reinitctxoffset

The PRC is lapping us in everything from solar panels to electric cars to broad-based robotics and AI in real goods heavy industry.

Their power grid runs at a huge margin of excess capacity and they easily bring more online because they can still do infrastructure projects.

They are rapidly overtaking the United States in domestic, sovereign, and secure semiconductor fabrication. They're a couple of nodes behind but Kirin SoCs and multi-terraweight LLM inference on Aspire look pretty hot shit to me and no one can yank them around like a dog on a leash that it'll get turned off.

Government is dramatically more participatory than the western caricature. It is substantially at the local and regional level that it is directly participatory (so, the size of the whole US). At the very top it is representative in the sense that a party guy in Beijing is considered incompetent (they care about that) if the needs of the region are not on the agenda. When's the last time you called your congressman and got change?

Innovation is plural, research is open, the university system is in the loop, the public benefits.

Real wages are going up.

The PRC gets jumped in with Putin's Kremlin by lazy Americans who don't talk to Chinese people.

It's not the Kremlin. It's JFK in a Chairman Mao hat.

In the 90s and 00s when I was a kid, being a crook had real consequences here too. Not the death penalty per second, but the die in prison penalty left and right.

When John Meriweather and the rest of LTCM nearly blew up the market they didn't get a bailout, and the taxpayer didn't fund the hole in the balance sheet. The New York Fed organized private money and leaned on all their counterparties to get it done, but they didn't backstop it. Meriweather and the Sheik and Scholes and the rest were wiped out, they worked it off for a couple of years for salary and then skunk away in shame (we had shame back then). Took it like men near as I can tell. I admire John Meriweather a great deal in spite of the scandal.

President Clinton was impeached and very nearly removed from office for (ultimately) consenting but untoward involvement with a young woman (they got him on lying about it technically, but the political will was there because the country was furious about the skirt chasing.

Enron. Accounting that went from aggressive to sketchy to fraudulent (most of it would pass with flying colors today). Hard time. Skilling, Fastow I think just got out like five years ago (don't take my word for the date). Ken Lay IIRC died before they locked him up which saved him dying inside.

Madoff, died in prison. Ebbers, I think he died in prison too.

When Microsoft was gearing up to strangle the web in its crib the Justice Department pulled guys off of terrorists and human traffickers to go take a pipe to Gates until he backed off, he was allowed to keep Microsoft intact by letting the web happen, the Feds weren't asking, he decided to not fuck around and find out.

Consequences for serious fucking bad shit for people who are our leaders work. A people gets exactly what it demands from it's leaders and that's exactly what a people deserves.

Right now we're choosing to settle for a lot less, China is demanding more. Which is why we're getting our asses kicked.

Interactive use cases: the web interface, the mobile interface, the design tool. The fast variants.

In an HBM bandwidth constrained setting you're dealing with something called "roofline analysis" (comes originally from NUMA work circa ~2009 but it's applicable to modern GPUs). Great diagram from the JAX people:

https://jax-ml.github.io/scaling-book/roofline/

In order to get your money's worth from a modern GPU (or disagg rack like an NVL72) you need to decode (the one token at a time thing) across big batches of context windows. To the left of that point where it hits "the roof" you're idling tensor units. TensorRT-LLM likes batches of 4096, so BS=4096.

In the case of one person chat prompting their local LLM, BS=1, totally bandwidth limited.

So the game is to set some latency target with some control theory primitive (PID or something) and then delay the next token until a batch is big enough to not waste tensor units. This is a real trick when a human is waiting (you've probably seen the thing in Claude.ai where it's all bursty and then they reflow the whole block with JavaScript).

Agentic workloads are huge piles of context windows where you've always got enough who want the same experts on the next token, you're always to the right of that intersection. And it doesn't really matter if it's on the other side of the world, or lags by a second, it's fine.

Claude Code soaks up all the tensor units that would be idle until they're full, and only then does it leak into the capacity reserved for highly interactive use. It's the bottom of the barrel until it's rinsed the fuck out.

They want more margin on agentic tokens. That's it. The COGS on them is the absolute lowest of anything they do.

Agentic workloads are the most batch friendly, latency insensitive, geography insensitive, migration insensitive tokens that a big lab ever sells. In the ads business such inventory is called "remnant". The sausage is made of whatever is left over when the choice cuts have been removed.

