Actually "non-stabilizerness" does describe it better than "magic".
> In quantum information theory, magic is a property that quantifies the computational resources needed to describe quantum states beyond stabilizer states.
> In 2024–2025, quantum magic was detected in top quark pairs produced at the Large Hadron Collider; it is the first observation of this property in fundamental particle collisions.
I looked up the original paper. It's an interesting read and foreshadows a lot of the current hot arguments around LLMs, but I'm not sure it's aged especially well:
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
However, from the perspective of work on language technology, it is far from clear that all of the effort being put into using large LMs to ‘beat’ tasks designed to test natural language understanding, and all of the effort to create new such tasks, once the existing ones have been bulldozed by the LMs, brings us any closer to long-term goals of general language understanding systems. If a large LM, endowed with hundreds of billions of parameters and trained on a very large dataset, can manipulate linguistic form well enough to cheat its way through tests meant to require language understanding, have we learned anything of value about how to build machine language understanding or have we been led down the garden path?
...
Contrary to how it may seem when we observe its output, an LM is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot.
...
Finally, we would like to consider use cases of large LMs that have specifically served marginalized populations. If, as we advocate, the field backs off from the path of ever larger LMs, are we thus sacrificing benefits that would accrue to these populations?
Especially in a world where a there's myriad open Chinese LLMs, it's not clear what policy changes are being recommended today. Gebru's paper explicitly advocates backing off from developing larger LMs than existed at the time, 6 years ago. Do those celebrating the paper continue to advocate that LLMs be scaled back to GPT2 level, for safety?
Opening an issue consisting only of some twitter clone screenshot with some "literally who" who found a bug called "Please Do Not Vibe Fuck Up This Software" ain't it. That's not a way to tell a maintainer that you disagree with the direction they're taking. This issue is entirely useless. A "fucked up vibe coded" bug report would have been better.
This nailed it. None of the bug reports even attempt to document the claimed "--compare-dest=" regression. I did ctrl-f and I didn't even see anyone mention "compare-dest" again? The people posting worthless AI rage comments could have asked Opus 4.8 to spin up rysnc 3.4.3 vs. 3.4.1, thoroughly document the regression and git bisect the commit that broke it and filed a 1000x more professional and useful bug report.
If you want society to value your human work more than AI work, try to avoid acting like a uniquely human bozo.
They're not mad about the bug, they're mad about the cause of the bug. This is like city council starts flinging stones with a trebuchet into streets, and you're expected to calmly file pothole reports instead of complaining about the trebuchet.
Bugs exist in human code too. The AI derangement crowd pounces on any small bug as evidence that AI is a trebuchet, and thinks that if only we didn't use AI there would never be any bugs (like five years ago when all software was perfect, was not being enshittified, and had 0 bugs).
The bugs were introduced because rsync had security issues(i.e. bugs) in it, presumably written into the code by humans.
It's really baffling to see so many people in this thread maintain the position that somehow software was clean and pristine until AI touched it with its evil.
Please try to at least put some sort of constructive argument forward, for example - I don't like AI because it might introduce more bugs than a careful human reviewer. Then we could discuss why a single maintainer is responsible for rsync and how they should handle the pressure of keeping it up to date - should they just stop making further changes, should they look for tools that might help them?
(By the way, if your position is that rsync was perfect before AI got its hands on it, you have a clear solution to all your problems - simply do not update to any newer versions)
Either way, move away from this absolutist nonsense that has no bearing to reality.
> It's really baffling to see so many people in this thread maintain the position that somehow software was clean and pristine until AI touched it with its evil.
Nobody maintains this position. Again, it's a strawman you made up, because it's easier to dismiss such "absolutist nonsense" than it is to just admit that these specific bugs were introduced as a direct result of careless AI usage.
If the developer is overwhelmed by the maintenance burden (they aren't, judging by how many AI commits they've been making to a large number of repositories), then that's an entirely different problem that deserves a good faith discussion, but delegating the work to AI is not the correct solution.
> By the way, if your position is that rsync was perfect before AI got its hands on it
Again, strawman, nobody said this either. In fact, quite the opposite - we want rsync to continue to be maintained by a human. If the current developer isn't interested in or capable of maintaining the project anymore, they should just say so instead of quietly letting AI take over, because then the likelihood of someone else stepping up to contribute would be much higher.
> I always imagined some twisted diminutive demonic swarm of insects
Behavioral ecologist Stephen Simpson has proposed the cannibalistic forced march hypothesis[36], that is, the forward motion of a locust swarm is essentially sustained by each individual’s imperative to avoid being eaten by the locust behind it: 1) Align their body axis with neighbors (parallel) to minimize the chances of a side-on attack and present their narrowest possible profile to the individual behind. 2) March forward to bite and feed on the abdomen of the locust immediately ahead.
A billion crazed insects marching through eating all your crops while cannibalizing each other does seem relatively twisted and demonic.
Also, there’s the stimulus causing nymphs to not be solitarious slow green grass-nibbles, but instead transform into armored, black and yellow, upright marching machines that mature into stronger-winged battle bugs. With soft tasty abdomens exposed to the soldier behind them…
Woah up until now I was thinking locusts were a different but related species, but indeed they are the exact same species as grasshopper, just with a different end-product of metamorphosis. And the trigger is just contact with other grasshopper/locusts.
Opus 4.7+ Max is a 10x engineer who wants to be left alone to work. When you talk to him, he infodumps on you to get you (his pointy haired idiot Dilbert boss) to go away.
I do know of moderate-size companies deploying OSS LLMs on their own GPU clusters, for ownership/security/maybe cost reasons. I'm somewhat surprised F500 companies are apparently just handing over all their data to the model providers.
Could be fantastic for small shops while it lasts. The big guys have to pay 10x for precious tokens.
To me, this kind of talk exhibits the very cultish and con side of the whole genAI train ... Generally, and more so with paid products, one should expect to get something that is ready to be used
Right like I bought an AWS EC2 m6a.metal instance expecting to get something that is ready to be used. Now being told to recite arcane "commands" from the cloud computing holy book. They claim their supposedly groundbreaking hypertext protocol isn't even accessible to mere mortals using a $6000/month EC2, the blame is definitely on you, the user, for not setting up the tool in the right way.
This sysadmin cloud cult is basically saying that the EC2 product is actually not much more than an empty box, and that it is your responsibility to augment it with third-party servers and interpreters and application source texts that make it finally useful. And you better be carefully selecting the tools you install.
an EC2 instance gives exactly what you're told you'll be getting. You pay for a VM in some public cloud, you get it.
It's not that Claude code isn't a finite product per-se, I certainly can find some value in it. What I'm saying is that people selling it, through the convenient talks of prominent voices on the Internet and gullible C-suites, are trying to make it look like it's the only software engineer the world will need from now on. What makes me mad is not the deceptive advertising, that's already everywhere, it's the fact that the industry is happily believing all of this. If you raise any doubt, it must be that you haven't tried with the right skill.
Seems like you can ask Polsia about Polsia. The zero-arr.vercel.app AI investigative journalist interviewing Polsia could be this generation's Frost/Nixon.
if you feed it back any of the claims, it defends them pretty well
it doesn't deny any of the stats and cites none of it is secret as it's all published on the public dashboard
at the same time, churn is expected for such a company, doesn't mean it's not a real business, especially as AI gets better and margins improve... I think their business model is real
probably worth 30MM too since market is untapped, cost of user acquisition is much less than average profit per user and if one of the businesses actually takes off, they have 100% vendor lock-in and take 20% off the top
reply