HNHacker News
TopNewBestAskShowJobs

nickysielicki

9,596 karma · joined July 28, 2014

email me: hackernews@nslick.com
submissionscomments
nickysielicki··on We Will Not Be Divided
> The government isn't banning Anthropic because using it harms national security. They are banning it in retribution for Anthropic taking a stand.

You might be completely right about their real motivations, but try to steelman the other side.

What they might argue in court: Suppose DoD wants to buy an autonomous missile system from some contractor. That contractor writes a generic visual object tracking library, which they use in both military applications for the DoD and in their commercial offerings. Let’s say it’s Boeing in this case.

Anthropic engaged in a process where they take a model that is perfectly capable of writing that object tracking code, and they try to install a sense of restraint on it through RLHF. Suppose Opus 6.7 comes out and it has internalized some of these principles, to the point where it adds a backdoor to the library that prevents it from operating correctly in military applications.

Is this a bit far fetched? Sure. But the point is that Anthropic is intentionally changing their product to make it less effective for military use. And per the statute, it’s entirely reasonable for the DoD to mark them as a supply chain risk if they’re introducing defects intentionally that make it unfit for military use. It’s entirely consistent for them to say, Boeing, you categorically can’t use Claude. That’s exactly the kind of "subversion of design integrity" the statute contemplates. The fact that the subversion was introduced by the vendor intentionally rather than by a foreign adversary covertly doesn’t change the operational impact.

nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
Obviously, but that's a huge number and some tens-of-billions amount of that absolutely does flow towards hyperscalers. Contractors need compute.
nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
Huawei did eventually end up on the entities list, but there was a gap between when it was initially announced and when it became law, and the divestment from contractors started immediately overnight.
nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
obviously, I was never suggesting that the DoD spends $961b a year on cloud computing.

Look, it’s a very simple question: Amazon has invested $8b into anthropic. Do you think if the DoD disappeared tomorrow that Amazon would lose more than $8b in revenue over the next 5 years?

I think you underestimate how large the DoD budget is and how many times that money changes hands in the pursuit of fulfilling contracts. $20b-$25b in revenue per year across all hyperscalers is a totally reasonable estimate.

nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
The counterparty risk on those buildout contracts is not the same as their equity investments. Amazon isn’t assuming the entirety of that buildout exposure as a vote of confidence or form of investment in anthropic; they’re hedging it with insurance, credit default swaps, and MAE clauses.

Those datacenters are AWS infrastructure that Amazon owns and can repurpose. The equity stake is the only part that’s truly at risk, and $8B is a rounding error on Amazon’s balance sheet.

nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
Source?

Half of that budget gets contracted out to Lockheed, Raytheon, Northrop, Boeing, General Dynamics, etc. Those companies absolutely do spend money on the hyperscalers.

nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
The cost of a company like Amazon or Google losing their piece of that $1T annual budget is greater than their exposure to the failure of Anthropic.
nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
* * *
nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
You expect hyperscalers to play chicken with the DoD?

The courts have historically been pretty consistent about giving the DoD whatever the fuck they want, going back to WW2 and even longer for the predecessors of the DoD. I agree that the next administration might reverse it, but the thing is, the government will stay irrational longer than Anthropic will remain solvent.

The US government told every American company to stop doing business with Huawei and they all did it overnight, even when it cost them billions. TSMC stopped fabricating for them, Google pulled Android licensing… The machinery of sanctions compliance is extremely well-oiled and companies fold instantly because the outcome of noncompliance is literally getting thrown in prison.

nickysielicki··on I am directing the Department of War to designate Anthropic a supply-chain risk
This could kill Anthropic.

The designation says any contractor, supplier, or partner doing business with the US military can’t conduct any commercial activity with Anthropic. Well, AWS has JWCC. Microsoft has Azure Government. Google has DoD contracts. If that language is enforced broadly, then Claude gets kicked off Bedrock, Vertex, and potentially Azure… which is where all the enterprise revenue lives. Claude cannot survive on $200/mo individual powerusers. The math just doesn’t math.

nickysielicki··on Statement from Dario Amodei on our discussions with the Department of War
Pagerank is not Claude.
nickysielicki··on Show HN: I ported Tree-sitter to Go
right, so why are you asking me to imagine one machine that can calculate using logic circuits and another that can calculate using a lookup table when we’re in agreement that they’re the same thing?

