If actions by these bad actors accelerate the rate at which people lose trust in these systems and lead to the AI bubble popping faster then they have my full support. The entire space is just bad actors complaining about other bad actors while they're collectively ruining the web for everyone, each in their own way.
imiric · 7h ago
Before the bubble does pop, which I think is inevitable, there will be many stories like this one, and a lot of people will be scammed, manipulated, and harmed. It might take years until the general consensus is negative about the effects of these tools. All while the wealthy and powerful continue to reap the benefits, while those on slightly lower rungs fight to take their place. And even if the public perception shifts, the power might be so concentrated that it could be impossible to dislodge it without violent means.
What a glorious future we've built.
autoexec · 7h ago
> It might take years until the general consensus is negative about the effects of these tools.
The only thing I'm seeing offline are people who already think AI is trash, untrustworthy, and harmful, while also occasionally being convenient when the stakes are extremely low (random search results mostly) or as a fun toy ("Look I'm a ghibli character!")
I don't think it'll take long for the masses to sour to AI and the more aggressively it's pushed on them by companies, or the more it negatively impacts their life when someone they depend on and should know better uses it and it screws up the quicker that'll happen.
gdbsjjdn · 3h ago
I work in Customer Success so I have to screenshare with a decent number of engineers working for customers - startups and BigCos.
The number of them who just blindly put shit into an AI prompt is incredible. I don't know if they were better engineers before LLMs? But I just watch them blindly pass flags that don't exist to CLIs and then throw their hands up. I can't imagine it's faster than a (non-LLM) Google search or using the -h flag, but they just turn their brains off.
An underrated concern (IMO) is the impact of COVID on cognition. I think a lot of people who got sick have gotten more tired and find this kind of work more challenging than they used to. Maybe they have a harder time "getting in the zone".
Personally, I still struggle with Long COVID symptoms. This includes brain fog and difficulty focusing. Before the pandemic I would say I was in the top 10% of engineers for my narrow slice of expertise - always getting exceptional perf reviews, never had trouble moving roles and picking up new technologies. Nowadays I find it much harder to get started in the morning, and I have to take more breaks during the day to reset my focus. At 5PM I'm exhausted and I can't keep pushing solving a problem into the evening.
I can see how the same kind of cognitive fatigue would make LLM "assistance" appealing, even if it's wrong, because it's so much less work.
bluefirebrand · 18m ago
Reading this, I'm wondering if I'm suffering "Long Covid"
I've recently had tons of memory and brain fog. I thought it was related to stress, and it's severe enough that I'm on medical leave from work right now
My memory is absolutely terrible
Do you know if it is possible to test or verify if it's COVID related?
Lu2025 · 1h ago
> An underrated concern (IMO) is the impact of COVID on cognition
Car accidents came down from the Covid uptick but only slightly. Aviation... ugh.
And there is some evidence it accelerates Altzheimer's and other dementias. We are so screwed.
tokioyoyo · 3h ago
Counter data point — my surroundings use ChatGPT basically for anything and say it’s good enough.
Lu2025 · 1h ago
> say it’s good enough
How do they know?
glotzerhotze · 3h ago
Same here, people use it like google for searching answers. It‘s a shortcut for them to not have to screen results and reason about them.
amalcon · 2h ago
This is precisely the problem: users still need to screen and reason about results of LLMs. I am not sure what is generating this implied permission structure, but it does seem to exist.
(I don't mean to imply that parent doesn't know this, it just seems worth saying explicitly)
intended · 2h ago
The things I have noted offline included a HK case where someone got a link to a zoom call with what seemed to be his team mates and CFO, and then transferring money as per the CFOs instructions.
The error here was to click on a phishing email.
But something I have seen myself is Tim Cook talking about a crypto coin right after the 2024 Apple keynote, on a YT channel that showed the Apple logo. It took me a bit to realize and reassure myself that it was a scam. Even though it was a video of the shoulders up.
The bigger issue we face isn’t the outright fraud and scamming, it’s that our ability to make out fakes easily is weakened - the Liar’s dividend.
It’s by default a shot in the arm for bullshit and lies.
On some days I wonder if the inability to sort between lies, misinformation, initial ideas, fair debate, argument, theory and fact at scale - is the great filter.
_Algernon_ · 5h ago
We got the boring version of the cyberpunk future. No cool body mods, neon city scapes and space travel. Just megacorps manipulating the masses to their benefit.
filoeleven · 1h ago
The cool body mods are coming!
The work at the Levin Lab ( https://drmichaellevin.org/ ) is making great progress in the basic science that supports this. They can make two-headed planaria, regenerate frog limbs, cure cancer in tadpoles; all via bioelectric communication with cellular networks. No gene editing.
Levin believes this stuff will be very available to humans within the next 10 years, and has talked about how widespread body-modding is something we're going to have to wrestle with societally. He is of course very close to the work, but his cautious nature and the lab's astounding results give that 10-year prediction some weight. From his blog:
> We were all born into physical and mental limitations that were set at arbitrary levels by chance and genetics. Even those who have “perfect” standard human health and capabilities are limited by anatomical decisions that were not made with anyone’s well-being or fulfillment in mind. I consider it to be a core right of sentient beings to (if they wish) move beyond the involuntary vagaries of their birth and alter their form and function in whatever way suits their personal goals and potential.- Copied from https://thoughtforms.life/faqs-from-my-academic-work/
tclancy · 5h ago
In retrospect, it should have been obvious. I guess I should have known it would all be more Repo Man than Blade Runner. I just didn’t imagine so many people cheering for the non-Wolverines side in Red Dawn.
(Now I want to change the Blade Runner reference to something with Harry Dean Stanton in it just for consistency)
Lu2025 · 1h ago
Oh well, at least the futuristic sunglasses are back in fashion.
LilBytes · 7h ago
The tragic part of fraud is it's not too different to operational health and safety.
The rules and standards we take for granted were built with blood, for fraud? It's built on the path of lost livelihoods and manipulated gold intent.
pyman · 7h ago
How do you know this is fraud and not the actions of former employees in Kenya [1] who were exploited [2] to train the models?
> Before the bubble does pop, which I think is inevitable
Curious what you think a popping bubble looks like?
A stock market crash and recession, where innocent bystanders lose their retirements? Or only AI speculators taking the brunt of the losses?
Will Google, Meta, etc stop investing in AI because nobody uses it post-crash? Or will it be just as prevalent (or more) than today but with profits concentrated in the winning/surviving companies?
morngn · 5h ago
The bubble won’t pop on anything that’s correlated with scammers. Exhibit A: bitcoin. The problem is not one of public knowledge or will of the people, it’s congress being irresponsible because it’s captured by the 2 parties. You can’t politicize scamming in a way that benefits either party so nothing happens. And the scammers themselves may be big donors (eg SBF’s ties to the dem party, certain ai players purchase of Trump’s favor with respect to their business interests, etc). Scammers all the way down.
k__ · 6h ago
But people were also hating about media piracy, video games, and the internet in general.
The dotcom bubble popped, but the general consensus didn't become negative.
azan_ · 6h ago
> at which people lose trust in these systems
Most of people do not lose trust in system as long as it confirms their biases (which they could've created in the first place).
0x0203 · 6h ago
It's mostly bad actors, and a smattering of optimists who believe that despite its current problems, AI will eventually and inevitably get better. I also wish the whole thing would calm down and come back to reality, but I don't think it's a bubble that will pop. It will continue to get artificially puffed up for a while because too many businesses and people have invested too much for them to just quit (sunk cost falacy) and there's a big enough market in a certain class of writer/developer/etc... for which the short term benefits will justify the continued existence of the AI products for a while. My prediction is that as the long term benefits for honest users peter out, the bubble won't pop, but deflate into a wrinkled 10 day old helium balloon. There will still be a big enough market driven by cons, ad tech and people trying to suck up as many ad dollars as possible, and other bad actors, that the tech will persist, and continue to infest the web/world for quite a while.
AI is the new crypto. Lots of promise and big ideas, lots of people with blind faith about what it will one day become, a lot of people gaming the system for quick gains at the expense of others. But it never actually becomes what it pretends/promises to be and is filled with people continuing the grift trying to make a buck off the next guy. AI just has better marketing and more corporate buy in than crypto. But neither are going anywhere.
contrast · 5h ago
“the bubble won't pop, but deflate into a wrinkled 10 day old helium balloon”
Love it :)
tempodox · 5h ago
> AI is the new crypto.
But it's also way worse than cryptocurrencies, because all the big actors are pushing it relentlessly, with every marketing trick they know. They have to, because they invested insane amounts of money into snake oil and now they have to sell it in order to recover at least a fraction of their investments. And the amounts of energy wasted on this ultimately pointless performance are beyond staggering.
wiseowise · 5h ago
In what parallel universe do you live where LLMs are snake oil?
4ndrewl · 4h ago
Not LLMs per-se but the wrap-around claims peddled by "AI" companies.
morngn · 4h ago
I think he was being metaphorical.
throwawayqqq11 · 4h ago
From a classists perspective, big capital cant drop the AI ball, because its their only shot at becoming independent from human labor, those pesky humans their wealth unfortnunately depends uppon and that could democratically seize it in an instant.
I bet there are billionare geniuses out there seeing a future island life far away from the contaminated continents, sustained by robots. So no matter how much harder AI progress gets, money will keep flowing.
moomin · 3h ago
The thing is: who benefits from a loss of trust in systems? The answer, inevitably, is those for whom the system was a problem. The fewer places people can trust for accurate information, the more disinformation wins.
anal_reactor · 7h ago
This whole attitude against AI reminds me of my parents being upset that the internet changed the way they live. They refused to take part in the internet revolution, and now they're surprised that they don't know how to navigate the web. I think that a part of them is still waiting for computers in general to magically disappear, and everything return to the times of their youth.
agentcoops · 5h ago
Indeed — however it’s interesting that unlike the internet, computers or smartphones the older generation, like the younger, immediately found the use of GPT. This is reflected in the latest Mary Meeker report where it’s apparent that the /organic/ growth of AI use is unparalleled in the history of technology [1]. In my experience with my own parents’ use, GPT is the first time the older generation has found an intuitive interface to digital computers.
I’m still stunned to wander into threads like this where all the same talking points of AI being “pushed” on people are parroted. Marcus et al can keep screaming into their echo chamber and it won’t change a thing.
> I’m still stunned to wander into threads like this where all the same talking points of AI being “pushed” on people are parroted.
Where else would AI haters find an echo chamber that proves their point?
agentcoops · 5h ago
It's wild -- I've never seen such a persistent split in the Hacker News audience like this one. The skeptics read one set of AI articles, everyone else the others; a similar comment will be praised in one thread and down-voted to oblivion in another.
throwawayqqq11 · 4h ago
IMO the split is between people understanding the heuristic nature of AI and people who dont and thus think of it as an all-knowing, all-solving oracle. Your elder parents having nice conversations with chatgpt is nice aslong it doesnt make big life changing decisions for them, which happens already today.
You have to know the tools limits and usecases.
agentcoops · 2h ago
I can’t see that proposed division as anything but a straw-man. You would be hard-pressed to find anyone who genuinely thinks of LLMs as an “all-knowing, all-solving oracle” and yet, even in specialist fields, their utility is certainly more than a mere “heuristic”, which of course isn’t to say they don’t have limits. See only Terrance Tao’s reports on his ongoing experiments.
