FWIW: I see this pattern each time I discuss MIT (drifting from the Institutional Propaganda partyline) on HN:
1. Upvotes
2. Time passes
3. Downvotes
I'm starting to suspect Institutional astroturf.
2,141 karma · joined June 28, 2022
1. Upvotes
2. Time passes
3. Downvotes
I'm starting to suspect Institutional astroturf.
No, I can't provide a cite. Insider information. But whistleblowers have been threatened with being "Shwartzed" since (e.g. have frivolous criminal charges filed, to make their lives hell). In every case I've seen, whistleblowers signed an NDA, a non-disparage, and backed down.
That's why I don't trust a !@#$% thing from MIT anymore. Lies and fraud get covered up. Once that starts happening, you can't trust anything.
I want to get back to the era before people went to MIT as a center of wealth, power, and prestige, and would lie and cheat to get tenured positions there, to where it was a place where nerds went to nerd together.
And to a place where every !@#$% done didn't lead to a self-hyping press release, but where statements were precise, measured, calibrated, and most importantly, scientifically-accurate.
MIT is so very, very overcapitalized.
These things don't lead to anything other than a rich guy getting his name on a building, and more corruption in academia. By far the best thing which could happen to MIT would be for 95% of the endowment to go up in a poof of smoke, bringing it back to nineties levels. That's the end of the period when the Institute did high-integrity research.
They're surprisingly hard to intentionally defeat. Growing a beard, putting on weight, etc. won't do it.
I looked into this some time ago, when I didn't want to be tracked. It turned out to be annoyingly hard. And e.g. "wear a hat and mask" makes you rather conspicuous.
The increasingly universal cameras are rather effective at knowing what you're doing at all times.
This was possible with old-school graphics cards in the nineties and early 00's, when they were mostly a framebuffer, with a bit of 2d acceleration.
FYI: For most commercial network equipment (internet backbone and similar), reliability and security are the highest priority. Most of that is Linux, BSD, and friends.
The provider needs to comply with specific rules, have specific certifications, and sign specific agreements. You check the boxes, and you're good to go.
Microsoft does that better than anyone. OpenAI and Anthropic don't do that at all. Google does that rarely and poorly. AWS is not bad, but not as good as Microsoft.
Azure was always my go-to for regulated applications in the cloud. Some do require e.g. on-prem or even air gap, where even Azure is out.
A trillion-dollar business can easily trade dollars for the privacy. A business with $1M to spend won't even get a phone call with OpenAI or Anthropic, who were the only* previous players in town for doing this.
Worst-case example: Bootstrapped startup working in military.
It's also the case that an open model enables many more intermediate-cost solutions. E.g. providers certified for specific applications, on-prem rentals, etc.
* Omitting Azure, which gives some privacy for some $$$ on their models, but not at the level of high-security.
...
> standard commercial monolithic kernels
Note the path from A to B.
Fully monolithic: Linux, FreeBSD, NetBSD, OpenBSD, Solaris, AIX, zOS.
Mostly monolithic: Windows, MacOS
To get to microkernels, we're reaching for QNX and Minix.
Monolithic kernels won *hard* with good reason.
> critical flight systems
Hard RTOS is a whole different domain of engineering. You're very confused if you think the same principles apply to something which makes sense in a mainstream OS.
Yes, it should be. I can name zero times in the past half-decade that it's been possible to soft-reset the hardware in my GPU.
The "reset the GPU switch" is the power switch on my computer.
That's true of most hardware. If my ethernet, wifi, or whatever else enter an unstable state, on paper, the monolithic kernel can do the same thing as the microkernel: unload the module, then reload the module. In practice, I've virtually never seen a driver able to pick up a piece of hardware in an unstable state and be able to "reset" it.
> With device drivers in separate processes, bugs in one should not corrupt the memory of others, so restarting a device driver should be enough, instead of rebooting the computer.
Device drivers can corrupt other devices easily enough. The hardware itself would need to be built for isolation.
> Any complex program must be designed from the beginning to be easy to test and debug. If the device drivers are distinct processes, there is no reason to be more difficult to debug them. On the contrary, because the interactions between them are more limited than when they live in a common address space, they should be easier to debug.
That's exactly the myth.
In practice, a complex data flow for a heisenbug might start at the network layer, go through the disk, and land on the GPU. Tracing that on a microkernel requires instrumentation which, while possible in theory, I've never seen in practice.
I'd love to see a hardware stack where everything is ECC, devices are isolated, etc. but for now, on a typical computer, plugging in a bad USB device can brick the whole system (regardless of the software on it). The assumption is that devices on your PCIe bus can be trusted.
> This can only make much easier the reasoning about how the entire system works.
You make very strong statements, without support.
There's a reason Linux runs on every Android phone, almost every router, and increasingly, devices like home microwaves.
I don't think this is a fair comparison. New tools take you higher. I'll never climb Everest. A good comparison is:
- Climbing your local mountain
- Helicopter to the top of Everest
- Spacecraft to the moon
If a helicopter just saved you a trip up, you'd be right, but it can take you places you wouldn't otherwise go.