I reject the premise of your analogy.

nickysielicki··on Show HN: I ported Tree-sitter to Go
never mind the fact that the model is constantly reseeding itself against the files it’s reading from your working directory, so the prompts are useless on their own.
nickysielicki··on Show HN: I ported Tree-sitter to Go
But logic circuits are look up tables.
nickysielicki··on Show HN: I ported Tree-sitter to Go
Well, that’s obviously bad.

But once you told it to stop cheating, did it eventually figure it out? I mean, correctly implementing fuzzer support for a project is entirely within the wheelhouse of current models. It’s not rocket science.

nickysielicki··on Show HN: I ported Tree-sitter to Go
Is that what this was all about? Depth and integrity? Rigor and hard work? Because I thought it was all about writing useful programs for computers.
nickysielicki··on Show HN: I ported Tree-sitter to Go
There never was a cohesive generic open source community. There are no meaningful group norms. This was and always will be a fiction.

I’m tempted to just start putting co-authored-by: Claude in every commit I make, even the ones that I write by hand, just to intentionally alienate people like you.

The best guardrails are linters, autoformatters, type checkers, static analyzers, fuzzers, pre-commit rules, unit tests and coverage requirements, microbenchmarks, etc. If you genuinely care about open source code quality, you should be investing in improving these tools and deploying them in the projects you rely on. If the LLMs are truly writing bad or broken code, it will show up here clearly.

But if you can’t rephrase your criticism of a patch in terms of things flagged by tools like those, and you’re not claiming there’s something architecturally wrong with the way it was designed, you don’t have a criticism at all. You’re just whining.

nickysielicki··on Show HN: I ported Tree-sitter to Go
I’m just curious, what would need to happen for you to change your opinion about this? Are you basically of the opinion that it’s not good enough today, never will be good enough in the future, and we should just wind back the clock 3 years and pretend these tools don’t exist?

It feels to me like a lot of this is dogma. If the code is broken or needs more testing, that can be solved. But it’s orthogonal: the LLM can be used to implement the unit testing and fuzz testing that would beat this library into shape, if it’s not already there. It’s not about adding a human touch, it’s about pursuing completeness. And that’s true for all new projects going from zero to one, you have to ask yourself whether the author drove it to completeness or not. That’s always been true.

You want people to hedge their projects with disclaimers that it probably sucks and isn’t production worthy. You want them to fess up to the fact that they cheated, or something. But they’re giving it away for free! You can just not use it if you don’t want to! They owe you nothing, not even a note in the readme. And you don’t deserve more or less hacker points depending on whether you used a tool to generate the code or whether you wrote it by hand, because hacker points don’t exist, because the value of all of this is (and always will be) subjective.

To the extent that the modern tools and models can’t oneshot anything, they’re going to keep improving. And it doesn’t seem to me like there’s any identifiable binary event on the horizon that would make you change your mind about this. You’re just against LLMs, and that’s the way it is, and there’s nothing that anyone can do to change your mind?

I mean this in the nicest way possible: the world is just going to move on without you.

nickysielicki··on Don't create .gitkeep files, use .gitignore instead (2023)
> Having 100 tiny gitignores makes finding out why something is excluded annoying. Our policy is one root level gitgnore and gitkeeps where required.

This is not a complicated or important enough problem to justify a team-wide policy. Let it work itself out naturally.

https://git-scm.com/docs/git-check-ignore makes it trivial to debug repo-wide gitignore behavior.

nickysielicki··on AI makes you boring
That's not what I'm saying. My point is that expertise, as in, credentials, institutional knowledge, accepted wisdom, was actively harmful to solving flight. The Wrights succeeded because they built a tool that made iteration cheap (the wind tunnel), tested 200 wing shapes without deference to what the existing literature said should work (Lilienthal's tables were wrong and everyone with "expertise" accepted them uncritically), and they closed the loop with reality by actually flying.

That's the same approach as vibe coding. Not "asking Claude to make a CRUD app.", but using it to cheaply explore solution spaces that an expert's priors would tell you aren't worth trying. The wind tunnel didn't do the thinking for the Wrights, it just made thinking and iterating cheap. That's what LLMs do for code.

The blog post's argument is that deep immersion is what produces original ideas. But what history shows is that deeply immersed experts are often totally wrong and the outsiders who iterate cheaply and empirically take the prize. The irony here is that LLM haters feel it falls victim to the Einstellung effect [1]. But the exact opposite is true: LLMs make it so cheap to iterate on what we thought in the past were suboptimal/broken solutions, which makes it possible to cheaply discover the more efficient and simpler methods, which means humans uniquely fall victim to the Einstellung effect whereas LLMs don't.

