Dario has his name on the OpenAI claims years ago that “gpt-2 is too powerful to release”, so he’s been pulling this base-of-the-brain-stem crap for years at this point. When you unpack what they’re saying, it boils down to “The Chinese and everyone working in the open are really scary and we should stop their progress”. Remind me again whose shoulders you sit on?
Even more disgusting, he did it with an unnecessarily long essay. You know that colleague who puts everything in ChatGPT and you’re forced to read all of it just in case there’s a buried surprise?
Anthropic is doing that to government.
The strategy only makes sense - and the essay is only feasible to write - if you have access to “a country full of sycophants in a datacenter.”
I’ve had huge success with TDD and nothing more formal than that; test frameworks are great for verifying “business correctness”. But they can also verify other aspects - eg. asserting that 3 queries ran instead of 103 for a request, to prove absence of N+1 queries.
Really any technique that lets the agent create its own verification is ideal, as it makes verification scale.
You’d really enjoy the book “The Democratic Coup D’Etat.” It looks at societies that had revolutions then installed democracy, starting with Portugal and moving to the US and several others, and tries to draw a through-line and find the common elements.
Focused on democratic turnarounds so its adjacent to your curiosity - but a great, enlightening read.
> This is what happens with poor internal quality. Progress is rapid initially, but as time goes on it gets harder to add new features. Even small changes require programmers to understand large areas of code, code that's difficult to understand. When they make changes, unexpected breakages occur, leading to long test times and defects that need to be fixed.
Seems familiar!
> Concentrating on high internal quality... is a rarer case, as it requires a skilled and disciplined team to make it happen. But we do occasionally see it.
Vibecoding vs. agentic coding. Same tools used different ways. Very different outcomes.
The most powerful thing you can do in 2026: lint your architecture in a pre-commit check.
Agents & humans can’t commit, or stop work, unless they’ve passed all lint checks. Passing the linter requires complying with the architecture. Now architecture violations can’t creep in and entrench themselves - they have to be fixed upfront. Code review is free from worries about architecture too.
Start by having your agent install some import-linter rules. When you hit its limits, have your agent write a custom AST-based checker (a short python script will do) that can look for imports that violate your rules. Resist its attempts to add a `# noqa: imports` escape hatch, because future agents will gleefully abuse that and you’ll be back to human in the loop (ask me how I know!)
Philosophy is often a key stepping stone to science. Wondering what could be, is taking someone to the next step. If we ever want to do something that isn't just small increments on the past this is needed.
The article doesn’t seem to mention it, but I’d be curious to know if this or similar duplication contributes to how many organisms - including humans - display symmetry and use it as a sign of genetic fitness.
Eg. Humans. Lots kv studies show people finding symmetrical faces attractive.
Maybe symmetry is the body’s way of showing off how good it is at producing the same outcome twice - and attraction comes from the idea that that body could do it for your offspring too.
No, it does not have any connection with symmetry whatsoever.
Genome duplication has happened in many animal groups. Wherever it happens, it enables the descendants to evolve into having bodies that are bigger and more complex, because each ancestral gene is replaced by multiple copies and then each copy evolves differently, becoming able to accomplish additional functions than in the ancestors.
Without a genome duplication, accumulating a similar number of genetic innovations would require a much longer time.
Another example besides the vertebrates is that the ancestor of spiders and scorpions also passed through a genome duplication event.
In animals, the genome duplications are very rare events, because the development of animal embryos is very complicated and after a genome duplication the embryos will usually fail to develop correctly and they would die.
On the other hand, in plants genome duplications and also hybridizations when a genome is doubled by combining together 2 genomes of some closely related plants, are very frequent.
Some of the most important crops have genomes that have been multiplied, either from the same species or by hybridization, i.e. which are called tetraploid or hexaploid, to mark the fact that they are doubled or tripled in comparison with the original diploid genomes. In cultivated plants, this genome multiplication has resulted in a higher productivity in comparison with their wild ancestors.
I don't think that has much to do. Symmetry is very handy in several fronts, you can code an organism geometry more easily, that same organism has an easier time learning how to navigate environments and another organism of the same species has an easier time detecting defects on it while weighing potential mating.
> ... I feel contempt for the author, because if you use AI to write, you are a waste of biomass. Let’s not mince words here. Someone who is so eager to replace themselves, that they would have a machine write in their stead, when the machine can’t even write good yet: what do you call that, if not contemptible? It’s like making yourself into a eunuch so Claude can fuck your wife.
Unfortunately due to how tasteless this passage was, I won’t be reading this or your future writing.
Alas, for rude, tasteless behavior, such as replacing your own authentic self-expression with the mellifluous spew of verbal diarrhea that bullshit machines slather across all surfaces they touch, rude, tasteless metaphors are the only fitting ones.
