> A team I talked to recently wired up an agent to do something simple: pull a metrics API every morning, reshape the JSON, and drop the result into a table. Clean idea. It worked on day one.
So a team of people with access to "intelligent" help did not immediately figure out that they could just have the model write a small script that does the job perfectly every time? That's "call a doctor" level of stupid that should be impossible in vivo. I'll believe a lot, but not this. The rest of the article is not much better - only an overexcitable LLM could believe anything there constitutes a deep insight worth writing down.
Do we need a new category for this sort of made-up, probably LLM-generated, garbage? Is this AI brainrot blogspam engineered to be upvoted by people who don't read the contents?
How do you not die of shame publishing something like this on your company blog?
Yes, from a programmer's perspective, this is insane. But the team may have been made up of people without coding experience who maybe didn't even know what a script was. I imagine there are a lot of people right now who "know AI" and not much else, for whom this would be a genuinely new insight...
Note that even before AI, there were a lot of manual data munging jobs that could famously be replaced by "a very small shell script".
I worry, this might become worse if we see the emergence of "100% nontechnical tech startups" where no one in the entire org knows how to code - because why would you, we have AI for that...
Not to mention federal politicians doing their best to kill the German economy and motivation to work at all, it's quite hard to imagine why any foreign engineer would chose to move to Saxony of all places. Have you been to the countryside in the Bundesland? It's bad. Neonazi and racism all over the place. Literally visible to the casual observer, you don't even have to talk to anyone. T-shirts, stickers, graffiti, stupid notices, ... people there don't hide it.
Dresden and especially Leipzig are quite nice though - lots of culture, music and art. Also close to Berlin. It's not exactly like you would have to travel through the countryside by carriage to get anywhere. I agree that the political and cultural situation outside the cities there is sad though.
I am frequently in Leipzig, it's indeed a very nice city. But it's also immediately obvious people with a (visible) migration background are a smaller minority than elsewhere (which is backed by stats). So for a lived experience, I would strongly suggest talking to those minorities. The reports I heard speak of a lot of casual racism.
While you don't have to travel by carriage, you may even use the airport, avoiding the Umland is still an alienating and saddening state of affairs you can hardly call acceptable. For example, Saxony has absolutely beautiful nature along the Elbsandsteingebirge/Sächsische Schweiz, but those areas are also notoriously hot spots of the violent far right (e.g. https://de.wikipedia.org/wiki/Skinheads_S%C3%A4chsische_Schw...). For me it felt moderately dangerous, as a white person. As an Indian, are you gonna join your colleagues on "workation" there, are you gonna bring up your worries? It's all fine, bro, racism in Saxony is all Wessi propaganda, bro. Just avoid leaving the safe zones.
I don't disagree and I don't think think that it's a good state of affairs. Whether acceptable is a legitimate category here i have some doubts about. There is definetly people that do not accept the status quo there (for good reasons) and try to become acively engaged in doing something about it. To state that it is not acceptable kinda evokes the image of some higher force that intervenes and changes the situation; idk seems weird. Would be nice if the xenophopic and far right resentfull mindset would stop holding it's grasp over such huge portions of the population. (Globally, too)
> To state that it is not acceptable kinda evokes the image of some higher force that intervenes and changes the situation
Yes, this higher force is morality and conscience. Going back to the original argument, Saxony isn't sponsoring these developments all by itself. It's a larger collective's effort. As a tax paying citizen, I am entitled to judge the situation and at the very least demand conditions for success, or ask for reallocation of resources to more fertile grounds.
As can be seen in this thread, the people living in these regions feel zero accountability or problem awareness. Unsurprisingly you even see the common "How dare you call us Nazis, we're just against multiculturalism and anything woke", ironically proving the point: Foreign engineers are not welcome. The foreign workers are not all going to be drawn from white Norway and Poland (what an idiotic assumption), but India and Asia etc.. They will make the landscape "multicultural". Not too mention the higher percentage of queer and other "woke" minorities in the tech sector, which are also "just" not welcome. After all, being not accepted at home may be another reason to seek shelter elsewhere.
Saxony is not full of Nazis, just people who kinda prefer the authoritarian ethnostate. Wessis need to be more accepting of the fact Pride parades are cancelled. And did you know, statistically you are far more likely to die by other means, than a hate crime in Saxony!