I created a new field of mathematics just yesterday, coincidentally. It wasn't a very good field of mathematics, though.
> Better security, better reliability, better modularity. Better security because in a muicrokernel [sic] system a bug in one driver only gives an attacker access to that subsystem (or maybe even that driver), as opposed to having unlimited access to the whole system like is the case with mainstream operating systems today. Better relaibility [sic] becasue [sic] a crash in one subsystem only crashes that subsystem instead of crashing everything; if windows were a microkernel system then the crowdstrike bug would have just stopped some IT security staff from getting telemetry instead of being front page news. If linux were a microkernel system then the linux kernel team wouldn't be responsible for merging every driver for every hardware device, and wouldn't be responsible for vetting code that they would have to become experts on the inner workings of every chip supported by the linux kernel to properly vet.
Is false. In particular:
- A kernel-level exploit can almost always be escalated. If I control your filesystem, you're SOL
- A kernel-level crash in any subsystem brings down the whole system. If your file system crashes, or your GPU is in an unstable state, there is no coming back
- Modularity is a question of clear boundaries. Whether those are static linking, dynamic linking, or XML-RPC doesn't change the level of modularity.
And so on.
Basically, none of the upsides have panned out. On the other hand, code complexity explodes. Microservices need to know who to call how. The whole "a crash in one subsystem only crashes that subsystem instead of crashing everything" means every subsystem needs defensive code for things going wrong elsewhere. KLOCs go up, bugs/KLOC stay constant, and things get less stable.
Plus, it's hard to reason about systemically, and without excellent telemetry, neigh-impossible to debug.
Examples of things with high-impact towards civics:
- Philosophy, especially political philosophy. Aristotle, Confucius, Marx, Smith, ...
- History, political science, economics, journalism
- Theory of law, studied academically. Lawyers, for the most part, have the opposite effect.
- Anthropology + international relations
Examples of things which currently don't:
- Most major in art, film, visual studies, literature, African/women's/LGBTQ/Asian/... studies, language majors, linguistics, ...
This is a statement about the practice of such majors. For example:
* Art/film/... ought to contribute, historically has contributed, and in much of Europe, actually still does. In most of the US, it leads to either unemployment or work in advertising, and doesn't teach the right things.
* Looking at the experiences of different minorities ought to contribute. In practice, most of these majors are dysfunctional self-admiration cults, with people citing each other's work, due to misaligned incentive structures in academic hiring.
* It's no coincidence that the most important work in philosophy, in the past century, was not done by philosophy majors.
* A Chinese / French / Russian / ... major will place you about one tier down from a Chinese / French / Russian / ... immigrant in terms of... just about everything.
In either case, many of these things do still have value, but it's not the same funding bucket.
It's easy to make the argument:
"If we invest $1M in education, we will have $10M in additional future economic output, $4M in future taxes, and $20M less in law enforcement / criminal prosecution / jail fees. It improves global competitiveness."
That's a no-brainer. Education is a very high ROI investment for a country. Like infrastructure spending or industrial policy, it's about cold, hard economics.
One step more complex -- but equally high ROI -- is towards having a functioning democracy. That's economics, but a bit more squishy.
Investing in the arts, humanities, and music is a good thing as well. However, that's a very different bucket of money. I wouldn't lump it in with the former two.
IP is the code and possibly the weights.
No one is stealing IP.
And I don't think anyone in their right mind would argue the AI industry isn't being fairly compensated.
To go further: most people in the world wouldn't feel bad at all if we found a way to slow what's likely the biggest socio-polical-economic change in human history down a bit.
This is an extreme claim, and incorrect. If you'd like to see counterexamples, you'll see many corrupt regimes in Africa, which did extreme harm to their own people. You'll see many regional powers.
One does not need to be a global superpower to be good or evil.
There's also nothing magical about Germany or Japan. Many countries had similar resources. Both chose to invest those resources into industrial militarism.
One can make the argument for a handful of countries which our outliers for land area or population, but in general, if any country chooses to invest in military and attack its neighbors, it has good odds of success.
> You have the UK, France, Germany, Japan, Russia, China and the US
Germany was an outlier, on the evil end, but otherwise, it's a selection of which facts one picks.
A comparison would require deciding which facts to compare on. For the US, the "evil" argument comes back to things like slavery and the genocide of the native peoples.
One can pick hundred of examples like Guantanamo Bay, fake vaccines in Pakistan, the Tuskegee Syphilis Study, the Tusla massacre, police violence, corrupt court, ...
The US does pretty nasty things, even if they don't always make US news or grade school textbooks.
> It's very cheap to label anything as propaganda without taking the time to appreciate whether it has any merit in terms of the overall behavior of a country or its people
This is an ad hominem, and a poorly placed one. You're discounting what people are telling you. A lot of the people here went through the US school system, learned "US rah rah rah" propaganda, and only deconditioned themselves as adults.
Many of us were where you are when we were younger.
I don't believe in US good, non-US bad. I also don't believe the same about my religion or my political party, for that matter.
How you measure depends on weights you assign (cultural system of values) and what information you use (media bias).
