Why do you think that you could trust what you read before? Is it now harder for you to distinguish false information, and if so, why?
While the professional looking text could have been already wrong, the likelihood was smaller, since you usually needed to know something at least in order to write convincing text.
For example, when applying to a job, your cover letter used to count as proof of work. The contents are less important than the fact that you put some amount of effort in it, enough to prove that you care about this specific vacancy. Now this basic assumption has evaporated, and job searching has become a meaningless two-way spam war, where having your AI-generated application selected from hundreds or thousands of other AI-generated applications is little more than a lottery.
When it becomes obvious, though, is when people let the LLM do the writing for them. The job search bit is definitely rough. Referrals, references, and actual accomplishments may become even more important.
The noise ground has been elevated 100x by LLMs. It was already bad before but it's accelerated the trend.
So, yes, we should have never been trusting anything online but before LLMs we could rely on our brains to quickly identify the bad. Nowadays, it's exhausting. Maybe we need a LLM trained on spotting LLMs.
This month, I, with decades of experience, used Claude Dev as an experiment to create a small automation tool. After countless manual fixes, it finally worked and I was happy. Until I gave thr whole thing a decent look again and realized what a piece of garbage I had created. It's exhausting to be on the lookout for these situations. I prefer to think things through myself, it's a more rewarding experience with better end results anyway.
I've been using Zed recently with its LLM integration so assist me in my development and its been absolutely wonderful, but one must control tightly what to present to the model and what to ask for and how.
"Human-curated encycolpedias are a great onramp to filling in knowledge gaps", that I can go with.
Maybe I should have worded it better as a "beginner" or "intermediate" knowledge onramp and/or filler. For example, I have asked it on occasion to translate into traditional Mandarin in parallel for every English response. It helps tremendously in trying to rebuild that bridge that may have been burned long ago.
1. I use dictation software to get my thoughts out as a stream of consciousness. 2. Then, I have ChatGPT or Claude refine it into something coherent based on a prompt of what I'm aiming for. 3. Finally, I review the result and make edits where needed to ensure it matches what I want.
This method has easily boosted my output by 10x, and I'd argue the quality is even better than before. As a non-native English speaker, this approach helps a lot with clarity and fluency. I'm not a great writer to begin with, so the improvement is noticeable. At the end of the day, I’m just a developer—what can I say?
I know not to trust it to be as precise as good research papers need to be, so I don't take its output, it usually helps me reorder points or use different transitions which make the material much more enjoyable to read. I also find it useful for helping to come up with an opening sentence from which to start writing a section.
I don't criticise you for it, because that strategy is both rational and popular. But you never checked the accuracy of your information before so you have no way of telling if it has gotten more or less accurate with the advent of AI. You were testing for whether someone of high social intelligence wanted you to believe what they said rather than if what they said was true.
Now we'll need to be fully diligent, which means more work, and also there'll be way more things to review.
The best we can hope for is prioritizing which things are worth checking. But even that gets harder because you go looking for sources and now those are increasingly likely to be LLM spam.
Unfortunately it's not clear how this could be adapted to the internet and international commerce without harming some of the open-ness aspects we'd like to keep.
AI does indeed threaten people's ability to identify whether they are reading work by a high status human and what the group consensus is - and that is a real problem for most people. But it has no bearing on how correct information was in the past vs will be in the future. Groups are smart but they get a lot of stuff wrong in strategic ways (it is almost a truism that no group ever identifies itself or its pursuit of its own interests as the problem).
Plenty of people care about the truth in order to get advantages over the ignorant. Beliefs aren't just about fitting in a group, they are about getting advantages and making your life better, if you know the truth you can make much better decisions than those who are ignorant.
Similarly plenty of people try to hide the truth in order to keep people ignorant so they can be exploited.
There are some fallacious hidden assumptions there. One is that "knowing the truth" equates to better life outcomes. I'd argue that history shows more often than not that what one knows to be true best align with prevailing consensus if comfort, prosperity and peace is one's goal, even if that consensus is flat out wrong. The list is long of lone geniuses who challenged the consensus and suffered. Galileo, Turing, Einstein, Mendel, van Gogh, Darwin, Lovelace, Boltzmann, Gödel, Faraday, Kant, Poe, Thoreau, Bohr, Tesla, Kepler, Copernicus, et. al. all suffered isolation and marginalization of some degree during their lifetimes, some unrecognized until after their death, many living in poverty, many actively tormented. I can't see how Turing, for instance, had a better life than the ignorant who persecuted him despite his excellent grasp of truth.