Quite frankly, you can't get to them (reactionary idiots) through rational discussion. It's been tried over and over again. We don't know how to do it, but talking to them isn't it. Meeting them with brainrot TikTok, seems to be somewhat effective (see Linke campaign), but in the end the simplistic answers win social media and the right got a natural edge there. They are living in informational bubbles and, in the East, due to liberal people leaving/fleeing (no judgement), these bubbles become lived realities. Worth noting: The high AfD turnouts are likely due to demographics, population density is really, really low across the east. Self-fulfilling prophecy. Of course there is historic baggage, too, but it's not just the old fucks, so at last let's start putting accountability where it belongs. The people there need to change, if they want the economic prosperity they feel entitled to. Rainald Grebe's "Brandenburg" is 20 years old: https://www.youtube.com/watch?v=uellmynA34U
And yes, some of this entitlement and mental retardation is also rapidly growing in the West. All of Germany is very sick, geriatric, mismanaged and racing towards precarious inequality. Whatever the East is cooking is not the solution, so let's not call it a virtue.
> A pardon absolves one of the sin as if it didn't happen, legally.
This is incorrect. A pardon is not an expungement. The conviction remains a usable historical fact and could still be referenced in later legal procedings.
Exact ramifications vary between innocence-based pardons, rehabilitiation-based pardons, and pure discretionary clemency.
In fact, part of accepting a pardon is accepting guilt. That can particularly be consequential if there is a civil case associated with the criminal charges. For example, if I'm charged with drunk driving and I run into someone's house, by accepting a pardon I have to admit that I'm guilty of drunk driving which the home owner can then use in their civil suit to extra money for the damage I caused.
This is part of the reason why people will sometimes not accept a pardon.
That link breaks for me, but I suspect I know what you are referring to. That talk from the various courts seems mostly like rhetoric more than an establishment of legal precedent. It is all implied meaning, since indeed you do not need to affirmatively proclaim your own guilt in order to accept a pardon. You can just accept delivery and be done with it. Whether someone else imputes guilt from that is [mostly] their problem.
Like if a pardon is issued before trial, under normal circumstances the prosecutor will drop charges and the pardonee does not need to accept it. Further, a prosecutor won't go after charges when someone is pardoned.
These are the cases where a pardon wouldn't imply guilt.
But generally speaking, pardons happen after a conviction and not before. Accepting a pardon ends appeals.
I think you're missing the point. If you are a felon, there is baggage that comes with it which varies depending on the state. Some felons can no longer vote or legally own a firearm. Some felons find it hard to find a place to rent. Unless of course, you've been pardoned.
I also even stipulated that people could not be made to forget about it. Yet, you then reiterate that after telling me I was incorrect.
I remember in the 1990s watching a bit of British TV, where an ad asked, "What do women in Los Angeles know about beauty that British women don't?" (obviously selling some "secret" "American" product).
In LA those are either French or Italian women, selfishly hoarding all the good skin cream. I then wondered who the French and Italian women think are the beauty-secret-holders...
There's a post every other month where some dude who put nonsense information online celebrates because it actually ended up in some frontier models weights.
If it's easy enough that some randos can do it for fun, what do you think happens when there's commercial interest behind it?
Obviously companies are going try nudging AI towards recommending whatever they're selling. It's a logical extension of SEO - and that's a 100 billion USD industry.
Additionally, if I believed myself to be in some sort of spending - err - AI race, I'd try to poison the data sets of my competitors by putting crap out there for others to ingest.
What does it mean, Is it like when somebody used some coding agent to develop a feature and later input prompts and a resulting PR can be used for training by a presumption that final PR was a correct implementation of a prompt?
Yea it’s rejection sampling, so you have an agent, you take a verifiable problem (people use lots of different verification signals but say unit tests etc) and have the agent attempt it K times. You accept the trajectories (all context, tool use etc, the entire log) that are positively verified and use these as training examples.
The trick is to find the examples that are just in between too difficult and too easy for the existing agent, these have the strongest training signals
The question is not whether it has happened or will continue to happen. Of course it will always be a problem to some extent.
Your original claim is that this will be enough of a problem to prevent models from improving in expert level knowledge. I completely disagree with this premise.
If the models fail to improve, it will likely be due to limitations in the transformer architecture rather than poisoned training data.
And even then, I doubt that the transformer is the best architecture we will ever come up with.
Clearly it doesn’t learn or think like a human does, since humans don’t need many gigabytes of text samples to learn to talk, so there is some room for improvement.
That no one has actually solved the underlying problem at all, and the generation of the example LLM has no bearing on the nature of the fundamental problem.
You are totally misunderstanding my argument then. As I said, garbage in garbage out. Your article is just an example of that. It’s pretty obvious that if you train an LLM on bad data, you will get bad output.
What I’m saying is that the AI labs are handling this not by fixing the “garbage out” part, but by minimizing the “garbage in” part.
The fact that all you could come up with was research (not an actual example of poisoning a real training set) from 2025 kind of proves that this isn’t some kind of widespread, unsolvable problem like you seem to be claiming.
