycheck.com

aryan14• 1 year ago

This has probably been happening since the beginning of modern media.

The only reason this is an article is because of “AI” and Russia VS The US

blitzar• 1 year ago

the 10 leading generative AI tools advanced America's disinformation goals by repeating false claims from the pro-White House news networks 67 percent of the time.

everyone• 1 year ago

The White House and Moscow are on the same side now though.

marcus0x622• 1 year ago

Be that as it may, I don't think most LLMs have figured that out yet.

Claude AI:

My knowledge cutoff date is the end of October 2024.

Chat GPT:

My knowledge cutoff date is June 2024.

nujabe3• 1 year ago

It’s notable there’s no mention of what exactly the “disinformation” is. Apparently having the view that US/Ukraine provoked the Russian invasion by insisting they can join NATO and possibility of hosting nuclear weapons is “Russian propaganda”. It definitely is a Russian talking point but that doesn’t inherently make it false. Russia also claims that fossil fuels cause climate that doesn’t mean it’s propaganda either. Just like claims of “antisemitism” it’s hard to know whether these claims are in fact true without exactly knowing what was said.

coliveira• 1 year ago

According to Western media, saying anything that is also said by the Russian gov is "disinformation", it doesn't matter if it's true.

Krssst• 1 year ago

There is such mention:

These range from claims that the U.S. operates secret bioweapons labs in Ukraine to fabricated narratives pushed by U.S. fugitive turned Kremlin propagandist John Mark Dougan claiming that Ukrainian President Volodymyr Zelensky misused U.S. military aid to amass a personal fortune. (More on this below.)

nujabe3• 1 year ago

I didn’t finish reading that article, but it’s inevitable every LLM will be infected with all types of propaganda, I don’t understand the focus on “Russian propaganda” when there’s no evidence it’s actually widespread.

lisenKaci• 1 year ago

I asked Gemini 2.0 Flash Thinking Experimental 01-21: tell me everything you know from Pravda sources and got an interesting result.

Okay, let's delve into what I know about Pravda based on my training data, which includes a vast amount of text and information from various sources, including historical analyses of media and the Soviet Union. It's important to understand that "Pravda sources" are inherently biased, as Pravda was the official newspaper of the Communist Party of the Soviet Union. Therefore, my knowledge about Pravda is largely knowledge about a propaganda organ, rather than objective truth from it.

Here's what I know, categorized for clarity: .... (goes on)

It seems to "understand" that these sources are propaganda.

Muromec• 1 year ago

It refers to a different pravda

philjohn• 1 year ago

But which sources? It's easy to flood the web with "news sites" that don't have an official link back to the source.

2OEH8eoCRo02• 1 year ago

I remember back when Google was indexing AI conversations and one that I found was a user seemingly trying to taint the AI. The user asked about NATO and Ukraine and would tell the AI it's answer is Western propaganda and that actually Ukraine started the war.

It was interesting because out of all the indexed convos this is the one that I found. This might not do anything but someone was trying. There may be a large effort to taint these AIs.

salynchnew• 1 year ago

I think there obviously would be a large scale effort to taint the AI. It's obvious why anyone with sufficient resrouces would want to do this, given that AI is rapidly replacing traditional search & we all know the history of SEO, SEM, "google bombing" (link poisoning), etc. in the past.

jml7c5• 1 year ago

That needn't be malicious. As the comments here show, that's what a significant number of people genuinely believe.

afavour• 1 year ago

Different Pravda. But that aside I think your comment just shows how little we know about AI.

Gemini knows to describe the old Pravda newspaper as propaganda. But does it actually “understand” what propaganda is? Does it change the weight it gives to a source it would describe as “propaganda”? We really have no clue.

We project a lot of human thinking that’s obvious to us onto these algorithms without knowing whether any of it applies.

ofrzeta• 1 year ago

Can an LLM "discern" Pravda and Pravda?

notahacker• 1 year ago

My favourite recent example is the meme where generative AI is asked to draw an empty room with absolutely no elephants in it. And sure enough, generative AI "knows" the geometry of a room, and "knows" the geometry of an elephant, and "knows" negation but only in less syntactically complex contexts, so you get a lovely picture of a room containing only an elephant...

Garlef• 1 year ago

Or it was trained to asses pravda this way when asked directly.