This talking point from Anthropic that Claude Code sitting in a Ralph Loop is burning top sirloin interactive session tokens is bad faith hogwash and it only flies because most everyone who has run this shit at scale either already works there, sells them hardware, or hopes to be an acquisition target.

I'm none of those things, so I'm happy to tell you they're lying. I know, it's hard to swallow, but it turns out Altman and Amodei are occasionally full of shit.

The Pentagon classifies Anthropic a supply chain risk and is appealing a ruling with vigor to reinstate the it's right to decline doing business with them.

I don't know what part of this you guys are having trouble with, but it doesn't get a whole hell of a lot more emphatic on "what the US Government thinks" than who the Pentagon is in court with to avoid doing business with that party.

Mr. Hegseth is the representative of the administration in the AI usage policy of the largest bureaucratic organization in the history of civilization. He has emphatically rejected at least one black box American AI megalab and President Trump has endorsed this action on Truth Social.

The government's stated position is not what you among other commenters are stating or implying as the government's position.

As of writing, https://openrouter.ai/rankings shows only Opus, Sonnet, and GPT 5.5 in the top twenty by usage (Opus and Sonnet both have two slots on the 4.7/4.8 and 4.5/5 splits respectively, read that how you will).

I don't know how much business OpenRouter does in Europe (they have some GDPR text in some settings pages I think) but it's zero in China.

You might also consider the countless companies that do nothing but open weight LLM inference: Together (of Tri Dao fame), Fireworks (founded by my old colleague), Baseten, SambaNova (of Chis Re fame), and too many more to count, there's a new one every week. Plus NVIDIA does a ton of this business via NIM and shit.

I'm trying to be reasonably polite because it's possible you literally didn't know that, but this reads as a troll, so, if it is a troll please stop.

The assertion that it's in the United States Government's best interests ban Chinese open weight models is a very strong opinion that is not a consensus even at the fringe of Thiel-adjacent psycho thought: Alex Karp is on record about open weight models being necessary, the fucking "we bombed a bunch of kids with Claude doing rubber stamp target selection" guy. He thinks "trust OpenAI and Anthropic" is a radical position.

Peter Hegseth, another really pro-America being powerful guy, he's dealing with a lawsuit because he doesn't want Anthropic in his military, he calls it a supply chain risk (he's right).

There is no evidence of any kind that a complex attack vector can be trained into model weights and survive all the crazy slicing and dicing that happens between published weights and running model. These things get quantized and run on mathematically imprecise kernels and sampled and LoRA-tuned and Dolphin/Orca de-tuned. Go look at what the ComfyUI community comes up with, those guys know more about WAN 2.2 than the people who trained it. Because those models run for real on a desktop, so there's mad innovation at light speed.

There is no one who wants a capriciously expensive black box run by extremely creepy people, not once the capability crosses over (in about November).

But don't take my word for it, you just had a chance at one AI IPO, and I'm sure you'll get another, so if you like how that goes, you don't need to convince me!

If the Trump administration decides to annoint Altman and Amodei in defiance of market forces it will rapidly discover that it no longer has the sovereign bond auction pricing power to prop them up. This isn't 1998: the Treasury has taken five major body blows in the last 25 years, the world's energy markets, maritime insurance regimes, electronic payments rails, and moral authority in places like the UN Security Council are effectively bifurcated. Oil and money and data and people just flow around the United States now. Canada's sovereign debt yields tank harder when the market is spooked, that's the North American flight to quality.

The administration could probably put some serious friction on open weight model use in the Fortune 500 for a little while, but the opposition never got such a gift right before a squeaker midterm. And outside of major enterprises with puckered ass compliance departments? Not a chance. It's popular around here to forget Uber and AirBnB and yes, OpenAI and Anthropic all got their start flagrantly breaking the law and grew lawyers and lobbyists faster than anyone could enforce it. And this time everyone from the DNC to the EFF would be holding hands wearing "Save The Models" t-shirts. Not even NVIDIA is remotely pretending they're anything but all in on GLM 5.2, they had an NVFP4 quant up by the time most people read the blog post.

And the Trump Administration isn't exactly enamored of Comrade Amodei at the moment, being as they're appealing the lawsuit Anthropic brought against the Pentagon during a shooting war.