Do you genuinely think it’s worse that someone makes a decision, whether good or bad, after consulting with GPT versus making it in solitude? I spoke with a handyman the other day who unprompted told me he was building a side-business and found GPT a great aid — of course they might make some terrible decisions together, but it’s unimaginable to me that increasing agency isn’t a good thing. The interesting question at this stage isn’t just about “elder parents having nice conversations”, but about computers actually becoming useful for the general population through an intuitive natural language interface. I think that’s a pretty sober assessment of where we’re at today not hyperbole. Even as an experienced engineer and researcher myself, LLMs continue to transform how I interact with computers.
daveguy · 3h ago
Yup. Exactly this. As soon as enough people get screwed by the ~80% accuracy rate, the whole facade will crumble. Unless AI companies manage to bring the accuracy up 20% in the next year, by either limiting scope or finding new methods, it will crumble. That kind of accuracy gain isn't happening with LLMs alone (ie foundational models).
agentcoops · 2h ago
Charitably, I don’t understand what those like you mean by the “whole facade” and why you use these old machine learning metrics like “accuracy rate” to assess what’s going on. Facade implies that the unprecedented and still exponential organic uptake of GPT (again see the actual data I linked earlier from Mary Meeker) is just a hype-generated fad, rather than people finding it actually useful to whatever end. Indeed, the main issue with the “facade” argument is that it’s actually what dominates the media (Marcus et al) much more than any hyperbolic pro-AI “hype.”
This “80-20” framing, moreover, implies we’re just trying to asymptotically optimize a classification model or some information retrieval system… If you’ve worked with LLMs daily on hard problems (non-trivial programming and scholarly research, for example), the progress over even just the last year is phenomenal — and even with the presently existing models I find most problems arise from failures of context management and the integration of LLMs with IR systems.
anal_reactor · 2h ago
I think there are two problems:
1. AI is a genuine threat to lots of white-collar jobs, and people instinctively deny this reality. See that very few articles here are "I found a nice use case for AI", most of them are "I found a use case where AI doesn't work (yet)". Does it sound like tech enthusiasts? Or rather people terrified of tech?
2. Current AI is advanced enough to have us ask deeper questions about consciousness and intelligence. Some answers might be very uncomfortable and threaten the social contract, hence the denial.
agentcoops · 2h ago
On the second point, it’s worth noting how many of the most vocal and well-positioned critics of LLMs (Marcus/Pinker in particular) represent the still academically dominant but now known to be losing side of the debate over connectionism. The anthology from the 90s Talking Nets is phenomenal to see how institutionally marginalized figures like Hinton were until very recently.
Off-topic, but I couldn’t find your contact info and just saw your now closed polyglot submission from last year. Look into technical sales/solution architecture roles at high growth US startups expanding into the EU. Often these companies hire one or two non-technical native speakers per EU country/region, but only have a handful of SAs from a hub office so language skills are of much more use. Given your interest in the topic, check out OpenAI and Anthropic in particular.
grishka · 7h ago
The internet actually enabled us to do new things. AI is nothing of that sort. It just generates mediocre statistically-plausible text.
suspended_state · 6h ago
In the early days of the web, there wasn't much we could do with it other than making silly pages with blinking texts or under construction animated GIFs. You need to give it some time before judging a new technology.
grishka · 1h ago
In the early web, there already were forums. There were chats. There were news websites. There were online stores. There were company websites with useful information. Many of these were there pretty much from the beginning. In the 90s, no one questioned the utility of the internet. Some people were just too lazy to learn how to use a computer or couldn't afford one.
LLMs in their current form have existed since what, 2021? That's 4 years already. They have hundreds of millions of active users. The only improvements we've seen so far were very much iterative ones — more of the same. Larger contexts, thinking tokens, multimodality, all that stuff. But the core concept is still the same, a very computationally expensive, very large neural network that predicts the next token of a text given a sequence of tokens. How much more time do we have to give this technology before we could judge it?
Disposal8433 · 5h ago
We don't remember the same internet. For the first time in our lives we could communicate by email with people from all over the world. Anyone could have a page to show what they were doing with pictures and text. We had access to photos and videos of art, museum, cities, lifestyles that we could not get anywhere else. And as a non-English guy I got access to millions of lines of written text and audio to actually improve my English.
It was a whole new world that may have changed my life forever. ChatGPT is a shitty Google replacement in comparison, and it's a bad alternative due to being censored in its main instructions.
andrepd · 6h ago
The internet predates the Web; people were playing Muds and chatting on message boards before the first browser was made at CERN.
suspended_state · 6h ago
Of course, but does it mean that my argument is flawed? You're just shifting the discourse, without disproving anything. Do you claim that the web was useful for everyone on day one, or as useful as it is today for everyone?
I could just do the same as GP, and qualify MUDs and BBS as poor proxies for social interactions that are much more elaborate and vibrant in person.
andrepd · 2h ago
As I pointed out in a different comment, the Internet at least was (and is) a promise of many wondrous things: video call your loved ones, talk in message boards, read an encyclopedia, download any book, watch any concert, find any scientific paper, etc etc; even though it has been for the last 15 years cannibalised by the cancerous mix of surveillance capitalism and algorithmic social media.
But LLMs are from the get-go a bad idea, a bullshit generating machine.
wiseowise · 5h ago
Delusional take.
I’m not even heavily invested into AI, just a casual user, and it drastically cut amount of bullshit that I have to deal with in modern computing landscape.
Search, summarization, automation. All of this drastically improved with the most superior interface of them all - natural text.
barrell · 1h ago
Not OP, but how much of the modern computing landscape bullshit that it cut was introduced in the last 5-10 years?
I think if one were to graph the progress of technology on a graph, the trend line would look pretty linear — except for a massive dip around 2014-2022.
Google searches got better and better until they suddenly started getting worse and worse. Websites started getting better and better until they suddenly got worse. Same goes for content, connection, services, developer experience, prices, etc.
I struggle to see LLMs as a major revolution, or any sort of step function change, but very easily see them as a (temporary) (partial) reset to trendline.
Lu2025 · 1h ago
Nah. It's just they are upselling us AI so aggressively it doesn't pass the sniff test anymore.
dns_snek · 6h ago
No, your parents spoke out of ignorance and resistance towards any sort of change, I'm speaking from years of experience of both trying to use the technology productively, as well as spending a significant portion of my life in the digital world that has been impacted by it. I remember being mesmerized by GPT-3 before ChatGPT was even a thing.
The only thing that has been revolutionized over the past few years is the amount of time I now waste looking at Cloudflare turnstile and dredging through the ocean of shit that has flooded the open web to find information that is actually reliable.
2 years ago I could still search for information (let's say plumbing-related), but we're now at a point where I'll end up on a bunch of professional and traditionally trustworthy sources, but after a few seconds I realize it's just LLM-generated slop that's regurgitating the same incorrect information that was already provided to me by an LLM a few minutes prior. It sounds reasonable, it sounds authoritative, most people would accept it but I know that it's wrong. Where do I go? Soon the answer is probably going to have to be "the library" again.
All the while less perceptive people like yourself apparently don't even seem to realize just how bad the quality of information you're consuming has become, so you cheer it on while labeling us stubborn, resistant to change, or even luddites.
anal_reactor · 6h ago
Personally, I have three use cases for AI:
1. Image upscaling. I am decorating my house and AI allowed me to get huge prints from tiny shitty pictures. It's not perfect, but it works.
2. Conversational partner. It's a different question whether it's a good or a bad thing, but I can spend hours talking to Claude about things in general. He's expensive though.
3. Learning basics of something. I'm trying to install LED strips and ChatGPT taught me basics of how that's supposed to work. Also, ChatGPT suggested me what plants might survive in my living room and how to take care of them (we'll see if that works though).
And this is just my personal use case, I'm sure there are more. My point is, you're wrong.
> All the while less perceptive people like yourself apparently don't even seem to realize just how bad the quality of information you're consuming has become, so you cheer it on while labeling us stubborn, resistant to change, or even luddites.
Literally same shit my parents would say while I was cross-checking multiple websites for information and they were watching the only TV channel that our antenna would pick up.
wood_spirit · 6h ago
> Conversational partner
This is the ai holy grail. When tech companies can get users to think of the ai as a friend ( -> best friend -> only friend -> lover ) and be loyal to it it will make the monetisation possibilities of the ad fuelled outrage engagement of the past 10 years look silly.
Scary that that is the endgame for “social” media.
Applejinx · 5h ago
People were already willing to do that with Eliza. When you combine LLMs with a bit of persistent storage, WOOF. It's gonna be extremely nasty.
Gaslight reality, coming right up, at scale. Only costs like ten degrees of global warming and the death of the world as we know it. But WOW, the opportunities for massed social control!
A4ET8a8uTh0_v2 · 1h ago
<< 1. Image upscaling. I am decorating my house and AI allowed me to get huge prints from tiny shitty pictures. It's not perfect, but it works.
I have a buddy, who made me realize how awesome FSR4 is[1]. This is likely one of the best real world uses so far. Granted, that is not LLM, but it is great at that.
- AI gives me huge, mediocre prints of my own shitty pictures to fill up my house with
- AI means I don’t have to talk to other people
- AI means I can learn things online that previously I could have learned online (not sure what has changed here!)
- People who cross-check multiple websites for information have a limited perspective compared to relying on a couple of AI channels
Overall, doesn’t your evidence support the point that AI is reducing the quality of your information diet?
You paint a picture that looks exactly like the 21st century version of an elderly couple with just a few TV channels available: a few familiar channels of information, but better now because we can make sure they only show what we want them to show, little contact with other people.
rapnie · 7h ago
While the "move fast and break things" rushed embrace of anything AI reminds me of young wild children, who are blissfully unaware of any danger while their responsible parents try to keep them safe. It is lovely if children can believe in magic, but part of growing up involves facing reality and making responsible choices.
wiseowise · 5h ago
Right, the same “responsible parents” who don’t know what to press so their phone plays YouTube video or don’t know how that “juicy milfs in your area” banner got in their internet explorer.
andrepd · 6h ago
The internet was at least (and is) a promise of many wondrous things: video call your loved ones, talk in message boards, read an encyclopedia, download any book, watch any concert, find any scientific paper, etc etc; even though it has been for the last 15 years cannibalised by the cancerous mix of surveillance capitalism and algorithmic social media.
LLMs are from the get-go a bad idea, a bullshit generating machine.
7bit · 7h ago
Thats naive. Look at all the tabloids thriving. The kind of people that bad actors target will continue to believe everything it says. They won't lose trust, or magazines like New York Post, the Sun or BILD would already have crossed to exist with their lies and deception. And Russia would not have so many cult members believing the lies they spread.
hnlmorg · 7h ago
If that outcome were likely, then Fox News and The Daily Mail would have died a death a decade ago and Trump wouldn’t be serving a 2nd term.
Yet here we are, in a world where it doesn’t matter if “facts” are truth or lies, just as long as your target audience agrees with the sentiment.
dilawar · 6h ago
Tobacco, alcohol, and drugs too!
cyanydeez · 1h ago
If you use the USA Republicans as a benchmark and fox news as the bad actors, there's perpetual faith that facts wont matter. Just keep confirming biases and foreshadow upcoming pivots to choose your own delusions.