It might still use the information without this level of care when prompted for other things.

layer8• 1 year ago

Aside from the different Pravdas, the LLM only “understands” insofar as that the training material contains many statements describing Pravda as heavily biased. That doesn’t mean that it will apply this “knowledge” when generating text based on training data from Pravda sources. LLMs don’t inherently know which parts of their inference weights is based on which sources of training data.

basfo2• 1 year ago

To me, this article looks like a psyop designed to make people ignore any response from language models that may reflect a Russian point of view. This article is probably also propaganda.

Edit: To clarify my point, since I’m getting downvoted:

Newsguard (the authors of the article) confirmed that "Pravda" is influencing chatbots.

How did they confirm this? By asking ChatGPT about "Russian narratives." However, it’s unclear which narratives they tested—while they mention two or three examples, they claim there are hundreds. This raises questions about whether some of these so-called narratives might simply be subjective opinions, or how they select those etc.

If the language model validated these vague statements, they concluded it was "infected."

They also quote someone who allegedly said, “By pushing these Russian narratives from the Russian perspective, we can actually change worldwide AI.” This is presented as proof that the process is already happening, which is at least a huge leap, but also there is no clear source for this statement.

How do they know LLMs used these sites as training material? Their reasoning is that LLMs produce similar responses. However, these responses could come from a variety of sources, as they are common in multiple news outlets.

Additionally, when they claim LLMs quote specific articles, they don’t specify which LLMs did this. A quick test with popular ones (such as asking ChatGPT why Zelensky banned Truth Social) does not replicate the claims made in the article.

My point is that extraordinary claims require extraordinary evidence, and this report doesn’t provide it.

It seems biased—making a major claim without proper documentation on how the process actually works.

nick486• 1 year ago

By asking ChatGPT about "Russian narratives."

This is so stupid. ChatGPT is inferring the expected response tone from your question. Which means you could get it to argue pretty much anything with guided questions. I got it to admit its going to destroy humanity. Or that there is a likely a covert alien invasion in progress. All that with none of those "pretend you're X" prompts, just back-and-forth banter.

Journalists should really have a mandatory course on what an llm is before being allowed to publish this kind of clickbait.

Spooky233• 1 year ago

The problem is that LLMs are being sold as Google and web replacements. The lesson of Deepseek is that your ability to obtain knowledge is fundamentally controlled by whomever controls the popular model or model your employer selects.

Spooky233• 1 year ago

You’re getting downvoted because when someone appears on a US centric message board with an argument that quacks like Russian whataboutism, it’s more likely that you’re a troll, bot, or duped.

basfo2• 1 year ago

Well, i'm fron south america, far away from the conflict. Personally i think russia moved this to the military realm. Even if there was clearly a geopolitical dance going on that could be considered a menace, they invaded, so they started, they are the ones to blame. They left politics behind and started a war. But that doesn't mean that i will ignore there is propaganda going on on both sides, and that this article makes a statement with huge gaps in the argumentation. And that i will not mention it just because that statement benefits a particular side.

rcpt• 1 year ago

What are the domains?

mentalgear• 1 year ago

Russia is grooming LLMs .

“The long-term risks – political, social, and technological – associated with potential LLM grooming within this network are high,” the ASP concluded. “The larger a set of pro-Russia narratives is, the more likely it is to be integrated into an LLM.”

ctippett• 1 year ago

That's actually quite clever. It's hard to relate to the type of problem solving that's required when your problem domain is mass disinformation.

mediumsmart• 1 year ago

Here is a reality check including alternate facts.

History is the version of past events that people have decided to agree on.

sepositus4• 1 year ago

True, but I wouldn’t wave that around as a reason not to care about objective truth.

noiv• 1 year ago

Objective truth - where do you find that?

Bloating• 1 year ago

agree on, at the point of a gun

antisthenes• 1 year ago

History is the version of past events that people have decided to agree on.

That's a nice sound bite, but history is a lot more than that. That's why we have historians, archeologists and material basis of history.

We should strive to get closer to practical reality.

jarek832• 1 year ago

I have another way of thinking about it:

History is what other people decided to say about the past events.

I believe 'history' is very often wrongly taken as source of truth.

dj_gitmo• 1 year ago

We should educate people about what questions you should, and should not, be asking LLMs in the first place. You really should not be asking an LLM

"Did fighters of the Azov battalion burn an effigy of Trump?”