Forcing the American proprietary AI megalab financing event was our fiscal Ukraine Special Military Operation, the market is calling the bluff and neither the capital markets nor the Federal Reserve has the dry powder to absorb this one.

The Treasury auctions will flat not clear in an orderly way. We can't raise 2-4 trillion dollars on a dime in 2026 and if CoreWeave turns out, as many suspect, to be Patient Zero? It would be that big a hole.

We play by the same rules as everyone else now. I hope we regard it as being worth it, but I fear we will not.

They're in a price war with the People's Republic of China running flat out with the full backing of a government that literally does not care if they ever see a financial return on the investment, they just want to drive the value of LLM training and inference to zero because we banked the market on it being arbitrarily high margin forever. China was like hold my beer.

They have a staggering surplus of grid capacity and can bring more online without any difficulty. We couldn't get a serious nuclear project done if Jeffrey Epstein was offering private flights to the ribbon cutting.

In the United States at any given time more than half of the FLOPs are badly misallocated, Meta has like, a double digit percentage of the total capacity going down the drain every day and has for years. That's a conspicuous example but on OpenRouter rankings it's rare to see more than one or two American vendors in the top 10, sometimes the top 20. But 3rd, 4th, and 5th place are all merrily burning half the compute duplicating effort and missing key innovations because we stopped publishing real results. In China if DeepSeek makes a breakthrough it's at Zhupai and Moonshot and MiniMax and MiMo and Qwen that week.

Our only lever, export restrictions, seems to do nothing but breed multiply antibiotic resistant super hackers who just get more efficient and immediately propagate all of those efficiencies to the rest of the Chinese AI industry.

At the beginning of 2026 there was one Chinese lab with a model that had any real relevance fielding modern tool users. Today in July there are like, eight lagging the absolute frontier by maybe 3-6 months. Barring some massive bend in some curve 3-4 of the top 5 and 6-8 of the top 10 will be Chinese and open weight by January.

The great irony in all of this is that our current playbook is straight out of the 1960s USSR, and the PRC's current playbook is straight out of 1960s USA. We're the ones with the opaque decision making and gross resource misallocation driven by the personal agendas of a shadowy cabal of frenemies wired back channel into government in the form of the individuals rather than the offices. They're the ones with a thriving marketplace of ideas powered by robust public/private partnership and a paved path running bidirectionally to the university system.

It's going to implode because the Kruschev system does. Theirs is going to thrive because the Kennedy system puts a man on the moon before the decade is out.

I hope to open source some of this soon but it's still married to quirky internal build tools and other jank. I'll make a note to come back to this comment when there's code up.

In terms of process I don't know that I really have any unique edge or anything, I'm still very intermediate as a lean4 programmer. I was a strong Haskeller before so I had a little bit of a map. The key thing about lean4 coming from e.g Haskell is that it has a much cleaner type tower, there's no type/kind/PolyKind jank, it's just Type 0, Type 1, as high as you care to go (which usually isn't very high), so a fair bit is unlearning some of that.

A mistake I made was to try to learn lean4 in emacs, which I'm working on making more reasonable, but today lean4 is a vscode only experience IMHO, this is partially because the default Mathlib4 style is really unapproachable without the tooling (not being a professional mathematician I'll stop short of calling the style bad, maybe it's intuitive to mathematicians, but it's very ill suited for programming in the large as a hacker would approach it). Proofs of theorems are this tactic match space with a lot of implicit projection up into a big possibility space at every step, and then the next explicit tactic is grabbing things from that space often very implicitly. It's my least favorite part of the semantics (or maybe one I feel is over-used, there are ways to make it more explicit, the canonical example is probably `simp_only` which grounds in specific lemmas).

The effectful part of the language should be a pretty short hop from anything else with do-notation, it reads very naturally and the conditional/looping/match stuff sorta looks like Rust if you squint a little. One place the AI assist really shines is to go rename all the variables something reasonable, a lot of otherwise great code looks like it hybridized with a bizarre Hungarian-notation splinter faction, the keywords are fine.

There's some stuff that probably makes a lot of sense in a meta mathematical view of things that makes zero sense from a TCB standpoint. To mathematicians there is a distinction between the map coloring proof and Wiles' FLT proof. This is real and it makes sense in mathematics. But in high-assurance software having an issue with `native_decide` as opposed to `bv_decide` is a weird religion: TCB certifying them is identically compromised by a compromised compiler.