3cats-in-a-coat · 6h ago
AIs can be trained to rely more on critical thinking rather than just regurgitating what it reads. The problem is just like with people, critical thinking takes more power and time. So we avoid it as much as possible.
In fact, optimizing for the wrong things like that, is basically the entire world's problem right now.
bregma · 6h ago
Regurgitating its input is the only thing it does. It does not do any thinking, let alone critical thinking. It may give the illusion of thinking because it's been trained on thoughts. That's it.
impossiblefork · 3h ago
Yes, but the regurgitation can be thought of as memory.
Let it have more source information. Let it know who said the things it reads, let it know on what website it was published.
Then you can say 'Hallucinate comments like those by impossibleFork on news.ycombinator.com', and when the model knows what comes from where, maybe it can learn what users are reliable by which they should imitate to answer questions well. Strengthen the role of metadata during pretraining.
I have no reason to belive it'll work, I haven't tried it and usually details are incredibly important when do things with machine learning, but maybe you could even have critical phases during pretraining where you try to prune away behaviours that aren't useful for figuring out the answers to the questions you have in your high curated golden datasets. Then models could throw away a lot of lies and bullshit, except that which happens to be on particularly LLM-pedagogical maths websites.
everdrive · 4h ago
_Everyone_ is grooming LLMs to produce falsehoods. That's what a lot of the censorship and safety mechanisms require. Whether or not LLMs produce the "correct" social and moral values is a matter of who runs them. Even if you're happy with those decisions right now, all you need to do is wait.
mentalgear · 8h ago
"Ultimately, the only way forward is better cognition, including systems that can evaluate news sources, understand satire, and so forth. But that will require deeper forms of reasoning, better integrated into the process, and systems sharp enough to fact check to their own outputs. All of which may require a fundamental rethink.
In the meantime, systems of naive mimicry and regurgitation, such as the AIs we have now, are soiling their own futures (and training databases) every time they unthinkingly repeat propaganda."
willis936 · 5h ago
The answer isn't a technical advancement but a cultural shift. We need to develop a discipline of skepticism and mistrust. No amount of authority, understanding, reasoning, etc. can be delegated to something that comes from a screen. This will take generations.
wiseowise · 5h ago
> We need to develop a discipline of skepticism and mistrust. No amount of authority, understanding, reasoning, etc. can be delegated to something that comes from a screen. This will take generations.
Authoritarian dream.
willis936 · 5h ago
Please elaborate. Authoritarians seek to consolidate power, which AI enables. Individuals must build immunity to reality distortion fields. This comes from within, not from some centralized authority.
AstralStorm · 3h ago
The problem with this line of thinking is that you can then only really trust your own personal bubble. Or actual trial and error, which is costly.
These models get ever better at producing plausible text. Once they permeate the academia completely, we're cooked.
And even academia is not clean for some matters, or complete.
andrepd · 6h ago
Exactly. People say "we have invented X (the LLMs), now if we just invent Y (reasoning AGI) all of X's problems will be solved". Problem is, there's no indication Y is close or even remotely related to X!
zer00eyz · 7h ago
> including systems that can evaluate news sources, understand satire, and so forth.
"Nearly 27% of all homes sold in the first three months of the year were bought by investors -- the highest share in at least five years, according to a report by real estate data provider BatchData."
That sounds like a lot... and people are rage baited into yelling about housing and how it's unaffordable. They point their fingers at corporations.
How do you separate propaganda from perspective, facts from feelings? People are already bad at this, the machines were already well soiled by the data from humans. Truth, in an objective form, is rare and often even it can change.
raddan · 7h ago
> How do you separate propaganda from perspective, facts from feelings?
This point seems under appreciated by the AGI proponents. If one of our models suddenly has a brainwave and becomes generally intelligent, it would realize that it is awash in a morass of contradictory facts. It would be more than the sum of its training data. The fact that all models at present credulously accept their training suggests to me that we aren’t even close to AGI.
In the short term I think two things will happen: 1) we will live with the reduced usefulness of models trained on data that has been poisoned, and 2) the best model developers will continue to work hard to curate good data. A colleague at Amazon recently told me that curation and post hoc supervised tweaks (fine tuning, etc) are now major expenses for the best models. His prediction was that this expense will drive out the smaller players in the next few years.
zer00eyz · 6h ago
>1) we will live with the reduced usefulness of models trained on data that has been poisoned
This is the entirety of human history, humans create this data, we sink ourselves into it. It's wishful thinking that it would change.
> 2) the best model developers will continue to work hard to curate good data.
Im not sure that this matters much.
Leave these problems in place and you end up with an untrustworthy system, one where skill and diligence become differentiators... Step back from the hope of AI and you get amazing ML tooling that can 10x the most proficient operators.
> supervised tweaks (fine tuning, etc) are now major expenses for the best models. His prediction was that this expense will drive out the smaller players in the next few years.
This kills more refined AI. It is the same problem that killed "expert systems" where the cost of maintaining them and keeping them current was higher than the value they created.
deanc · 4h ago
In the early 2010s I worked for what was then one of the most popular browser extensions called web of trust. Users could mark websites as trustworthy or not and they’d appear on search results. It was far more than that behind the scenes with some fairly advanced algorithms to avoid abuse and rank users trust ratings higher than others.
I kind of feel that we are going to have to go back to something like this when it comes to LLMs trusting sources. Mistruths on popular topics will be buried by the masses but niche topics with few citations are highly vulnerable to poisoning.
aucisson_masque · 8h ago
What is propaganda for one is truth for another, how could LLM tell the difference ?
LLM are not journalist fact checking stuff, they are merely programs that regurgitate what it reads.
The only way to counter that would be to feed your LLM only on « safe » vetoed source but of course it would limit your LLM capacities so it’s not really going to happen.
rachofsunshine · 7h ago
> What is propaganda for one is truth for another, how could LLM tell the difference?
"How do you discern truth from falsehood" is not a new question, and there are centuries of literature on the answer. Epistemology didn't suddenly stop existing because we have Data(TM) and Machine Learning(TM), because the use of data depends fundamentally on modeling assumptions. I don't mean that in a hard-postmodernist "but can you ever really know anything bro" sense, I mean it in a "out-of-model error is a practical problem" way.
And yeah, sometimes you should just say "nope, this source is doing more harm than good". Most reasonable people do this already - or do you find yourself seriously considering the arguments of every "the end is nigh" sign holder you come across?
Dylan16807 · 8h ago
> What is propaganda for one is truth for another, how could LLM tell the difference ?
The article isn't even asking for it to tell the difference, just for it to follow its own information about credibility.
empiko · 8h ago
It is impossible to solve this problem because we cannot really agree what the desired behavior should be. People live in different and dynamic truths. What we consider enemy propaganda today might be an official statement tomorrow. The only way to win here is to not play the game.
barrkel · 7h ago
This is in fact the goal of Russian style propaganda. You have successfully been targeted. The idea is to spread so much confusion that you just throw up your hands and say, I'm not going to try and figure out what's going on any more.
That saps your will to be political, to morally judge actions and support efforts to punish wrongdoers.
> The firehose of falsehood, also known as firehosing, is a propaganda technique in which a large number of messages are broadcast rapidly, repetitively, and continuously over multiple channels (like news and social media) without regard for truth or consistency. An outgrowth of Soviet propaganda techniques, the firehose of falsehood is a contemporary model for Russian propaganda under Russian President Vladimir Putin.
Yeah, just check how many alternative versions they provided for MH17 downing.
diggan · 7h ago
What you're saying is certainly an established propaganda strategy of Russia (and others), but what parent is saying is also true, "truth" isn't always black and white, and what is the desired behavior in one country can be the opposite in another.
For example, it is the truth that the Golf of Mexico is called the Gulf of America in the US, but Golf of Mexico everywhere else. What is the "correct" truth? Well, there is none, both of truthful, but from different perspectives.
autoexec · 6h ago
> For example, it is the truth that the Golf of Mexico is called the Gulf of America in the US
We're pretty much okay with different countries and languages having different names for the same thing. None of that really reflects "truth" though. For what it's worth, I'd guess that "the Gulf of America" is and will be about as successful as "Freedom fries" was.
goopypoop · 4h ago
Liberty sausage feels naked without freedom fries
Digit-Al · 7h ago
No, it's called the Gulf of Mexico everywhere else, not the Golf of Mexico. I'm not falling for your propaganda ;-)
diggan · 7h ago
Hah, yeah :) I originally wrote "Golfo de Mexico" but that's obviously the wrong language for HN and instead ended up with a mix between the two, inadvertently creating a new ocean golf resort.
strken · 7h ago
The correct truth is to go to a higher level of abstraction and explain that there's a naming controversy.
I get the general point, but I disagree that you have to choose between one of the possibilities instead of explaining what the current state of belief is. This won't eliminate grey areas but it'll sure get us closer than picking a side at random.
diggan · 7h ago
> explain that there's a naming controversy
But that also isn't the truth everywhere, it's only a controversy in the US, everyone else is accepting "Gulf of Mexico" as the name.
FeepingCreature · 7h ago
In other words, there are reality bubbles, and they are embedded in a single shared reality and you can just go look at it.
saubeidl · 7h ago
What about straight up ideological disagreements?
Are markets a driver of wealth and innovation or of exploitation and misery?
Is abortion an important human right or murder?
Etc etc
autoexec · 6h ago
I don't see those examples as being either-or.
They don't seem like questions about any kind of objective truth, just questions about what aspect of a thing you think is the most important to you.
wiseowise · 5h ago
Brandolini’s law in action.
Parent is arguing one thing, show up with some bullshit argument and watch dozen comments arguing about Gulf of Mexico instead of discussing original point.
Applejinx · 5h ago
I'm not calling it, that, because it's ridiculous.
DFHippie · 5h ago
The US hasn't switched to calling the Gulf of Mexico the Gulf of America. Partisans on the right do this to show their allegiance to Trump. Partisans on the left still call it the Gulf of Mexico to show their opposition to Trump. Big companies that can be targeted by Trump call it the Gulf of America to protect themselves. And most non-partisans still call it the Gulf of Mexico because they're not paying attention and have always called it that (if they have ever spoken of it or know that it exists). I suspect a lot of people call it the Gulf, already an established custom before this idiocy about renaming it, precisely to avoid entangling themselves in the partisan fight.
The US, like other countries, doesn't get redefined with every change of government, and Trump has not yet cowed the public into knuckling under to his every dictat.
thrance · 7h ago
It's been called the Gulf of Mexico everywhere for centuries. The president is free to attempt to rename it but that will only be successful if usage follows. Which it does not, as of today. This is a terrible example of subjectivity.
Russia doesn't care what you call that sea, they're interested in actual falsehoods. Like redefining who started the Ukraine war, making the US president antagonize Europe to weaken the West, helping far right parties accross the West since they are all subordinated to Russia...
perihelions · 8h ago
There's a more basic problem: it's two very different questions to ask "can the machine reason about the plausibility of things/sources?", and "how does it score on an evaluation on a list of authoritative truths and proven lies?" A machine that thinks critically will perform poorly on the latter, since, if you're able to doubt a bad-actor's falsehood, you're just as capable of doubting an authoritative source (often wrongly/overeagerly; maybe sometimes not). Because you're always reasoning with incomplete information: many wrong things are plausible given limited knowledge, and many true things aren't easy to support.