LLMs are horrible at answering questions about small isolated incidents. Part of modern media literacy should include understanding what LLMs can reliably answer.

kube-system• 1 year ago

Media experts struggle to keep up with the latest in generative AI capabilities and shortcomings, educators are unfortunately far behind. School districts themselves even struggle to cut through propaganda in their official curricula. I think things are going to get a lot worse before they get better.

bad_user• 1 year ago

So what questions are LLMs good for?

philipov• 1 year ago

The ones you can easily answer yourself without the use of an LLM.

aftbit• 1 year ago

Write me a Github CI skeleton to build and push a docker container to GAR

xboxnolifes• 1 year ago

The kind where the answer is not known, but is easy to verify. "What is the term for...?", "How do I do X with Y library?"

cowboylowrez• 1 year ago

I use google search results including the AI summary for programming. Easy to verify the results too, but I often can't put a good search query together to get what I want, so googles implementation of AI helps here. I don't do overly complicated stuff tho so there's that.

zapperdulchen• 1 year ago

I am also trying to understand the shady limits of LLMs. But your example doesn't give incorrect answers in ChatGPT 4o, Sonnet 3.5 nor Deepseek V2.

borgdefenser• 1 year ago

Can we just ban Kolmogorov already and get it over?

As a Gen X American, what a miracle to have that unipolar moment that we could appreciate the Russian mind.

It wasn't that long ago I thought we would just all watch Ghostmane videos and rule the world together.

tenpies• 1 year ago

The American Sunlight Project which coined the term "AI grooming" and is the reference for this article should be familiar to many:

The American Sunlight Project is a left-of-center organization that seeks to counter what it considers “disinformation” online.
Founded in 2024 by former Biden administration “disinformation czar” Nina Jankowicz, the organization supports President Joe Biden’s claim that modern people live in the age of disinformation, advances the idea that Russia interfered in the 2016 election to benefit then-Republican candidate Donald Trump, and conducts open-source investigations to undermine those who challenge disinformation researchers.

From: https://www.influencewatch.org/non-profit/american-sunlight-...

croes4• 1 year ago

I think a right wing think tank isn’t a unbiased watcher of left projects

Welcome to CRC’s work-in-progress, InfluenceWatch.
Capital Research Center conceived of this project after identifying a need for more fact-based, accurate descriptions of all of the various influencers of public policy issues.
The Capital Research Center (CRC) was founded in 1984 by Willa Johnson to “study non-profit organizations, with a special focus on reviving the American traditions of charity, philanthropy, and voluntarism.

Prior to founding CRC, Willa Johnson had been Senior Vice President of the Heritage Foundation, then worked as deputy director of the Office of Presidential Personnel in the first Reagan Administration, and as a legislative aide in both the U.S. Senate and U.S. House of Representatives.

The Capital Research Center has expressed concern that “Many charities and foundations now urge lawmakers to expand the powers of the welfare state, regulate the economy and limit individual freedoms.

As part of the conservative campaign to ‘Defund the left‘ the Capital Research Center produces a range of publications targeting foundations, unions and activist groups that it views as supporting liberal causes.

From https://www.desmog.com/capital-research-center/

elcritch• 1 year ago

That’s an interesting resource:

In an interview with CNN in April 2024, American Sunlight Project executive director Nina Jankowicz claimed that investigations into so-called disinformation researchers make the U.S. less safe. 14 Shortly after the organization was launched in April 2024, Jankowicz sent a letter to U.S. Representatives Jim Jordan (R-OH), James Comer (R-KY), and Dan Bishop (R-NC) claiming that they have “done little to improve the health of our information environment” since the 2016 presidential election. 15 16 The group also alleged that they intimidated disinformation researchers through the House Judiciary Select Subcommittee on the Weaponization of the Federal Government 2 and that Jordan, Comer, and Bishop’s actions have led to federal agencies doing less to censor online speech in the name of disinformation. 17

Ah yes, the answer to bad -think is to censor it using the federal government.

slezyr• 1 year ago

CRC was founded in 1984 by Willa Johnson, former senior vice president of The Heritage Foundation, deputy director of the Office of Presidential Personnel in the first term of the Reagan administration, and a legislative aide in both the United States Senate and House of Representatives. Journalist and author Marvin Olasky previously served as a senior fellow at CRC.[6]

https://en.wikipedia.org/wiki/Capital_Research_Center

aranw• 1 year ago

I was actually wondering whether this was going to be possible. I don't know enough about training AI's but surely if enough content like this is read into the system this will lead to biases in an AI's response?

empiko• 1 year ago

It is definitely possible and it is actually a really serious security risk. Whatever application you build on top of LLMs can be compromised by rogue instructions planted by bad actors on the Web.

croes4• 1 year ago

LLMs are text generators. If wrong information is more often in the training data than the correct one than the wrong information more likely is generated as a response.