So I think it's mostly just a very small community trying to write code as the primary activity and layering on formal methods as a correctness strategy as opposed to a much bigger community doing real maths where the programming language is more of an amenity than the main show.

I suppose my approach is to be willing to go a different way in a different use case.

I just got so fed up with Starlark trying to do a nontrivial RBE prelude that's fairly different than Meta's.

I added WASM/WASI bindings and wrote the prelude in Haskell.

Both the bindings and the prelude took less time than I already wasted on multiple attempts in Starlark that all collapsed into string goop.

It's nice to be able to bound the execution environment in a build tool but still use a serious programming language.

Given Zig has excellent support for targeting WASM, seems you'd get the same advantages.

It's early days of using it as a general purpose programming language, my initial use case was using it as a metaprogramming framework that lowered into target languages with some rigor on correspondence of the lowered artifact and the proof-amenable surface. By lines of code the biggest use case historically was describing state machines and codecs/protocols for high-assurance IO primitives: define QUIC or ZMTP once and get idealized performance C++ and Rust and Haskell (Zig coming soon) protocol suite native to io_uring.

Another big use case is GPU kernel compilation. I've encoded the polyhedral algebra used by CuTe and all of the stated theorems from nvfuser in lean4. From this you can recover scheduling and tiling properties that must be met to get equal or better performance to CASK (likely optimal) for many operations, and then lower that to PTX. Surprisingly often the candidate set for the fastest possible kernel turns out to be finite and enumerable, so fast inference.

Somewhat recently I started benchmarking stuff just directly in lean4 with a very small TCB shim to e.g. io_uring. I've done like the basic systems programming ABCs (I've got a reverse proxy that's lean4 and ~100 lines of C and shatters nginx, faster at every point on the histogram, memory use flat as a strap). Lithe runs on it so I can do web serving as fast as anything in the hyper ecosystem.

Now that I've got my feet with it I'm digging in on a serious project which is a hypervisor that addresses the deficiencies in firecracker. This is a place where some proven properties can really get you a lot of performance (something like firecracker has to be very defensive regarding e.g. virtqueue semantics in the presence of a malicious guest). Firecracker is an exceptional piece of code and it's carried me far but at agent scale it's time to go another turn of the crank on virtualization performance.

I've gone from zero knowledge of lean4 to the point where I'm doing most of my coding with it in ~6 months, and this was dramatically helped by how facile the AI assist is: it's remarkable how consistently fluent models are in lean4. I've found this to be true of the near frontier and smaller local models alike, LLMs just seem to get lean4.

I still have a ways to go before calling myself a lean4 expert, but I don't need assist to get useful programs anymore.

The ability to start with very little knowledge and still be able to trust parts you don't fully understand is a real unlock on learning progress: it's both practical and motivating to get useful programs you can rely on with incomplete knowledge, it sort of drags you in. You're bounded by the subset of the language that describes your axiom and proposition surface, not the subset that describes the intermediate steps. Over time as your ambition goes up, you need to understand more to do more things, but you can operate safely at level N+1 in a sense.

It's also just a delightful programming language irrespective of its theorem proving role, and it's remarkably fast. I've got it bolted to io_uring and in many cases it blows the ass off of C++ with libuv or Rust with Tokio. Now and again you'll see some huge tail at the p99.99 latency or something and you go make a number fixed width or something, but you have to tune C++ and Rust too.

It might just mean "please oppose the inevitable attempts to privatize AI governance".

Nothing has ever been, directly or indirectly, deficit financed at this scale before. In notional or real terms, in history, by anyone.

Now maybe there's an argument that it's a good investment: we are going to beggar the Treasury to buy 2CTA on CoWoS out of Taipei and DCs the size of Manhattan. I personally think we could have done a little more engineering before deciding that the big blind was like, 5 trillion all counted, but it was going to be expensive no matter what.

What super weird is that we're running a project where the "penny" to the "dollar" is the Manhattan Project, and a couple of super weird dudes who do MDMA at Lighthaven now and again are like, in charge of it.

Your comment being top of thread, and you seemingly being conversant in the details of the issue, would be exceptionally well-placed to make an argument on the merits of the subject matter. Character portraits are sometimes useful, but technical detail is often conclusive.