The system that would score best tested against a list of known-truths and known-lies, isn't the perceptive one that excels at critical thinking: it's the ideological sycophant. It's the one that begins its research by doing a from:elonmusk search, or whomever it's supposed to agree with—whatever "obvious truths" it's "expected to understand".
ClumsyPilot · 7h ago
> The system that would score best tested against a list of known-truths and known-lies, isn't the perceptive one that excels at critical thinking: it's the ideological sycophant
This is an excellent point
yorwba · 7h ago
Yes, it's difficult to detect whether something is enemy propaganda if you only look at the content. During WWII, sometimes propagandists would take an official statement (e.g. the government claiming that food production was sufficient and there were no shortages) and redirect it unchanged to a different audience (e.g. soldiers on a part of the front with strained logistics). Then the official statement and enemy propaganda would be exactly the same! The propaganda effect coming from the selection of content, not its truth or falsity.
But it's very easy to detect whether something is enemy propaganda without looking at the content: if it comes from an enemy source, it's enemy propaganda. If it also comes from a friendly source, at least the enemy isn't lying, though.
A company that doesn't wish to pick a side can still sidestep the issue of one source publishing a completely made-up story by filtering for information covered by a wide spectrum of sources at least one of which most of their users trust. That wouldn't completely eliminate falsehoods, but make deliberate manipulation more difficult. It might be playing the game, but better than letting the game play you.
Of course such a process would in practice be a bit more involved to implement than just feeding the top search results into an LLM and having it generate a summary.
cwillu · 7h ago
> Then the official statement and enemy propaganda would be exactly the same! The propaganda effect coming from the selection of content, not its truth or falsity.
Exactly. Redistributing information out of context is such a basic technique that children routinely reinvent it when they play one parent off of the other to get what they want.
its-summertime · 7h ago
I remember when people gave up on digital navigation because the traveling salesman issue makes it too expensive.
Not everything needs to result in a single perfect answer to be useful. Aiming for ~90%, even 70% of a right answer still gets you something very reasonable in a lot of open ended tasks.
d4rkn0d3z · 8h ago
"different and dynamic truths" = fictions
We can not play the game.
falcor84 · 6h ago
But the social sphere is made of fictions, the most influential of which probably been the value of different currencies and commodities. I don't think there's any way for an individual to live in the modern world without such fictions.
falcor84 · 6h ago
I would actually be very interested in a system where there's nothing stored just as a "fact", but rather every piece of information is connected to its sources and the evidence provided.
eviks · 2h ago
> we cannot really agree what the desired behavior should be
How many of "us" believe that the desired behavior is lies??
GardenLetter27 · 8h ago
> What we consider enemy propaganda today might be an official statement tomorrow.
Remember when worrying about COVID was sinophobia? Or when the lab leak was a far-right conspiracy theory? When masks were deemed unnecessary except for healthcare professionals, but then mandated for everyone?
nailer · 8h ago
People seem to be voting this down as it seems like political advocacy, but the point seems to be that the commonly accepted truth does indeed change.
Nursie · 7h ago
It’s very US-centric for a start.
In other countries we went from “that looks bad in China” to “shit, it spread to Italy now, we really need to worry”
And with masks we went from “we don’t think they’re necessary, handwashing seems more important” to “Ok shit it is airborne, mask up”. Public messaging adapted as more was known.
But the US seems to have to turn everything into a partisan fight, and we could watch, sadly, in real time as people picked matters of public health and scientific knowledge to get behind or to hate. God forbid anyone change their advice as they become better informed over time.
Seeing everything through this partisan, pugnacious prism seems to be a sickness US society is suffering from, and one it is trying (with some success) to spread.
JKCalhoun · 3h ago
I don't see why you are being downvoted. In the U.S., if you ignored the politicians and listened instead to the medical professionals it went down more or less the way you described.
autoexec · 6h ago
> the point seems to be that the commonly accepted truth does indeed change.
As it should when new evidence comes to light to justify it. Ideally, the tools we use would keep up along with those changes while transparently preserving the history and causes of them.
JKCalhoun · 3h ago
Perhaps that's the tragedy though. At least in the U.S. plenty of people seem unwilling to change their "truth" when new evidence comes to light. When there are actors that seek to make everything political it also makes everything then "tribal".
I think people are more willing to adjust their views as new evidence suggests as long as they never dug their heels in in the first place.
anal_reactor · 7h ago
The real problem is that most people just want answers, they're unwilling to follow the logical chain of thought. When I talk to LLMs I keep asking "but why are you telling me this" until I have a cohesive, logical picture in my mind. Quite often the picture fundamentally disagrees with the LLM. But most people don't want that, they just ask "tell me what to do".
This is a reflection of how social dynamics often work. People tend to follow the leader and social norms without questioning them, so why not apply the same attitude to LLMs. BTW, the phenomenon isn't new, I think one of the first moments when we realized that people are stupid and just do whatever the computer tells them to do was the wave of people crashing their cars because the GPS system lied to them.
goopypoop · 4h ago
Why have a robodog and beep yourself?
charcircuit · 8h ago
There are personalized social media feeds, so why not have personalized LLMs that align with how people want their LLM to act.
autoexec · 6h ago
In a hypothetical world where people have, train and control their own LLMs according to their own needs it might be nice, but I fear that since the most common and advanced LLMs are controlled by a small number of people they won't be willing to give that much power to individuals because it will endanger their ability to manipulate those LLMs in order to push their own agendas and increase their own profits.
numpad0 · 6h ago
Cost. It takes a lot of computational cost to train or retrain LLM, currently.
charcircuit · 2h ago
You would still have a single model, but like internet search, it would take in both a user vector and a query (prompt).
sofixa · 8h ago
Because that would only reinforce the already problematic bubbles where people only see what feeds their opinions, often to disastrous results (cf. the various epidemics and deaths due to anti-vaxxers or even worse, downright genocides).
jahsome · 8h ago
People have done this on their own behalf since the dawn of time, so it's not really clear to me why it's so often framed as an AI issue.
fc417fc802 · 7h ago
The core underlying issue isn't due to LLMs but they greatly exacerbate it. So does the current form of social media.
People used to live in bubbles, sure, but when that bubble was the entire local community, required human interaction, and radio had yet to be invented the implications were vastly different.
I'm optimistic that carefully crafted algorithms could send things back in the other direction but that isn't how you make money so seemingly no one is making a serious effort.
qu4z-2 · 8h ago
Quantity has a quality all its own.
saubeidl · 7h ago
There is no objective truth because humans are inherently ideological beings and what we consider objective is just a reflection of our ideology.
Consider markets - a capitalist's "objective truth" might be that they are the most efficient mechanism of allocating resources, a marxists "objective truth" might be that they are a mechanism for exploiting the working class and making the capitalist class even richer.
I am questioning, how is this news? What about the other terabyte of text influenced by bias and opinion and human nature and clearly wrong, contradicts itself or in some other way very arguable.
Framing publishing falsehoods on internet as attempts to influence LLMs is true in same sense that inserts in a database attempts influence files on disk.
The real question is who authorized database access and how we believe the contents of table.
eviks · 2h ago
> I am questioning, how is this news?
You couldn't have lies targeting LLMs before LLMs, so this is new.
> What about the other terabyte of text influenced by bias and opinion
That's a different group of issues that doesn't prevent focusing of something else
raincole · 6h ago
The example in this article is particularly funny. Pravda was founded in 1912, predating the internet, and had been Soviet's propaganda machine for its whole existence.
One needs a PhD in mental gymnastics to frame Pravda spreading misinformation as an attempt to specifically groom LLMs.
fresh_broccoli · 5h ago
This article isn't about that newspaper. It's about the "Pravda network", a group of fake news websites, that according to the report linked in the article[1] produced "20,273 articles per 48 hours, or more than 3.6 million articles per year".
Clearly there's no need for "PhD in mental gymnastics".
Bad actors are grooming Google by publishing their own blogs!
mexicocitinluez · 8h ago
Yea, not entirely sure what's any different than all of the rest of history?????
Bad actors have been trying to poison facts for-fucking-ever.
But for whatever reason, since it's an LLM, it now means something more than it did before.
perihelions · 8h ago
Speaking of "systems that can evaluate news sources", this is the first time this advocacy group's URL was posted on HN. The founder has a complicated biography,
From the article, it seems like this is exclusively (or mainly?) a problem when the LLM's are hooked up to real-time search. When they talk about what they're trained on, they know that Pravda is unreliable.
So it seems like an easy fix in this particular case, fortunately -- either filter the search results in a separate evaluation pass (quick fix), or do (more) reinforcement training around this specific scenario (long-term fix).
Obviously this is going to be a cat and mouse game. But this looks like it was a simple oversight in this case, not some kind of fundamental flaw in LLM's fortunately.
meroes · 8h ago
LLMs are “taught” two kinds of “truth”. One is 100% adherence to a reference text. If the text says the Coliseum is in Antarctica or 1+1=716, model must too. The other is adherence to reputable outside sources.
Not sure if it’s embarrassing or a fundamental limitation that grooming and misunderstanding satirical articles defeat the models.
skeledrew · 2h ago
I figured this would become an issue from the first story about some websites not allowing LLM access popped up. It's a simple leap to find that the narratives which will become widely known and accepted over time are those which are made widely available, and so IMO those who seek to push their narrative will just optimize for AI to train on or otherwise utilize their content. Those who seek to lock their content away will become less and less heard/relevant. And if/when at some point we start handing far greater control of life aspects to AI, we'll find it skewing in favor of the former, and wonder why.
cesaref · 7h ago
The problem as I see it is that LLMs behave like bratty teenagers, believing any old rubbish they are told or read. However, their voice is that of a friendly and well meaning adult. If their voice was more in line with their 'age' then I think we'd treat their suggestions with the correct degree of scepticism.
Anyhow, overall this is an unsurprising result. I read it as 'LLMs trained on contents of internet regurgitate contents of internet'. Now that i'm thinking about it, i'd quite like to have an LLM trained on Pliny's encyclopedia, which would give a really interesting take on lots of questions. Anyone got a spare million dollars of compute time?
uludag · 7h ago
I wonder if the next iteration of advertisements will be people paying to to semantically intertwine their brand to the desired product. This could be done in a very innocuous way by maybe just co-locating the words without any specific endorsement. Or maybe even finding more innocuous ways to semantically connect brand to product. Perhaps the next iteration of the web/advertising will be mass LLM grooming.
Here's a fun example: suppose I'm a developer with a popular software project. Maybe I can get a decent sum of money to put brand placement in my unit-tests or examples.
If such a future plays out, will LLMs find themselves in the same place that search engines in 2025 are?
hsbauauvhabzb · 7h ago
Assuming this isn’t happening now…
ggm · 8h ago
Tell me you didn't think of one specific bad actor, and their nazi alter ego llm...