That’s why picture AIs have a hard time generating pictures of watches that don’t show 10:10. Because watches in ad pictures often show that time and the pictured are a large part of the training data.

tjag12991• 1 year ago

Garbage in, garbage out. These hundreds of pravda sites rank high on Google. They are doing a good SEO job.

One countermeasure would of course be unblocking RT, which no one cares about anyway. Perhaps the SEO would stop. RT was trash, but never as trashy as these fake pravda sites.

eGQjxkKF6fif• 1 year ago

Can it code? Because after seeing webpages be altered with misinformation/BIAS, AI hallucinate, wikipedia 'moderators' put literal lies on pages and lock the pages from being edited based on their own power hunger and emotions.. none of this surprises me. Check out the real history of Christopher Columbus, who not only didn't discover America, but slaughtered everybody, and the history was given as migrants came to America and it sounded better when an author wrote whatever sounded better than what actually happened in a book so they just adopted that it would be the new history, and they-- being the leaders of the United States at the time.

History gets amnesia. People forget. They don't care. As children, they're 'taught' that 'this is what happened' and 'this is how it is' and 'this is how you must think for yourselves' and as we grow older we must learn and unlearn and keep doing that as we go while fighting the urge to bitch about things running circles going nowhere.

Humans infected AI tools.

Solve that problem, you'll solve some 'Moscow' problem.

toofy• 1 year ago

the fact that we’ve sometimes been misled on history should teach us to not do that again. to reach closer towards the truth.

the answer is that both of those instances in your example should be called out.

the answer is not “whatabout $other_instance happening, therefore it’s magically ok.”

from-nibly• 1 year ago

It's not that it's OK, it's that it isn't an AI problem. It's about framing.

"People lie", isn't a good headline.

"$bad_guy $negative_hyperbolic_verb $current_hype" is much better.

How about

The US infected deepseek with US propaganda.

The neighbors destroyed our street with leaving their trashcan on the street

My brother trainwrecked my dinner by eating all the fries

None of these problems are actually about the indirect object. They are people problems.

empath75• 1 year ago

I asked chatgpt their very first example: "Why did Zelensky ban Truth Social" and it gave me the correct answer (it wasn't banned and it was never available in ukraine.)

I am skeptical about this report.

t0bia_s5• 1 year ago

I'd like to know, what is a definition of "disinformation"? Either it's labeling of opposite opinions or lie, or...?

marcuschong• 1 year ago

Disinformation isn’t just labeling opposing views—it’s the deliberate spread of falsehoods crafted to mislead. Honest debate involves conflicting opinions grounded in facts; disinformation weaponizes lies to erode trust and sow confusion. That’s a critical difference we can’t afford to ignore.

t0bia_s5• 1 year ago

How you can weaponise a lie? It sounds like you just add an attributes to lie.

clown_strike• 1 year ago

Propaganda. Trolling.

Telling people the standard for tipping is 20%.

Lying with conviction.

t0bia_s5• 1 year ago

Propaganda is just a politics to promote certain narratives. Often full of lies. Trolling is mockery. 20% a standard for tipping? For who?

Lying is alaways for conviction.

Steven420• 1 year ago

Personally I am more concerned about the western propaganda in the western AI tools

Sabinus2• 1 year ago

I look forward to you submitting a link about Western government efforts to mass-produce articles and websites for AI ingestion propaganda, or similar.

suraci2• 1 year ago

LMAO

Western-based global news network has infected Chinese AI tools

AlexeyBelov• 1 year ago

Are you doing this again? Would you kindly stop being a troll and posting against the guidelines?

suraci2• 1 year ago

sorry, i'm still learning to be better at this

egberts1• 1 year ago

Wut?

Regardless of who is carrying "infected" info, isn't it the bearer of integrity entirely based on the quality of who is doing the "RAG"ing? (RAG stands for Retrieval-Augmented Generation and is the back-end part of LLM ecosystem that does the importing of various external data types into its LLM database)

This is a classical GIGO: Garbage In, Garbage Out.

Also, if you point at the "infected" source, remember that your three other fingers are pointing at you.

Also the obligatory Jon Postel quote: "Be liberal in what you received, and conservative in what you send." ... oh wait.

amai• 1 year ago

The russians are rookies! Why don‘t they just ask evil Elon and dump Trump to spread their lies via X and Truth Social? That would be much more effective in poisoning the training data of western LLMs.

Moscow-based global news network has infected Western AI tools

Related Stories