When dealing with almost everything you do day by day, you have to rely on the credibility of the source of the information you have. Otherwise how could you know that the can of tuna you're going to eat is actually tuna and not some venomous fish? How do you know that you should do what your doctor told you? Etc. etc.
But isn't your third sentence True?
All of that is gone now. You have LLMs spitting their excrement directly onto the web without so much as a human giving it a once-over.
They didn't say that and that's not a fair or warranted extrapolation.
They're talking about a heuristic that we all use, as a shorthand proxy that doesn't replace but can help steer the initial navigation in the selection of reliable sources, which can be complemented with fact checking (see the steelmanning I did there?). I don't think someone using that heuristic can be interpreted as tantamount to completely ignoring facts, which is a ridiculous extrapolation.
I also think is misrepresents the lay of the land, which is that in the universe of nonfiction writing, I don't think that there's a fire hose of facts and falsehoods indistinguishable in tone. I think there's in fact a reasonably high correlation between the discernible tone of impersonal professional and credible information, which, again (since this seems to be a difficult sticking point) doesn't mean that the tone substitutes for the facts which still need to be verified.
The idea that information and misinformation are tonally indistinguishable is, in my experience, only something believed by post-truth "do you own research" people who think there are equally valid facts in all directions.
There's not, for instance, a Science Daily of equally sciency sounding misinformation. There's not a second different IPCC that publishes a report with thousands of citations which are all wrong, etc. Misinformation is out there but it's not symmetrical, and understanding that it's not symmetrical is an important aspect of information literacy.
This is important because it goes to their point, which is that something has changed, in the advent of LLMS. That symmetry may be coming, and it's precisely the fact that it wasn't there before that is pivotal.
I suppose a related problem is that we can't know if the human who posted the article, actually agrees with it themselves.
(Or if they clicked "Generate" and don't actually care, or even have different opinions)
AIs are good at writing professional looking text because it's a low bar to clear. It doesn't require much intelligence or expertise.
AIs are getting good at precisely imitating your voice with a single sample as reference, or generating original music, or creating video with all sorts of impossible physics and special effects. By your rationale, nothing “requires much intelligence or expertise”, which is patently false (even for text writing)
https://en.wikipedia.org/wiki/Michael_Crichton#Gell-Mann_amn...
They give a superficially smart look, but really they regurgitate without deep understanding.
That is pretty much true also for other media, such as audio and video. Before digital stuff become mainstream pics are developed in the darkroom, and film are actually cut with scissors. A lot of effort are put into producing the final product. AI has really commoditized for many brain related tasks. We must realize the fragile nature of digital tech and still learn how to do these by ourselves.
...or...the likelihood of text being really wrong pre-LLMs was worse because you needed to be a well-capitalized player to pay your thoughts into public discourse. Just look at our global conflicts and you see how much they are driven by well-planned lobbying, PR, and...money. That is not new.
How did you know this unless you also had the same or more knowledge than the author?
It would seem to me we are as clueless now as before about how to judge how skilled a writer is without requiring to already posses that very skill ourselves.
Reading was a form of connecting with someone. Their opinions are bound to be flawed, everyone's are - but they're still the thoughts and words of a person.
This is no longer the case. Thus, the human factor is gone and this reduces the experience to some of us, me included.
They're not connecting with the relative correctness of each note, but feeling a human, creative connection with an artist expressing herself.
If you wanna advocate for unplugged music being more gratifying, I don't disagree, but acting like the autotune is what people are getting out of Taylor Swift songs is goofy.
I think that says more about media-technology, corporate ecosystems, and overall population-growth than about music itself.
What I think OP was saying and I agree with is that connection, that knowing no matter what was said or how flawed or what motive someone had I trusted there was a human producing the words. I could guess and reasons the other factors away. Now I don’t always know if that is the case.
If you’ve ever played a multiplayer game, most of the enjoyable experience for me is playing other humans. We’ve had good game AIs in many domains for years, sometimes difficult to distinguish from humans, but I always lost interest if I didn’t know I was in fact playing and connecting with another human. If it’s just some automated system I could do that any hour of the day as much as I want but it lacked the human connection element, the flaws, the emotion, the connection. If you can reproduce that then maybe it would be enjoyable but that sort of substance has meaning to many.