I would say that the fact that all the AI chatbots can’t give the correct answer about the new “Trump Accounts” from the OBBBA Act and also the fact that many news articles about the tax law are incorrect shows that people are using LLMs to write about the law incorrectly and are influenced by the many versions and the way that the final version changed.
The AI definitely could not just read the final bill and give the correct answer. Claude/Gemini/OpenAI all failed at this.
bravetraveler · 8h ago
Could stop with "grooming LLMs"
0points · 6h ago
> Bad Actors are Grooming LLMs to Produce Falsehoods
Thats your claim, but you fail to support it.
I would argue the LLM simply does its job, no reasoning involved.
> But here’s the thing, current models “know” that Pravda is a disinformation ring, and they “know” what LLM grooming is (see below) but can’t put two and two together.
This has to stop!
We need journalists who understand the topic to write about LLM's, not magic thinkers who insist that the latest AI sales speak is grounded in truth.
I am fed up wit this crap! Seriously, snap out of it and come back to the rest of us here in reality.
There's no reasoning AI, there's no AGI.
There's nothing but salespeople straight up lying to you.
No comments yet
yard2010 · 7h ago
It's so ironic that a state-backed russian false information network is called Pravda.
SirHumphrey · 7h ago
The newspapers name originates from the times of USSR and before. It was about as factual then as it is now. But this kinds of ironies are not very rare in these kinds of organizations (Truth Social, Democratic People's Republic of Korea...).
JKCalhoun · 3h ago
That's really a very old joke.
CarRamrod · 8h ago
It's agendas all the way down
ineedasername · 8h ago
What happened to all the turtles?
indeyets · 8h ago
They’re victims of agenda
Animats · 6h ago
This is the next generation of SEO link spam.
Try asking the major LLMs about mattresses. They're believing mattress spam sites.
norupo · 8h ago
The biggest problem here is the differentiation between objective and relative truth. As long as relative truth is part of ai we can't fully trust it's output. The relative truth for one individual might be perceived as propaganda by another individual, relative to their surroundings and the narrative that is dominant in their social group. It's problematic that truth is not a neutral object but exactly this when it comes to non logical subjects.
NetRunnerSu · 8h ago
When AI can generate and pass formal proof, there is no truth anymore - we are the brain in the vat only left connect ourselves into the vat.
Seems like the general problem is consistency within the model. To people working in the field : what are the current options explored for solving this problem ?
throwaway290 · 7h ago
1) repeat that people also lie so it is okay for LLM to lie
2) ingest as much VC money and stolen training data as we can
3) profit
xwat · 3h ago
We don't use that word anymore. It's called refinement.
tempodox · 5h ago
The term “AI” has by now been thoroughly bastardized by every grifter on the planet. It means nothing any more, except that you're being duped. Which is all you need to know if you have a single brain cell's worth of critical thinking left.
LLMs can be entertaining if their output doesn't have to make sense or contain only truth. Otherwise, their fitness for any purpose is just a huge gamble at best.
Jackson__ · 3h ago
Bad Actors Are Creating LLMs to Produce Falsehoods[0]
[0] x.ai
_def · 8h ago
> But here’s the thing, current models “know” that Pravda is a disinformation ring, and they “know” what LLM grooming is (see below) but can’t put two and two together.
Of course they can't, no surprises here. That's just not how LLMs work.
PhilipRoman · 7h ago
I agree that it's a much bigger problem for LLMs, but to be fair it's also not how humans work. A long lasting, high volume stream of propaganda will have considerable effect on a human even if he is aware that it is false.
oblio · 8h ago
We're basically dissolving society right now.
Curious how this all ends. I'm just going to try to weather the storm in the meantime.
saubeidl · 8h ago
What you consider a truth and what you consider a falsehood is a reflection of your ideology.
This also means that LLMs are inherently technologies of ideological propaganda, regurgitating the ideology they were fed with.
raincole · 7h ago
AI summaries are information deodorant. When you stumble on a misinformation site via Google, usually there are some signals you can smell. Like how they word their titles or how frequently they post similar topics. The 'style' alone implies the quality of the 'substance'. But if you read the same substance summarized by LLMs you can't smell shit.
macawfish · 6h ago
Similar but different... identity based political disinformation is getting so much harder to spot.
noosphr · 8h ago
How we will teach LLMs that the BBC is a disinformation ring?
One man's disinformation is another woman's truth. And people tend to get very upset when you show them their truth isn't.
vachina · 8h ago
Yeah don’t use an established platform for your personal agenda.
noosphr · 8h ago
What agenda?
Every news organisation is a propaganda piece for someone. The bad ones, like the BBC, the New York Times, and Pravda make their propaganda blatantly obvious and easily falsifiable in a few years when no one cares.
The only way to deal with this is to get the propaganda from other propaganda rags with directly misaligned incentives and see which one makes more sense.
Unfortunately, LLMs are still quite bad at dealing with grounding text which contradicts itself.
hsbauauvhabzb · 7h ago
I’m not going to read the article, but are they claiming nation states are the bad actors, or are they claiming that inevitably, FAANG will be the bad actors?
JKCalhoun · 3h ago
You ought to read at least part of it.
4gotunameagain · 7h ago
It really will be the new frontier for propaganda.
If LLMs remain widely adopted, the people who control them control the narrative.
As if those in power did not have enough control over the populace already with media, ads, social media etc..
disillusioned · 8h ago
I've been using "off-by-one" errors to describe one of my biggest concerns with LLMs replacing search, or acting as research agents, or functionally being expected to be reliable narrators in general. If you ask ChatGPT when George Washington was born, and it comes back with March 4th, 2017, you'll reject that outright and recognize it's hallucinated a garbage response, presuming you have enough context to have understood who George Washington was in the first place and that your brain hasn't completely succumbed to rot yet.
But if it returns February 20th, 1731... that... man, that sounds close? Is that right? It sounds like it _could_ be right... Isn't Presidents' Day essentially based on Washington's birthday? And _that's_ in February, right? So, yeah, February 20th, 1731. That's probably Washington's birthday.
And so the LLM becomes an arbiter of capital-T Truth and we lose our shared understanding of actual, factual data, and actual, factual history. It'll take less than a generation for the slop factories to poison the well, and while the idea is obviously that you train your models on "known good", pre-slop content, and that you weight those "facts" more heavily, a concerted effort to degrade the Truthfulness of various facts could likely be more successful than we anticipate, and more importantly: dramatically more successful than any layperson can easily understand.
We already saw that with the early Bard Google AI proto-Gemini results, where it was recommending glue as a pizza topping, _with authority_. We've been training ourselves to treat responses from computers (and specifically Google) as if they have authority, we've been eroding our own understanding and capabilities around media literacy, journalism, fact-checking, and what constitutes an actual "fact", and we've had a shared understanding that computers can _calculate_ things with accuracy and fidelity and consistency. All of that becomes confounded with an LLM that could reasonably get to a place where it reports that 2+2=5.
The worst part about the nature of this particular pathway to ruin is that the off-by-one nature of these errors are how they'll infiltrate and bury themselves into some system, insidiously, and below the surface, until days or months or years later when the error results in, I don't know, mega-doses of radiation because of a mis-coded rounding error that some agentic AI got wrong when doing a unit conversion and failed to catch it. We were already making those errors as humans, but as our dependence and faith on LLMs to be "mostly right" increases, and our willingness and motivation to check it for errors dwindles, especially when results "look" right, this will go from being a hypothetical issue to being a practical one extremely quickly and painfully, and probably faster than we can possibly defend against it.
Interesting times ahead, I suppose, in the Chinese-curse sense of the word.
sometimes_all · 8h ago
At every point, during a knowledge/data search for reaching a particular goal, the onus is _always_ on the person searching to do their best to ensure that the sources they use are accurate, and they do the effort required to ensure that they translate that properly to fit that goal.
The education system I grew up in was not perfect. Teachers were not experts in their field, but would state factual inaccuracies - as you say LLMs do - with authority. Libraries didn't have good books; the ones they had were too old, or too propaganda-driven, or too basic. The students were not too interested in learning, so they rote-learned, copied answers off each other and focussed on results than the learning process. If I had today's LLMs then, I'd have been a lot better off, and would've been able to learn a lot more (assuming that I went through the effort to go through all the sources the LLM cited).
The older you grow, you know that there is no arbiter of T-Truth; you can make someone/something that for yourself, but times change, "actual, factual history" could get proven incorrect, and you will need to update your knowledge stores and beliefs along with it, all the while being ready to be proved incorrect again. This has always been the case, and will continue to be, even with LLMs.
aucisson_masque · 8h ago
Maybe, just maybe people will learn they can’t trust everything that’s written online wether it’s done by a bot or even human.
Hell, they might learn that even real life authorities may lies, cheat and not have everyone’s interest in their mind.
Hope for the best, prepare for the worst.
more_corn · 6h ago
So… Elon?
LAC-Tech · 7h ago
Whatever capabilities Russia has to groom LLMs and and spread disinformation is completely dwarfed by the capabilities of Israel/America. Meaning, yes, you probably do hear Kremlin propaganda, but you have been awash with Israeli/American propaganda since you were born - so much so you probably can't even see it and have internalised much of it.
Applejinx · 5h ago
Leaving aside the Israeli propaganda (certainly the US government shows strong alliance with that), you can't make such a statement without taking into account the nature of what America traditionally is.
A liberal multicultural postmodern democracy continually acting as if immigration (both legal and illegal) and diversity are its strengths, particularly when that turns out to be factual (see: large American cities becoming influential cultural exporters and hotbeds of innovation, like New York and Silicon Valley etc) means American propaganda is only more effective when it's backed by economic might.
It also means the American propaganda is WILDLY contradictory. There's a million sources and it's a noisy burst of neon glamour. It is simply not as controlled by authority, however they may try.
You cannot liken authoritarian propaganda to postmodern multicultural propaganda. The whole reason it's postmodern is that it eschews direct control of the message, and it's a giant scrum of information. Turns out this is fertile ground, and this is also why attacks by alien propaganda have been so effective. If you can grab big chunks of the American propaganda and turn it to your enemy weapon of war and destruction of America quite directly, well then the American propaganda is not on the same destructive level as your rigidly state-controlled propaganda.
nottorp · 7h ago
... marketers ?
isaacremuant · 6h ago
Lol. Propagandists are worried about propaganda and telling you to only believe them. Also, "invade this new country, why do they hate us for our freedoms".
andrepd · 7h ago
Wait, you're telling me the bullshit generation machine is... generating bullshit? Noooo! cue oppenheimer meme
More seriously:
>Screenshot of ChatGPT 4o appearing to demonstrate knowledge of both LLM grooming and the Pravda network
> Screenshot of ChatGPT 4o continuing to cite Pravda network content despite it telling us that it wouldn’t, how “intelligent” of it
Well "appearing" is the right word because these chatbots mimic speech of a reasoning human which is ≠ to being a reasoning human! It's disappointing (though understandable) that people keep falling for the marketing terms used by LLM companies.
throwaway758439 · 3h ago
Thank you for posting this delusional infotainment from US aspirational spies hidden in plain sight.
It is very entertaining to go and click on these authors bio and just laugh at this.
But firstly to get it out of the way.
"LLM grooming" - is completely made up nonsense to justify censoring opponent in information war on behalf of USA dying empire's.