You can rank in the extremes (e.g. North Korea as worse than Belgium), since they come out that way by almost any set of information and values. Comparing the US to most other countries, there isn't a clear ordering. If you believe the things you wrote, I think the other comment summed it up well: "You sound like you've swallowed pro-US-propaganda hook, line and sinker."
Most countries have similar propaganda, by the way.
As many have pointed out, they're collecting data in ways much less aggressive than Anthropic itself about what Anthropic does.
Anthropic doesn't like it, but I don't see this as "Chinese companies stealing IP," any more than if Google tried to ban competitors from seeing how Google Docs or an Android phone behaves, or Ford trying to ban anyone from Toyota from seeing what their car looks like.
Please stop calling them "attacks." It's distillation training. It's looking at what Anthropic does -- as a block box -- and trying to duplicate or beat it. It's how progress happens.
There's the good "us" and the bad "them."
There's no reason to report on the odds of a meteor falling on my head tomorrow. There is every reason to consider the odds of:
- Nuclear apocalypse (esp. Cold War era)
- Bioweapon
- Climate change leading to disaster
- AI apocalypse
- Etc.
None of those are infinitesimal odds. That contrasts with, say, a zombie virus.
If you have a 5% chance of thermonuclear war each decade for 10 decades, you'll:
- Hear similar annoying statements
- They'll be true
With AI, we don't know if it's one week or one decade. This means we should assign probabilities and consider all possibilities, not get annoyed.
"The bet of using AI to speed up AI research is starting to pay off.
OpenBrain continues to deploy the iteratively improving Agent-1 internally for AI R&D. Overall, they are making algorithmic progress 50% faster than they would without AI assistants—and more importantly, faster than their competitors. The AI R&D progress multiplier: what do we mean by 50% faster algorithmic progress?
Several competing publicly released AIs now match or exceed Agent-0, including an open-weights model. OpenBrain responds by releasing Agent-1, which is more capable and reliable.28
People naturally try to compare Agent-1 to humans, but it has a very different skill profile. It knows more facts than any human, knows practically every programming language, and can solve well-specified coding problems extremely quickly. On the other hand, Agent-1 is bad at even simple long-horizon tasks, like beating video games it hasn’t played before. Still, the common workday is eight hours, and a day’s work can usually be separated into smaller chunks; you could think of Agent-1 as a scatterbrained employee who thrives under careful management.29 Savvy people find ways to automate routine parts of their jobs.30
OpenBrain’s executives turn consideration to an implication of automating AI R&D: security has become more important. In early 2025, the worst-case scenario was leaked algorithmic secrets; now, if China steals Agent-1’s weights, they could increase their research speed by nearly 50%.31 OpenBrain’s security level is typical of a fast-growing ~3,000 person tech company, secure only against low-priority attacks from capable cyber groups (RAND’s SL2).32 They are working hard to protect their weights and secrets from insider threats and top cybercrime syndicates (SL3),33 but defense against nation states (SL4&5) is barely on the horizon."
That's precisely where we are.
This is eerie. It's like a time traveler. The only delta is Anthropic is in the role of OpenAI.
and
Safety
Are opposites.
A good example of how this works is cocaine.
Capitalism and competition isn't always good governance. It works brilliantly in many places, such as restaurants or commodity goods. It fails completely for medicine or banking. It's in between for tech or education, but it's clearly failing for AI.
The key question is on direction of LLMs. Right now, LLMs are taking over human jobs. If the cost of silicon+power < cost of human being doing the same work, what rational reason is there to employ a human being?
If this applies to SWEs, lawyers, business analysts, many research scientists, .... this situation could persist for a long, long time. While capital costs less than the inputs of labor (nominal food, housing, etc.), there is no need for labor.
The key question is about continued progress in models, and of the tooling around them:
- Plateau: Old silicon obsoletes in due course
- Rise quickly: Old silicon maintains value for a long time
The way OpenAI describes it is ...
... concerning.
"Our goal is for models to learn about the world, not about private individuals. Privacy Filter helps make that possible." This means they're using sensitive PII to train models.
A smart AI will re-identify all the information -- including that in the 96% -- in a snap. That's already a solved problem.
He didn't win it. It was won by a team of students / collaborators / mentees, who felt he deserved it. I can't disagree with them. Among the nicest people in the world.
I don't think anyone meant it in the sense of "You're a Nobel Prize Winner," so much as "We couldn't have done this without your mentorship, and you deserve to hold onto this." He certainly doesn't consider himself to be a Nobel Prize winner.
- His breaking up images into grids was a poor-man's convolution. Render each letter. Render the image. Dot product.
- His "contrast" setting didn't really work. It was meant to emulate a sharpen filter. Convolve with a kernel appropriate for letter size. He operated over the wrong dimensions (intensity, rather than X-Y)
- Dithering should be done with something like Floyd-Steinberg: You spill over errors to adjacent pixels.
Most of these problems have solutions, and in some cases, optimal ones. They were reinvented, perhaps cleverly, but not as well as those standard solutions.
Bonus:
- Handle above as a global optimization problem. Possible with 2026-era CPUs (and even more-so, GPUs).
- Unicode :)