It’s interesting to see a calculator quickly spit out correct complex arithmetic but when you see a human do it, it’s more impressive or at least interesting, because you know the natural capability is lower and that they’re flawed just like you are.
I like to think of these ambiguities of "trust" as something like:
1. Trusting their identity
2. Trusting their intentions
3. Trusting their judgement about what to do
4. Trusting their competence to execute the task
Nor is trust meant to mean something is absolute and unquestionable. I may trust someone, but with enough evidence I can withdraw trust.
Nope. A lot of people just wrote stuff. There were always plenty of word salad blogs (and arguably entire philosophy journals) out there.
Imagine that AI tools sway generated sentences to be slightly close, in summarisation space, to the phrase "eat dirt" or anything. What would happen?
AI otoh can produce texts that are quite accurate globally with some totally random hallucinations here and there. It makes it quite harder to identify
I also did, but LLM increased the volume of content, which forces my brain first try to identify if content is generated by LLMs, which is consuming a lot of energy and makes brain even less focused, because now it's primary goal is skimming quickly to identify, instead of absorbing first and then analyzing info
You already know some of the more trustworthy sources of information, you don't need to read a random blog which will require a lot more effort to verify.
Even here on hackernews, I ignore like 90% of the spam people post. A lot of posts here are extremely low effort blogs adding zero value to anything, and I don't even want to think whether someone wasted their own time writing that or used some LLM, it's worthless in both cases.
However with AI anyone can generate absurd torrential flows of bullshit at a rate where, with your finite human time and energy, the only winning move is to reject out of hand any piece of media that you can sniff out as AI. It's a solution that's imperfect, but workable, when you're swimming through a sea of slop.
We are all becoming QA of this super automated world.
A human communicator is, in a sense, testifying when communicating. Humans have skin in the social game.
We try to educate people, we do want people to be well-informed and to think critically about what they read and hear. In the marketplace of information, we tend very strongly to trust non-delusional, non-hallucinating members of society. Human society is a social-confidence network.
In social media, where there is a cloak of anonymity (or obscurity), people may behave very badly. But they are usually full of excuses when the cloak is torn away; they are usually remarkably contrite before a judge.
A human communicator can face social, legal, and economic consequences for false testimony. Humans in a corporation, and the corporation itself, may be held accountable. They may allocate large sums of money to their defence, but reputation has value and their defence is not without social cost and monetary cost.
It is literally less effort at every scale to consult a trusted and trustworthy source of information.
It is literally more effort at every scale to feed oneself untrustworthy communication.
On second thought, perhaps AI/LLM generated content is better illustrated with it being like eating the regurgitated sludge called cud. Nothing new, but it fills the gut.
Do you think it would be possible before internet, before AI?
Bad actors, poorly written/sourced information, sensationalism etc have always existed. It is nothing new. What is new is the scale, speed and cost of making and spreading poor quality stuff now.
All one needs today is a laptop and an internet connection and a few hours, they can wreak havoc. In the past, you'd need TV or newspapers to spread bad (and good) stuff - they were expensive, time consuming to produce and had limited reach.
This can only be done with a sentiment that was, at least partially, already there. And may very well happen naturally eventually
It takes a bit more than that.
If you can place a rumor or lie in front of the right person/people to amplify, it will be amplified. It will spread like wildfire, and by the time it is fact checked, it will have done at least some damage.
Did the person who posted do the manipulation, or did JD Vance and Donald Trump do it?
You might question new information, but you certainly do not actually verify it. So all you can hope to do is sense-checking - if something doesn't sound plausible, you assume it isn't true.
This depends on having two things: having trustworthy sources at all, and being able to relatively easily distinguish between junk info and real thorough research. AI is a very easy way for previously-trustworthy sources to sneak in utter disinformation without necessarily changing tone much. That makes it much easier for the info to sneak past your senses than previously.
Now it's mostly zero effort LLM-generated SEO spam, and the low-wage workers lost their jobs.
If one spends a lot of years reading a lot of stuff, they come to this conclusion, that most of it cannot be trusted. But it takes lots of years and lots of material to see it.
If they don't, they don't.