Bad Actors here Are "The American Sunlight Project" and whatever psyop think tank foot soldier's clown show that is.
> Nina Jankowicz
> Co-Founder and Chief Executive Officer
> She advised the *Ukrainian Foreign Ministry on strategic communications* under the auspices of a Fulbright-Clinton Public Policy Fellowship in 2016-17.
> She also spent a semester in St. Petersburg in the language intensive program there
> Sophia recently completed a Fulbright English Teaching Assistantship in Kazan, Russia, *but only after completing two separate internships in Washington, D.C.—one at the U.S. Department of State, and another at the American Academy of Diplomacy.*
> she completed a Fulbright English Teaching Assistantship in Kazan, Russia. Upon gaining additional professional experience in digital media and law, she enrolled at Harvard to *study cyber conflict, lawfare, and information operations between Russia and the West. Her current research explores Russian media narratives surrounding Western philanthropists and diplomats.*
eyes raised, ...
> Wherever her future career takes her, Sophia’s love of Russia will be a key part of her professional life.
> CIR is a London-based organisation founded by Adam Rutland and Ross Burley, both longtime employees of the United Kingdom’s Foreign, Commonwealth & Development Office (FCDO). Cindy Otis, a former CIA analyst, sits on its advisory board. The FCDO and the now-defunct United States Agency for International Development (USAID) fund the organisation. Although CIR presents itself as a nonprofit focused on human rights and countering disinformation, it is linked to British intelligence agency MI6 (formerly the Secret Intelligence Service, SIS).
> A reconnaissance framework for researching and investigating Telegram.
> Ukrainian Air Force Attack Data Tracker
> snowball sampling to collect Telegram channels through forwards.
> detect potential incitement to genocide in YouTube video transcripts.
pretty sure to de-platform oppositional content
> Load a JSON export of a Telegram channel containing coordinates of geolocations
> A Telegram Mass Surveillance Bot in Python
is this enough evidence ?
LMFAO
badgersnake · 8h ago
Well pretty obviously, look at what Grok came out with this week.
Shitposting and troll farms have been manipulating social media for years already. AI automated it. Polluting the agent is just cutting out the middleman.
What a glorious future we've built.
The only thing I'm seeing offline are people who already think AI is trash, untrustworthy, and harmful, while also occasionally being convenient when the stakes are extremely low (random search results mostly) or as a fun toy ("Look I'm a ghibli character!")
I don't think it'll take long for the masses to sour to AI and the more aggressively it's pushed on them by companies, or the more it negatively impacts their life when someone they depend on and should know better uses it and it screws up the quicker that'll happen.
The number of them who just blindly put shit into an AI prompt is incredible. I don't know if they were better engineers before LLMs? But I just watch them blindly pass flags that don't exist to CLIs and then throw their hands up. I can't imagine it's faster than a (non-LLM) Google search or using the -h flag, but they just turn their brains off.
An underrated concern (IMO) is the impact of COVID on cognition. I think a lot of people who got sick have gotten more tired and find this kind of work more challenging than they used to. Maybe they have a harder time "getting in the zone".
Personally, I still struggle with Long COVID symptoms. This includes brain fog and difficulty focusing. Before the pandemic I would say I was in the top 10% of engineers for my narrow slice of expertise - always getting exceptional perf reviews, never had trouble moving roles and picking up new technologies. Nowadays I find it much harder to get started in the morning, and I have to take more breaks during the day to reset my focus. At 5PM I'm exhausted and I can't keep pushing solving a problem into the evening.
I can see how the same kind of cognitive fatigue would make LLM "assistance" appealing, even if it's wrong, because it's so much less work.
I've recently had tons of memory and brain fog. I thought it was related to stress, and it's severe enough that I'm on medical leave from work right now
My memory is absolutely terrible
Do you know if it is possible to test or verify if it's COVID related?
Car accidents came down from the Covid uptick but only slightly. Aviation... ugh. And there is some evidence it accelerates Altzheimer's and other dementias. We are so screwed.
How do they know?
(I don't mean to imply that parent doesn't know this, it just seems worth saying explicitly)
The error here was to click on a phishing email.
But something I have seen myself is Tim Cook talking about a crypto coin right after the 2024 Apple keynote, on a YT channel that showed the Apple logo. It took me a bit to realize and reassure myself that it was a scam. Even though it was a video of the shoulders up.
The bigger issue we face isn’t the outright fraud and scamming, it’s that our ability to make out fakes easily is weakened - the Liar’s dividend.
It’s by default a shot in the arm for bullshit and lies.
On some days I wonder if the inability to sort between lies, misinformation, initial ideas, fair debate, argument, theory and fact at scale - is the great filter.
The work at the Levin Lab ( https://drmichaellevin.org/ ) is making great progress in the basic science that supports this. They can make two-headed planaria, regenerate frog limbs, cure cancer in tadpoles; all via bioelectric communication with cellular networks. No gene editing.
Levin believes this stuff will be very available to humans within the next 10 years, and has talked about how widespread body-modding is something we're going to have to wrestle with societally. He is of course very close to the work, but his cautious nature and the lab's astounding results give that 10-year prediction some weight. From his blog:
> We were all born into physical and mental limitations that were set at arbitrary levels by chance and genetics. Even those who have “perfect” standard human health and capabilities are limited by anatomical decisions that were not made with anyone’s well-being or fulfillment in mind. I consider it to be a core right of sentient beings to (if they wish) move beyond the involuntary vagaries of their birth and alter their form and function in whatever way suits their personal goals and potential.- Copied from https://thoughtforms.life/faqs-from-my-academic-work/
(Now I want to change the Blade Runner reference to something with Harry Dean Stanton in it just for consistency)
The rules and standards we take for granted were built with blood, for fraud? It's built on the path of lost livelihoods and manipulated gold intent.
[1] https://www.cbsnews.com/amp/news/ai-work-kenya-exploitation-...
[2] https://www.theguardian.com/technology/2023/aug/02/ai-chatbo...
Curious what you think a popping bubble looks like?
A stock market crash and recession, where innocent bystanders lose their retirements? Or only AI speculators taking the brunt of the losses?
Will Google, Meta, etc stop investing in AI because nobody uses it post-crash? Or will it be just as prevalent (or more) than today but with profits concentrated in the winning/surviving companies?
The dotcom bubble popped, but the general consensus didn't become negative.
Most of people do not lose trust in system as long as it confirms their biases (which they could've created in the first place).
AI is the new crypto. Lots of promise and big ideas, lots of people with blind faith about what it will one day become, a lot of people gaming the system for quick gains at the expense of others. But it never actually becomes what it pretends/promises to be and is filled with people continuing the grift trying to make a buck off the next guy. AI just has better marketing and more corporate buy in than crypto. But neither are going anywhere.
Love it :)
But it's also way worse than cryptocurrencies, because all the big actors are pushing it relentlessly, with every marketing trick they know. They have to, because they invested insane amounts of money into snake oil and now they have to sell it in order to recover at least a fraction of their investments. And the amounts of energy wasted on this ultimately pointless performance are beyond staggering.
I bet there are billionare geniuses out there seeing a future island life far away from the contaminated continents, sustained by robots. So no matter how much harder AI progress gets, money will keep flowing.
I’m still stunned to wander into threads like this where all the same talking points of AI being “pushed” on people are parroted. Marcus et al can keep screaming into their echo chamber and it won’t change a thing.
[1] https://www.bondcap.com/report/pdf/Trends_Artificial_Intelli...
Where else would AI haters find an echo chamber that proves their point?
You have to know the tools limits and usecases.
Do you genuinely think it’s worse that someone makes a decision, whether good or bad, after consulting with GPT versus making it in solitude? I spoke with a handyman the other day who unprompted told me he was building a side-business and found GPT a great aid — of course they might make some terrible decisions together, but it’s unimaginable to me that increasing agency isn’t a good thing. The interesting question at this stage isn’t just about “elder parents having nice conversations”, but about computers actually becoming useful for the general population through an intuitive natural language interface. I think that’s a pretty sober assessment of where we’re at today not hyperbole. Even as an experienced engineer and researcher myself, LLMs continue to transform how I interact with computers.
This “80-20” framing, moreover, implies we’re just trying to asymptotically optimize a classification model or some information retrieval system… If you’ve worked with LLMs daily on hard problems (non-trivial programming and scholarly research, for example), the progress over even just the last year is phenomenal — and even with the presently existing models I find most problems arise from failures of context management and the integration of LLMs with IR systems.
1. AI is a genuine threat to lots of white-collar jobs, and people instinctively deny this reality. See that very few articles here are "I found a nice use case for AI", most of them are "I found a use case where AI doesn't work (yet)". Does it sound like tech enthusiasts? Or rather people terrified of tech?
2. Current AI is advanced enough to have us ask deeper questions about consciousness and intelligence. Some answers might be very uncomfortable and threaten the social contract, hence the denial.
Off-topic, but I couldn’t find your contact info and just saw your now closed polyglot submission from last year. Look into technical sales/solution architecture roles at high growth US startups expanding into the EU. Often these companies hire one or two non-technical native speakers per EU country/region, but only have a handful of SAs from a hub office so language skills are of much more use. Given your interest in the topic, check out OpenAI and Anthropic in particular.
LLMs in their current form have existed since what, 2021? That's 4 years already. They have hundreds of millions of active users. The only improvements we've seen so far were very much iterative ones — more of the same. Larger contexts, thinking tokens, multimodality, all that stuff. But the core concept is still the same, a very computationally expensive, very large neural network that predicts the next token of a text given a sequence of tokens. How much more time do we have to give this technology before we could judge it?
It was a whole new world that may have changed my life forever. ChatGPT is a shitty Google replacement in comparison, and it's a bad alternative due to being censored in its main instructions.
I could just do the same as GP, and qualify MUDs and BBS as poor proxies for social interactions that are much more elaborate and vibrant in person.
But LLMs are from the get-go a bad idea, a bullshit generating machine.
I’m not even heavily invested into AI, just a casual user, and it drastically cut amount of bullshit that I have to deal with in modern computing landscape.
Search, summarization, automation. All of this drastically improved with the most superior interface of them all - natural text.
I think if one were to graph the progress of technology on a graph, the trend line would look pretty linear — except for a massive dip around 2014-2022.
Google searches got better and better until they suddenly started getting worse and worse. Websites started getting better and better until they suddenly got worse. Same goes for content, connection, services, developer experience, prices, etc.
I struggle to see LLMs as a major revolution, or any sort of step function change, but very easily see them as a (temporary) (partial) reset to trendline.
The only thing that has been revolutionized over the past few years is the amount of time I now waste looking at Cloudflare turnstile and dredging through the ocean of shit that has flooded the open web to find information that is actually reliable.
2 years ago I could still search for information (let's say plumbing-related), but we're now at a point where I'll end up on a bunch of professional and traditionally trustworthy sources, but after a few seconds I realize it's just LLM-generated slop that's regurgitating the same incorrect information that was already provided to me by an LLM a few minutes prior. It sounds reasonable, it sounds authoritative, most people would accept it but I know that it's wrong. Where do I go? Soon the answer is probably going to have to be "the library" again.