The only way people will get to self-expression that's worth hearing is by working their way through the self-expression that wasn't.
In a better world that would happen in school and as a normal part of growing up, but at least in the US, that mostly doesn't happen.
If humans use LLMs to write, it's much more likely they will never uncover the unique perspectives they could share with the world.
So, for those whose parents and teachers failed them, embracing, encouraging, and engaging their clumsy, self-centered rudeness is, I think, the best path forward.
Granted, when it is clumsy, self-centered, and rude, one of the more helpful things we can do is offer the critiques that clearly and coherently point that out, as you've done here.
...though I stand by my claim that using LLMs to write is actually deeply offensive and rude, a rejection of what it means to be human, and that it should be met with strong rejections.
We also probably need to lower our standards somewhat, if not for the only reason to allow ourselves to grow.
One of the best and worst things to ever happen to self-expression was the internet. While it exposes us to so many new ways of thinking, it can also be repressive in so many ways, not to mentions limits many of us from engaging with those in the immediate area. Also it holds onto our worst outputs forever!
You might see why I really like the point about embracing imperfect communication and expression for so many reasons.
I think it's fine to strongly reject non-human communication for certain (maybe even many purposes), but there are a lot of "production" type situations (boilerplate code, CRUD systems) that don't require a human grinding away at it, and humanity benefits from using it's time more wisely. Defining what is "intellectual" work versus this "repetitive" work is kinda tough, and it is probably different for everyone. There is a certain amount of grinding necessary, and some move through it faster than others.
Unfortunately, due to how rude this comment was, I won't be reading this or your future writing.
Because obviously someone who is rude must be wrong about everything. If someone calls someone a waste of biomass, by the rule of conservation of insults that must imply they are not one. Also we must be civil and polite to people who are destroying civilization.
Tastelessness aside, it also shows that author doesn’t (or refuses to) understand why someone may decide to delegate a documentation task to a subpar agent.
Laziness.
Yes, conceptually it’s something about surrendering one’s voice and agency to a subpar machine. Or something like that. (Though that persistence-suggestive neutering metaphor is probably a unwarranted exaggeration.) In practice though it’s more like “I don’t want to write anything, but some poorly written document I’ll just proofread to be not too blatantly wrong beats having absolutely nothing. PRs welcome.”
It might be not the best decision, sure. Quite arguably, a wrong one. Still, I find it concerning that it’s sufficient for the author to dehumanize someone, even in a jest of edginess. Like wtf dude chill down, as if the world isn’t mad enough already.
But the problem is people are not just delegating formulaic procedural prose to AI. They're using AI to write entire scientific papers, so now reviewers have to use Pangram[0] to screen submissions. Literary magazines have the same problem[1]. Maybe those people should know that their behaviour is bad.
Ah, yes, although that’s a different situation from the linked post, more disrespectful (and potentially nefarious) than a sloppy readme. In those cases, most likely, more than just laziness is involved.
Arguably, someone who has chosen to replace their own human expression with machine words has already dehumanized themselves - although this is perhaps a too-literal reading the word “dehumanized”?
I very explicitly tried to separate this third-party perception and actual first-party intent. Well, my guess of it, of course - but I find it hard to believe someone decides to LARP Adeptus Mechanicus and goes along the lines of “why don’t I cede my voice to the Machine Spirit”, while “why don’t I task this tool to write notes instead of doing it myself” is a lot more plausible.
Choices and outcomes are two very different things.
This is a very very intense claim, and if true, would represent a monumental institutional failure across hundreds or even thousands of disparate organizations.
When DNA matching was introduced, we discovered that at minimum 10% of people on death row were innocent. Death row cases are among the most litigated and examined cases. So, 10% is a reasonable floor, and we're already in double digits.
That stat is off by a couple orders of magnitude. The total number of death penalty convictions overturned by DNA evidence is 29 (as of 2025). There are a couple thousand death row inmates right now, and the denominator here is all the people who were on death row in the last 20+ years. That's a rate of significantly <1%.
"we discovered that at minimum 10% of people on death row were innocent"
How did we do that? I never heard this: certainly 10% of people on death row weren't exonerated by DNA? This is some kind of shaky extrapolation I assume?
When they choose the "DNA loci" to do SRT "matching" in the first place they convinced themselves it was a unique fingerprint and there never would be any duplicates in the database.
It only took a few years.
They've since changed and expanded the standard "DNA loci" to compensate.
A few years ago, one of my coworkers was arrested for a domestic violence complaint. Looking into his case, I found an extremely specific lurid description of the allegations -- and then I found the same lurid description copy/pasted to every other person recently arrested for the same crime. I'm probably getting the specific terms wrong, but I did click through to see it on a government website, because my first suspicion was the aggregator, but no, the police just had a boilerplate story full of specifics which could not possibly apply to each and every person they carelessly slapped it onto. This absolutely blew my mind at the time, but it fits with smaller subsequent observations. In any case: a double digit percentage of institutional failure does not upset my priors about how carefully the police operate.