I literally just grabbed a random link. I’ve seen dozens of real life examples of poisoning.
The poisoning issue makes it so that no one can use the internet for training anymore, because more and more internet content is poisoned as a side effect - or poisoned intentionally. And .001% of poisoned data is enough to screw things up if included in the training data.
It’s also one reason why Google search results have been getting so much worse - it’s hard to not find a SEO page with subtly (or not so subtly) wrong AI slop on almost every topic you can imagine. Most folks won’t recognize it, but that’s what is going on if you know what to look for.
One other way of putting it is the ouroborus problem - more and more internet content is AI generated, because of people trying to game the system, and they are making it is indistinguishable from real content as possible to get by the AI detection algorithms.
Anyone trying to train on it just ends up eating the shit from another LLM, which poisons it.
Another name for it is ‘model collapse’, which also doesn’t have a known solution yet.
You’ve seen actual model poisoning? Or have you seen a model return the wrong answer due to what it saw in a search result? Or were they hallucinations perhaps? How do you know it’s due to poisoned training data?
And do you even realize how much data 0.001% of the training data for a frontier models is? They’re trained on 10s of trillions of tokens, meaning you’d need hundreds of millions of tokens of poisoned data.
Some of these problems you mention could become real barriers to models improvements, though there are plenty of countermeasures, such as by focusing on high quality data sources like I mentioned before.
We’ve already probably gotten as much as we’re ever going to get from simply scraping more and more unstructured text from the web as a way to improve model performance.
The type of training being done now is around tool use and solving specific types of problems better, which is the type of training data you simply don’t find lying around on the web.
You’re expecting me to know your job? Give me a break.
I’m wondering the same thing. You keep talking of some grand poisoning problem but can’t point to any specific public information except an article saying that it’s possible. As if that was ever in doubt.
Most Europeans do not think about some ill-defined "communist threat" at all. The majority of negative opinions about China stem from economic worries.
Also FYI: the CCP has officially adopted the designation "socialist democracy" for themselves*, so I don't think you're going to bother them much by using that term. You'll have to get more specific about what you think their "democracy" should look like for them to start giving you the side-eye.
* Many places that are not really recognizable as democracies from a western POV do this. People, we have democracy at home!
You are off the rails. Ask China why they don’t like Taiwan’s political system, or better why they don’t adopt it to make the “unification” more attainable. The Communist party is everything in China.
The reason most people reach for Kubernetes is because it's cool. The entire infra the vast majority of Kubernetes users have could run on a single bare metal machine with a second one for redundancy.
To be fair: using Kubernetes anyways builds the skill just in case you become one of the 0.1% who actually need it down the line.
You can hire an Azure or Google Kubernetes devops guy and he will be equally comfortable on your AWS EKS kubernetes cluster. And when he leaves, you don't have a six week onboarding process with the new guy to learn all the ins and outs of your totally bespoke, non-standard container orchestration system that was cobbled together by two devs with no operations experience.
K3S takes about 5 minutes to setup the first time and you instantly have an entire universe of standardized operational tooling. I wouldn't touch docker compose with a 20 foot pole for production work.
As soon as you work in a team, it’s irrelevant whether the project actually needs it. There will be someone who convinces stakeholders that it is necessary and then you just have to fall in line and learn the skills knowing that it is most likely one of the 99.9% of projects where it is just overkill.
Until your project has some success, and it turns out all those "complex" features actually turn out to be extremely useful.
Which is exactly what is happening with us, too bad we didn't choose K8S from the get-go and stuck with a "simpler" tool (gaining very little in the process).
Your rebuttal to the parent claiming that almost nobody needs k8s is to bring up a workload almost nobody runs? It seems to me like your argument reinforces the parent's, not undermines it.
This estimate includes things like a car, a partially paid off house and other assets.
Most of that wealth cannot easily be converted to cash which you'd need to start a company.
Also that's median. Germany is a country with a median age of 45. So yeah, someone who likely worked for 20+ years will likely have saved around 100k, I don't think you realize how that's an argument in favor of what I just stated...
So a team of people with access to "intelligent" help did not immediately figure out that they could just have the model write a small script that does the job perfectly every time? That's "call a doctor" level of stupid that should be impossible in vivo. I'll believe a lot, but not this. The rest of the article is not much better - only an overexcitable LLM could believe anything there constitutes a deep insight worth writing down.
Do we need a new category for this sort of made-up, probably LLM-generated, garbage? Is this AI brainrot blogspam engineered to be upvoted by people who don't read the contents?
How do you not die of shame publishing something like this on your company blog?
reply