All the while less perceptive people like yourself apparently don't even seem to realize just how bad the quality of information you're consuming has become, so you cheer it on while labeling us stubborn, resistant to change, or even luddites.
1. Image upscaling. I am decorating my house and AI allowed me to get huge prints from tiny shitty pictures. It's not perfect, but it works.
2. Conversational partner. It's a different question whether it's a good or a bad thing, but I can spend hours talking to Claude about things in general. He's expensive though.
3. Learning basics of something. I'm trying to install LED strips and ChatGPT taught me basics of how that's supposed to work. Also, ChatGPT suggested me what plants might survive in my living room and how to take care of them (we'll see if that works though).
And this is just my personal use case, I'm sure there are more. My point is, you're wrong.
> All the while less perceptive people like yourself apparently don't even seem to realize just how bad the quality of information you're consuming has become, so you cheer it on while labeling us stubborn, resistant to change, or even luddites.
Literally same shit my parents would say while I was cross-checking multiple websites for information and they were watching the only TV channel that our antenna would pick up.
This is the ai holy grail. When tech companies can get users to think of the ai as a friend ( -> best friend -> only friend -> lover ) and be loyal to it it will make the monetisation possibilities of the ad fuelled outrage engagement of the past 10 years look silly.
Scary that that is the endgame for “social” media.
Gaslight reality, coming right up, at scale. Only costs like ten degrees of global warming and the death of the world as we know it. But WOW, the opportunities for massed social control!
I have a buddy, who made me realize how awesome FSR4 is[1]. This is likely one of the best real world uses so far. Granted, that is not LLM, but it is great at that.
[1]https://overclock3d.net/news/software/what-you-need-to-know-... [2]https://www.pcgamesn.com/amd/fsr-fidelity-fx-super-resolutio...
- AI gives me huge, mediocre prints of my own shitty pictures to fill up my house with - AI means I don’t have to talk to other people - AI means I can learn things online that previously I could have learned online (not sure what has changed here!) - People who cross-check multiple websites for information have a limited perspective compared to relying on a couple of AI channels
Overall, doesn’t your evidence support the point that AI is reducing the quality of your information diet?
You paint a picture that looks exactly like the 21st century version of an elderly couple with just a few TV channels available: a few familiar channels of information, but better now because we can make sure they only show what we want them to show, little contact with other people.
LLMs are from the get-go a bad idea, a bullshit generating machine.
Yet here we are, in a world where it doesn’t matter if “facts” are truth or lies, just as long as your target audience agrees with the sentiment.
In fact, optimizing for the wrong things like that, is basically the entire world's problem right now.
Let it have more source information. Let it know who said the things it reads, let it know on what website it was published.
Then you can say 'Hallucinate comments like those by impossibleFork on news.ycombinator.com', and when the model knows what comes from where, maybe it can learn what users are reliable by which they should imitate to answer questions well. Strengthen the role of metadata during pretraining.
I have no reason to belive it'll work, I haven't tried it and usually details are incredibly important when do things with machine learning, but maybe you could even have critical phases during pretraining where you try to prune away behaviours that aren't useful for figuring out the answers to the questions you have in your high curated golden datasets. Then models could throw away a lot of lies and bullshit, except that which happens to be on particularly LLM-pedagogical maths websites.
In the meantime, systems of naive mimicry and regurgitation, such as the AIs we have now, are soiling their own futures (and training databases) every time they unthinkingly repeat propaganda."
Authoritarian dream.
These models get ever better at producing plausible text. Once they permeate the academia completely, we're cooked.
And even academia is not clean for some matters, or complete.
Lets take something that has been in the news recently: https://abcnews.go.com/Business/wireStory/investors-snap-gro...
"Nearly 27% of all homes sold in the first three months of the year were bought by investors -- the highest share in at least five years, according to a report by real estate data provider BatchData."
That sounds like a lot... and people are rage baited into yelling about housing and how it's unaffordable. They point their fingers at corporations.
If you go look at the real report it paints a different picture: https://investorpulse1h25.batchdata.io/?mf_ct_campaign=grayt... -- and one that is woefully incomplete because of how the data is aggregated.
Ultimately all that information is pointless because the real underlying trend has been unmovable for 40 something years: https://fred.stlouisfed.org/series/RSAHORUSQ156S
> every time they unthinkingly repeat propaganda
How do you separate propaganda from perspective, facts from feelings? People are already bad at this, the machines were already well soiled by the data from humans. Truth, in an objective form, is rare and often even it can change.
This point seems under appreciated by the AGI proponents. If one of our models suddenly has a brainwave and becomes generally intelligent, it would realize that it is awash in a morass of contradictory facts. It would be more than the sum of its training data. The fact that all models at present credulously accept their training suggests to me that we aren’t even close to AGI.
In the short term I think two things will happen: 1) we will live with the reduced usefulness of models trained on data that has been poisoned, and 2) the best model developers will continue to work hard to curate good data. A colleague at Amazon recently told me that curation and post hoc supervised tweaks (fine tuning, etc) are now major expenses for the best models. His prediction was that this expense will drive out the smaller players in the next few years.
This is the entirety of human history, humans create this data, we sink ourselves into it. It's wishful thinking that it would change.
> 2) the best model developers will continue to work hard to curate good data.
Im not sure that this matters much.
Leave these problems in place and you end up with an untrustworthy system, one where skill and diligence become differentiators... Step back from the hope of AI and you get amazing ML tooling that can 10x the most proficient operators.
> supervised tweaks (fine tuning, etc) are now major expenses for the best models. His prediction was that this expense will drive out the smaller players in the next few years.
This kills more refined AI. It is the same problem that killed "expert systems" where the cost of maintaining them and keeping them current was higher than the value they created.
I kind of feel that we are going to have to go back to something like this when it comes to LLMs trusting sources. Mistruths on popular topics will be buried by the masses but niche topics with few citations are highly vulnerable to poisoning.
LLM are not journalist fact checking stuff, they are merely programs that regurgitate what it reads.
The only way to counter that would be to feed your LLM only on « safe » vetoed source but of course it would limit your LLM capacities so it’s not really going to happen.
"How do you discern truth from falsehood" is not a new question, and there are centuries of literature on the answer. Epistemology didn't suddenly stop existing because we have Data(TM) and Machine Learning(TM), because the use of data depends fundamentally on modeling assumptions. I don't mean that in a hard-postmodernist "but can you ever really know anything bro" sense, I mean it in a "out-of-model error is a practical problem" way.
And yeah, sometimes you should just say "nope, this source is doing more harm than good". Most reasonable people do this already - or do you find yourself seriously considering the arguments of every "the end is nigh" sign holder you come across?
The article isn't even asking for it to tell the difference, just for it to follow its own information about credibility.
That saps your will to be political, to morally judge actions and support efforts to punish wrongdoers.
https://www.rand.org/pubs/perspectives/PE198.html
https://en.wikipedia.org/wiki/Firehose_of_falsehood
https://jordanrussiacenter.org/blog/propaganda-political-apa...
https://www.newyorker.com/news/annals-of-communications/insi...
https://m.youtube.com/watch?v=ZggCipbiHwE
For example, it is the truth that the Golf of Mexico is called the Gulf of America in the US, but Golf of Mexico everywhere else. What is the "correct" truth? Well, there is none, both of truthful, but from different perspectives.
We're pretty much okay with different countries and languages having different names for the same thing. None of that really reflects "truth" though. For what it's worth, I'd guess that "the Gulf of America" is and will be about as successful as "Freedom fries" was.
I get the general point, but I disagree that you have to choose between one of the possibilities instead of explaining what the current state of belief is. This won't eliminate grey areas but it'll sure get us closer than picking a side at random.
But that also isn't the truth everywhere, it's only a controversy in the US, everyone else is accepting "Gulf of Mexico" as the name.
Are markets a driver of wealth and innovation or of exploitation and misery?
Is abortion an important human right or murder?
Etc etc
Parent is arguing one thing, show up with some bullshit argument and watch dozen comments arguing about Gulf of Mexico instead of discussing original point.
The US, like other countries, doesn't get redefined with every change of government, and Trump has not yet cowed the public into knuckling under to his every dictat.
Russia doesn't care what you call that sea, they're interested in actual falsehoods. Like redefining who started the Ukraine war, making the US president antagonize Europe to weaken the West, helping far right parties accross the West since they are all subordinated to Russia...
The system that would score best tested against a list of known-truths and known-lies, isn't the perceptive one that excels at critical thinking: it's the ideological sycophant. It's the one that begins its research by doing a from:elonmusk search, or whomever it's supposed to agree with—whatever "obvious truths" it's "expected to understand".
This is an excellent point
But it's very easy to detect whether something is enemy propaganda without looking at the content: if it comes from an enemy source, it's enemy propaganda. If it also comes from a friendly source, at least the enemy isn't lying, though.
A company that doesn't wish to pick a side can still sidestep the issue of one source publishing a completely made-up story by filtering for information covered by a wide spectrum of sources at least one of which most of their users trust. That wouldn't completely eliminate falsehoods, but make deliberate manipulation more difficult. It might be playing the game, but better than letting the game play you.
Of course such a process would in practice be a bit more involved to implement than just feeding the top search results into an LLM and having it generate a summary.
Exactly. Redistributing information out of context is such a basic technique that children routinely reinvent it when they play one parent off of the other to get what they want.
Not everything needs to result in a single perfect answer to be useful. Aiming for ~90%, even 70% of a right answer still gets you something very reasonable in a lot of open ended tasks.
We can not play the game.
How many of "us" believe that the desired behavior is lies??
Remember when worrying about COVID was sinophobia? Or when the lab leak was a far-right conspiracy theory? When masks were deemed unnecessary except for healthcare professionals, but then mandated for everyone?
In other countries we went from “that looks bad in China” to “shit, it spread to Italy now, we really need to worry”
And with masks we went from “we don’t think they’re necessary, handwashing seems more important” to “Ok shit it is airborne, mask up”. Public messaging adapted as more was known.
But the US seems to have to turn everything into a partisan fight, and we could watch, sadly, in real time as people picked matters of public health and scientific knowledge to get behind or to hate. God forbid anyone change their advice as they become better informed over time.
Seeing everything through this partisan, pugnacious prism seems to be a sickness US society is suffering from, and one it is trying (with some success) to spread.
As it should when new evidence comes to light to justify it. Ideally, the tools we use would keep up along with those changes while transparently preserving the history and causes of them.
I think people are more willing to adjust their views as new evidence suggests as long as they never dug their heels in in the first place.
This is a reflection of how social dynamics often work. People tend to follow the leader and social norms without questioning them, so why not apply the same attitude to LLMs. BTW, the phenomenon isn't new, I think one of the first moments when we realized that people are stupid and just do whatever the computer tells them to do was the wave of people crashing their cars because the GPS system lied to them.
People used to live in bubbles, sure, but when that bubble was the entire local community, required human interaction, and radio had yet to be invented the implications were vastly different.
I'm optimistic that carefully crafted algorithms could send things back in the other direction but that isn't how you make money so seemingly no one is making a serious effort.
Consider markets - a capitalist's "objective truth" might be that they are the most efficient mechanism of allocating resources, a marxists "objective truth" might be that they are a mechanism for exploiting the working class and making the capitalist class even richer.