Shouldn't it be the exact opposite here ? The burden of proof is the other way around.
The big claim is here: the state has grandiose claims that the overwhelming majority is fair, but there is no proof of it.
Therefore the state should prove that more than 90% of the cases are legitimate, fair, not coerced, and not motivated by the pressure to interrupt the proceedings.
97% of people choose plea deals or out-of-court settlement, it is a huge amount.
It means that in real practice, not imaginary internet, people who face court consider that justice is a big machine that can crush you no matter if you are innocent or not.
In the best case you are acquitted at the end, but you are guaranteed to bear the financial burden, fear and stress as a punishment.
Being held in jail before trial is a very convincing reason to plea deal too.
It's a system engineered to make pleading the only reasonable option, no matter if you did anything or not.
That is true--the checks and balances the founding fathers fought so hard for were thrown out the window with overlegislation and expansion of prosecutorial discretion in 20th century. To make a convincing argument that "double digits" of cases involve fabricated evidence, you still need to explain why prosecutors would engage in fraud at this massive scale. Just laziness? Collecting scalps? The incentives run that way in some limited cases, e.g., prosecutor up for election, post-reconstruction south. But you need some explanation there.
Yeah, again, there are some incentives to fabricate evidence like career advancement. Now why should those, on a mass scale, outweigh disincentives like getting caught in an adversarial process and (presumably) some qualm about regularly convicting innocents and regularly letting guilty parties run free in communities. Easy to argue in particular cases but I haven't heard the basis for a trend.
What adversarial process? If the prosecutor loses the case, the defendant doesn't go to jail but still receives a very big punishment and the prosecutor loses nothing. And prosecutors never prosecute themselves for false prosecution.
This sounds like you’re imagining how prosecutors as a group sort of feel about things, generally, and that this notion you’ve thought of outweighs the demonstrable real-world system where prosecutors are awarded for convictions, full stop.
A burden of proof is associated with an individual claim. There’s no “burden of proof in the other direction” - what you’ve actually done is created a second burden of proof and also - worse - attempted to distract from the original point.
It is disingenuous to weasel out of proving one claim by making another, or saying “look over here”
Also, outrageous claims in opposite directions can both be bullshit.
That is like claiming that double digit percentage of software bugs and vulnerabilities were intentionally put there by malicious software engineers. Its outrageous to claim its that high.
Even single digit percent is hard to believe, but its possible, but double digits you are talking China or Russia levels of state corruption and even there I doubt its that high.
~~Please point to the place where I said your claim was outrageous.~~
Edit; upon closer examination. I did imply in my last paragraph that your claim was outrageous. Bit of a gaffe considering I’m the agitator here. My apologies.
An important thing you should recognize: the judicial system is painfully nontransparent in such a way that even figuring this sort of thing takes an extensive amount of time and is often even impossible. I've personally gone down a similar route (did some journalism for a bit) by trying to understand how shotspotter is used in prosecution, many of which resulted in false arrests and many, many years of life lost across all the people arrested falsely from it.
If you would like to begin trying to answer these, I recommend starting with submitting some FOIAs. Considering your stance seems to be that you won't believe what others are telling you -- I promise you that you'll be surprised.
If you believe parallel construction should be illegal (it sure seems like it is unconstitutional to me), then 100% of prosecutions that rely on it are unjust. I don't think anyone truly knows how common it is, though, and that's by design. Double-digits wouldn't shock me at all.
It's more monolithic than you would think due to shared culture over the internet. There's a whole narrative about sheepdogs (them), sheep (us) and wolves (the bad guys).
The main obstacle is the denominator. How many conversations were there (before dna testing) where testable tissue was preserved? You would want to test some random sample of them against the person convicted.
We have the highest proportion of imprisoned citizens in the world.
This is done because there's an exception in our constitution for slavery "as punishment for a crime" and well all know that capitalism loves slave labor.
Weapons.
Know what’s not? Words. Text. Tool calls.
Dario has his name on the OpenAI claims years ago that “gpt-2 is too powerful to release”, so he’s been pulling this base-of-the-brain-stem crap for years at this point. When you unpack what they’re saying, it boils down to “The Chinese and everyone working in the open are really scary and we should stop their progress”. Remind me again whose shoulders you sit on?
Even more disgusting, he did it with an unnecessarily long essay. You know that colleague who puts everything in ChatGPT and you’re forced to read all of it just in case there’s a buried surprise?
Anthropic is doing that to government.
The strategy only makes sense - and the essay is only feasible to write - if you have access to “a country full of sycophants in a datacenter.”
reply