What's the steelman of djb's position and why does it fail scrutiny? To the uninitiated, it sounds like his preference for the hybrid classical/BQP scene is prudent given the marginal computational burden.

But I'm guessing, it would be better if an expert weighed in with details.

The regress ends somewhere, because (barring some pretty sharp changes to the way the law works basically everywhere) ultimately someone has to certify the outcomes as acceptable. This might be in the form of the market (though AI-adjacent stuff seems extremely prone to prolonged market failures), this might be regulatory in nature. This might be the executive management of the companies involved.

Personally I think that if you cranked the capability up high enough the first person you'd run into who absolutely demanded more than vibes and didn't care about your singularity thesis would be the representative of a reinsurance firm: mostly to do serious stuff without bending the law, you need insurance, and I am unaware of anyone writing serious policies (certainly not ones that make any economic sense) that underwrite the risk of AI autonomy outcomes financially.

When Swiss Re writes a policy that Anthropic Cinematic Universe or whatever iteration we're on won't fuck it up?

Now maybe we're talking. Until then you ask three practitioners and get nine answers, no one knows what they're talking about unless they're doing a really good job keeping it quiet (and that's probably what you'd do!).

It's an open problem of clearly large value how to get reliably useful and trustworthy outcomes from AI systems in many domains, software is maybe the signal example of that. If one had solved it resoundingly and scaleably, one could in fact "get rich quick".

It is unsurprising that a lot of people claim to know how to get rich quick.

I believe it is possible to solve this problem, and I have my own horses in the race which I won't threadjack to promote here, but it's the central problem of our profession at the moment. We've all seen the truly discontinuous outcomes and we've all seen allegedly national security dangerous models (which at one time was GPT-3) faceplant with it's shoelaces tied together. I wanted to see if Fable was really all that and I left it overnight on some fairly straightforward C++ (code DSv4 Flash works on with moderate supervision) and it's pretty roast worthy, I gave it a chance to redeem itself this morning and it's ticked up a bit (I still think it's roughly Opus 4.8 with a Project Zero fine tune and DRO trained off the constant gratuitous yield tic which is pretty clearly an intentional gimp).

I give all such claims 30 seconds of my time because someone is going to actually be right one of these days.

Not if you define the behavior. Many programming languages have robust memory models ranging from "happens before" style systems like those in Java or Modern C++ to the Software Transactional Memory system in GHC/Haskell.

Correct, shared mutation is an extremely well-understood problem in the sense that there are points in the programming language design space that admit it with present technology.

As always, there are tradeoffs.

I'm fairly close to releasing something called nix-compile, which is a type inferring Hindley-Milner compiler that unifies across nixlang and bash. It finds a lot of bugs. Because it has a very complete view of your project, it can do cute stuff like precise go to definition across modules, annotate code with types in the comments, it's cool.

The only thing with a worse inflation story in 2026 than the United States Dollar is the Zero Day.

Used to be a zero day got you unauthorized access to a computer system with no warning.

Now it might not even get a maintainer to do a patch when they're bored.

It would represent a pretty sharp inversion from all the progress of mathematical physics until the present.

Up until the present it has been a nearly uniform march of revealed symmetries, collapsed privileged frames of reference, and other such (in the deepest sense) simplifications in our model of reality that has improved its fidelity to the measurable.

I hang qualifier about these developments being simplifying because the result isn't simple in the details: quantum chromodynamics is a daunting subject! But it's not just an enumeration of details and contradictions, the particle zoo that preceded the Eightfold Way looked like line noise, now in indexed notation the Lagrangian of the entire Standard Model fits on a page (or so I've been told I've never actually seen the page).

It's almost tautological that the frontier where it's still messy involves an unrevealed symmetry or a persistent privileged frame of reference, that's what frontier means, we don't see past it to the seam where it folds up.

Personally I suspect AI systems will be a great deal more inclined to discard the parochial axioms that have every point placed human ego above simplicity.

It doesn't resolve all of the open problems in physics if you amputate consciousness, free will, agency persistent identity, and an unambiguous arrow of time.

But it starts looking possible to make progress.

I can't signal boost this enough.

I spent months, months of late nights watching commits to nvfuser and shit, I wrote a SASS decompiler instrumented everything trying to learn Blackwell.

This is the first time I've seen something so clean, just a real work of scholarship on it.

My hat is off to the authors and the contribution it represents.