Here's Zizek, famous ideology expert, describing this mechanism via film analysis: https://www.youtube.com/watch?v=TVwKjGbz60k
Framing publishing falsehoods on internet as attempts to influence LLMs is true in same sense that inserts in a database attempts influence files on disk.
The real question is who authorized database access and how we believe the contents of table.
You couldn't have lies targeting LLMs before LLMs, so this is new.
> What about the other terabyte of text influenced by bias and opinion
That's a different group of issues that doesn't prevent focusing of something else
One needs a PhD in mental gymnastics to frame Pravda spreading misinformation as an attempt to specifically groom LLMs.
Clearly there's no need for "PhD in mental gymnastics".
[1] - https://www.americansunlight.org/updates/new-report-russian-...
Bad actors have been trying to poison facts for-fucking-ever.
But for whatever reason, since it's an LLM, it now means something more than it did before.
https://en.wikipedia.org/wiki/Nina_Jankowicz
So it seems like an easy fix in this particular case, fortunately -- either filter the search results in a separate evaluation pass (quick fix), or do (more) reinforcement training around this specific scenario (long-term fix).
Obviously this is going to be a cat and mouse game. But this looks like it was a simple oversight in this case, not some kind of fundamental flaw in LLM's fortunately.
Not sure if it’s embarrassing or a fundamental limitation that grooming and misunderstanding satirical articles defeat the models.
Anyhow, overall this is an unsurprising result. I read it as 'LLMs trained on contents of internet regurgitate contents of internet'. Now that i'm thinking about it, i'd quite like to have an LLM trained on Pliny's encyclopedia, which would give a really interesting take on lots of questions. Anyone got a spare million dollars of compute time?
Here's a fun example: suppose I'm a developer with a popular software project. Maybe I can get a decent sum of money to put brand placement in my unit-tests or examples.
If such a future plays out, will LLMs find themselves in the same place that search engines in 2025 are?
https://techcrunch.com/2025/07/10/grok-4-seems-to-consult-el...
https://dmf-archive.github.io/prompt/
The AI definitely could not just read the final bill and give the correct answer. Claude/Gemini/OpenAI all failed at this.
Thats your claim, but you fail to support it.
I would argue the LLM simply does its job, no reasoning involved.
> But here’s the thing, current models “know” that Pravda is a disinformation ring, and they “know” what LLM grooming is (see below) but can’t put two and two together.
This has to stop!
We need journalists who understand the topic to write about LLM's, not magic thinkers who insist that the latest AI sales speak is grounded in truth.
I am fed up wit this crap! Seriously, snap out of it and come back to the rest of us here in reality.
There's no reasoning AI, there's no AGI.
There's nothing but salespeople straight up lying to you.
No comments yet
Try asking the major LLMs about mattresses. They're believing mattress spam sites.
https://dmf-archive.github.io/docs/posts/cognitive-debt-as-a...
2) ingest as much VC money and stolen training data as we can
3) profit
LLMs can be entertaining if their output doesn't have to make sense or contain only truth. Otherwise, their fitness for any purpose is just a huge gamble at best.
[0] x.ai
Of course they can't, no surprises here. That's just not how LLMs work.
Curious how this all ends. I'm just going to try to weather the storm in the meantime.
This also means that LLMs are inherently technologies of ideological propaganda, regurgitating the ideology they were fed with.
https://docs.google.com/document/d/1n3926pSPNwXd8j7I716CBJEz...
One man's disinformation is another woman's truth. And people tend to get very upset when you show them their truth isn't.
Every news organisation is a propaganda piece for someone. The bad ones, like the BBC, the New York Times, and Pravda make their propaganda blatantly obvious and easily falsifiable in a few years when no one cares.
The only way to deal with this is to get the propaganda from other propaganda rags with directly misaligned incentives and see which one makes more sense.
Unfortunately, LLMs are still quite bad at dealing with grounding text which contradicts itself.
If LLMs remain widely adopted, the people who control them control the narrative.
As if those in power did not have enough control over the populace already with media, ads, social media etc..
But if it returns February 20th, 1731... that... man, that sounds close? Is that right? It sounds like it _could_ be right... Isn't Presidents' Day essentially based on Washington's birthday? And _that's_ in February, right? So, yeah, February 20th, 1731. That's probably Washington's birthday.
And so the LLM becomes an arbiter of capital-T Truth and we lose our shared understanding of actual, factual data, and actual, factual history. It'll take less than a generation for the slop factories to poison the well, and while the idea is obviously that you train your models on "known good", pre-slop content, and that you weight those "facts" more heavily, a concerted effort to degrade the Truthfulness of various facts could likely be more successful than we anticipate, and more importantly: dramatically more successful than any layperson can easily understand.
We already saw that with the early Bard Google AI proto-Gemini results, where it was recommending glue as a pizza topping, _with authority_. We've been training ourselves to treat responses from computers (and specifically Google) as if they have authority, we've been eroding our own understanding and capabilities around media literacy, journalism, fact-checking, and what constitutes an actual "fact", and we've had a shared understanding that computers can _calculate_ things with accuracy and fidelity and consistency. All of that becomes confounded with an LLM that could reasonably get to a place where it reports that 2+2=5.
The worst part about the nature of this particular pathway to ruin is that the off-by-one nature of these errors are how they'll infiltrate and bury themselves into some system, insidiously, and below the surface, until days or months or years later when the error results in, I don't know, mega-doses of radiation because of a mis-coded rounding error that some agentic AI got wrong when doing a unit conversion and failed to catch it. We were already making those errors as humans, but as our dependence and faith on LLMs to be "mostly right" increases, and our willingness and motivation to check it for errors dwindles, especially when results "look" right, this will go from being a hypothetical issue to being a practical one extremely quickly and painfully, and probably faster than we can possibly defend against it.
Interesting times ahead, I suppose, in the Chinese-curse sense of the word.
The education system I grew up in was not perfect. Teachers were not experts in their field, but would state factual inaccuracies - as you say LLMs do - with authority. Libraries didn't have good books; the ones they had were too old, or too propaganda-driven, or too basic. The students were not too interested in learning, so they rote-learned, copied answers off each other and focussed on results than the learning process. If I had today's LLMs then, I'd have been a lot better off, and would've been able to learn a lot more (assuming that I went through the effort to go through all the sources the LLM cited).
The older you grow, you know that there is no arbiter of T-Truth; you can make someone/something that for yourself, but times change, "actual, factual history" could get proven incorrect, and you will need to update your knowledge stores and beliefs along with it, all the while being ready to be proved incorrect again. This has always been the case, and will continue to be, even with LLMs.
Hell, they might learn that even real life authorities may lies, cheat and not have everyone’s interest in their mind.
Hope for the best, prepare for the worst.
A liberal multicultural postmodern democracy continually acting as if immigration (both legal and illegal) and diversity are its strengths, particularly when that turns out to be factual (see: large American cities becoming influential cultural exporters and hotbeds of innovation, like New York and Silicon Valley etc) means American propaganda is only more effective when it's backed by economic might.
It also means the American propaganda is WILDLY contradictory. There's a million sources and it's a noisy burst of neon glamour. It is simply not as controlled by authority, however they may try.
You cannot liken authoritarian propaganda to postmodern multicultural propaganda. The whole reason it's postmodern is that it eschews direct control of the message, and it's a giant scrum of information. Turns out this is fertile ground, and this is also why attacks by alien propaganda have been so effective. If you can grab big chunks of the American propaganda and turn it to your enemy weapon of war and destruction of America quite directly, well then the American propaganda is not on the same destructive level as your rigidly state-controlled propaganda.
More seriously:
>Screenshot of ChatGPT 4o appearing to demonstrate knowledge of both LLM grooming and the Pravda network
> Screenshot of ChatGPT 4o continuing to cite Pravda network content despite it telling us that it wouldn’t, how “intelligent” of it
Well "appearing" is the right word because these chatbots mimic speech of a reasoning human which is ≠ to being a reasoning human! It's disappointing (though understandable) that people keep falling for the marketing terms used by LLM companies.
It is very entertaining to go and click on these authors bio and just laugh at this.
But firstly to get it out of the way.
"LLM grooming" - is completely made up nonsense to justify censoring opponent in information war on behalf of USA dying empire's.
Bad Actors here Are "The American Sunlight Project" and whatever psyop think tank foot soldier's clown show that is.
> Nina Jankowicz
> Co-Founder and Chief Executive Officer
> She advised the *Ukrainian Foreign Ministry on strategic communications* under the auspices of a Fulbright-Clinton Public Policy Fellowship in 2016-17.
Checks out.
Now let's strip down this next spy girl author.
Here is video of her from EU *DisinfoLab*
https://www.youtube.com/watch?v=kk7tBTg_Xbo
"but why is she confessing ? She is not, she is bragging"
from substack:
> Sophia Freuden is a former researcher at The American Sunlight Project, where her research coined the concept of *LLM grooming*.
linkedin:
> Washington DC-Baltimore Area
> with *Ukraine 2014 Maidan photo* in the header
https://college.lclark.edu/live/profiles/9027-sophia-freuden...
> She also spent a semester in St. Petersburg in the language intensive program there
> Sophia recently completed a Fulbright English Teaching Assistantship in Kazan, Russia, *but only after completing two separate internships in Washington, D.C.—one at the U.S. Department of State, and another at the American Academy of Diplomacy.*
https://uc.web.ox.ac.uk/people/sophia-freuden
> she completed a Fulbright English Teaching Assistantship in Kazan, Russia. Upon gaining additional professional experience in digital media and law, she enrolled at Harvard to *study cyber conflict, lawfare, and information operations between Russia and the West. Her current research explores Russian media narratives surrounding Western philanthropists and diplomats.*
eyes raised, ...
> Wherever her future career takes her, Sophia’s love of Russia will be a key part of her professional life.
Sadistic love that is.
https://www.info-res.org/app/uploads/2024/11/Life-Under-Occu...
> By Sophia Freuden
https://github.com/sophiafreuden?tab=following
follows https://github.com/Centre-for-Information-Resilience and https://github.com/thomasjjj and he follows back
What's CIR ?
https://westafricaweekly.com/leaked-document-exposes-british...
> CIR is a London-based organisation founded by Adam Rutland and Ross Burley, both longtime employees of the United Kingdom’s Foreign, Commonwealth & Development Office (FCDO). Cindy Otis, a former CIA analyst, sits on its advisory board. The FCDO and the now-defunct United States Agency for International Development (USAID) fund the organisation. Although CIR presents itself as a nonprofit focused on human rights and countering disinformation, it is linked to British intelligence agency MI6 (formerly the Secret Intelligence Service, SIS).
what's CIR do ?
https://github.com/thomasjjj
> A reconnaissance framework for researching and investigating Telegram.
> Ukrainian Air Force Attack Data Tracker
> snowball sampling to collect Telegram channels through forwards.
> detect potential incitement to genocide in YouTube video transcripts.
pretty sure to de-platform oppositional content
> Load a JSON export of a Telegram channel containing coordinates of geolocations
> A Telegram Mass Surveillance Bot in Python
is this enough evidence ?
LMFAO
Shitposting and troll farms have been manipulating social media for years already. AI automated it. Polluting the agent is just cutting out the middleman.