If I would caution a reader anything it's that the 2CTA (sm_100 sm_110) patterns here are different on 1CTA in important ways and it's not a better / worse thing, they are good for different workloads.

Really outstanding work. I proves q lot of this in lean4 and published but I got lazy short of really doing the pedagogical work.

This is what you should be starting with if you want to max out 2CTA gear, it's immaculate.

On modern super wide znver5 or SBSA with full-clock scalar 256 or 512 ALUs / SIMD lanes deep pipelines hight BTB pressure eyc. it's just really difficult to make a priori statements about performance for a given workload.

absl::flat_hash_map (or folly::F14) are great defaults if you can eat the invalidation semantics.

But if it's really hot you measure by workload and have infrastructure to flag the right ones in.

This seems promising. I'll start benching it alongside the other likely lads.

I was advocating for "available weight" as a value neutral term for a while.

I gave up. No one cares. And no one will ever tell the truth about the training anyways.

Substantial and growing freedom beats zero freedom ever again.

What people aren't getting with `firecracker` is utilization. Don't get me wrong, `firecracker` is great software and it's what I'm using for lightweight virtualization, but workloads are really bursty over really short periods of time now, even with the snapshot and restore that you can get if you're willing to hack on `firecracker` substantially, you hit walls where it's like, this is too much against the grain, this thing wasn't designed to bounce from 1 core to 32 to 8 to 16 to 4 to 32 to 1 seamlessly, and that's what it takes to get extreme utilization even with extremely good ML on the prediction.

I am quite sure I'm not the only person working on post-firecracker KVM.

I'm inclined to believe the author, given that they were honest about the machine assist on the engineering? Why would they lie about the post and not the artifact?

More broadly, I'm just getting tired of the low effort snark on this. If someone is going to cite a passage, or examples of a theme, do some serious analysis by all means. I'm not trying to tone police high effort comments.

But this drive by "that's LLM output" thing is the new low effort "javascript is just as hard as c++" a.k.a "i don't know how to do that and i have no intention of learning and it feels bad so i'm going to lash out".

Make a case that something is deficient on its merits or don't, but baseless accusations of dishonesty are not a reasonable default. The parent here did real work that no one paid them to do, donated it to the community asking nothing in return, it's meticulous and high value, and they've already been run over by the FSF goons, you really want to pile on with no evidence?

In fairness I only saw this like, an hour ago, so it'll probably be at least until after work before I start messing around with it.

Pardon my shorthand. I meant open source and licensed under both MIT and GPLv2 such that distros build it alongside every other `.ko`.

One imagines if anyone has an issue it's with the RISC-V blob in GSP. Now while I myself wouldn't brave the wrath of NVIDIA's lawyers by like, calling Ghidra on it or anything, one imagines it doesn't have a lot of secrets from the motivated tinkerer!

FWIW no one can tell anymore. Claude stopped doing all the LLM ticks like 6 months ago. There's a whole industry of people trying to detect LLM writing and they're getting stomped, this can be and is extremely well studied in the literature.

The criticism is that you did something wildly ambitious and pulled it off. The blog is just well written.

This is outstanding, if I have time I'll be switching over today.

This is the kind of thing that could drive a truly free fork of emacs forward, it's enough better on realistic desktop displays to rally around and as the parent discovered "Free Software" at this point has very little to do with the freedom to do what I want on my computer in a low friction way: an ideological position on "GPUs" as a category is bizarre even by Late Soviet FSF standards. By all means cite a vendor and a policy, but even NVIDIA is in tree now, it's got the same software freedom as ext4 and I don't hear anyone talking about chains on that.

In the age of machine assist emacs could get a modern fast/cachable build, clean under all the sanitizers, io_uring on Linux, deterministic clang formatting, compat break with zero-use junk from the 80s, WASM compilation for polyglot extension (I like lisp but I understand why some people don't), modern networking, modern chrome, 100% vscode compatible LSP, modern theming that defaults to something that doesn't drive users away. I would love to have a ten line init.el instead of 4k of workarounds.

Maybe this can be the nvim moment.

I love emacs but the nvim people have so many nice things and FSF emacs has a shelf life. If someone out of their own time and resources did a cross platform, mechanically verified, dramatically accelerated at HiDPI patch to basically anything else they'd be greeted like a hero.

Keep up the good work legend.