[1]: https://en.wikipedia.org/wiki/Einstellung_effect

nickysielicki··on AI makes you boring
> AI models are extremely bad at original thinking, so any thinking that is offloaded to a LLM is as a result usually not very original, even if they’re very good at treating your inputs to the discussion as amazing genius level insights.

This is repeated all the time now, but it's not true. It's not particularly difficult to pose a question to an LLM and to get it to genuinely evaluate the pros and cons of your ideas. I've used an LLM to convince myself that an idea I had was not very good.

> The way human beings tend to have original ideas is to immerse in a problem for a long period of time, which is something that flat out doesn’t happen when LLMs do the thinking. You get shallow, surface-level ideas instead.

Thinking about a problem for a long period of time doesn't bring you any closer to understanding the solution. Expertise is highly overrated. The Wright Brothers didn't have physics degrees. They did not even graduate from high school, let alone attend college. Their process for developing the first airplanes was much closer to vibe coding from a shallow surface-level understanding than from deeply contemplating the problem.

nickysielicki··on Google Public CA is down
What qualifies as a non-compliant certificate?
nickysielicki··on 10 years building vertical software: are we cooked?
> Imagine being a vim or emacs user and have those replaced by something you have to type entire sentences for functionality.

This has already happened for a non-negligible amount of people on this site. I still use magit to prepare commits and review diffs. Everything else is English and Claude code or opencode.

Ironically, part of the reason I like emacs and formerly liked vim was because it reduced the amount of time my hands had to leave the keyboard. I simply look where I want to type, press the chord to jump with avy, and then begin typing. New tools are spiritually aligned with this goal. I look at the screen, I think about what I don’t like, and now instead of translating the criticism into code changes, I just stream my thoughts directly into the tool.

nickysielicki··on Anthropic raises $30B in Series G funding at $380B post-money valuation
How does their top tier subscription compare in usage limits to the $200/mo Claude usage limits?
nickysielicki··on Keeping 20k GPUs healthy
> A deep dive on why these beastly cards fail so frequently compared to all other common current day hardware would be fascinating!

P=CV²f

nickysielicki··on Keeping 20k GPUs healthy
Totally matches my experience, and it feels bizarre inside-looking-out that nobody else talks about it. Hardware from 2010-2020 was remarkably stable, and CPUs are still as stable as they were, but we've had this large influx of money spent on these chips that fall over if you look at them funny. I think it leads to a lot of people thinking, "we must be doing something wrong", because it's just outside of their mental model that hardware failures can occur at this rate. But that's just the world we live in.

It's a perfect storm: a lot of companies are doing HPC-style distributed computing for the first time, and lack experience in debugging issues that are unique to it. On top of that, the hardware is moving very fast and they're ill equipped to update their software and drivers at the rate required to have a good experience. On top of that, the stakes are higher because your cluster is only as strong as its weakest node, which means a single hardware failure can turn the entire multi-million dollar cluster into a paperweight, which adds more pressure and stress to get it all fixed. Updating your software means taking that same multi-million dollar cluster offline for several hours, which is seen as a cost rather than a good investment of time. And a lot of the experts in HPC-style distributed computing will sell you "supported" software, which is basically just paying for the privilege of using outdated software that lacks the bug fixes that your cards might desperately need. That model made sense in the 2010s, when linux (kernel and userspace) was less stable and you genuinely needed to lock your dependencies and let the bugs work themselves out. But that's the exact opposite of what you want to be doing in 2026.

You put all of this together, and it's difficult to be confident whether the hardware is bad, or going bad, or whether it's only manifesting because they're exposed to bugs, or maybe both. Yikes, it's no fun.

nickysielicki··on Opening the AWS European Sovereign Cloud
It's a regulatory requirement:

> Replicating a broadly practiced mitigation mechanism that is established in EU institution and government hiring practices, operational control and access will be restricted to EU citizens located in the EU to ensure that all operators have enduring ties to the EU and to meet the needs of our customers and partners.

- https://www.aboutamazon.eu/news/aws/aws-european-sovereign-c...

nickysielicki··on Deno has made its PyPI distribution official
But that's kind of my point, they already are.

You can write python bindings for basically everything, everything is python-adjacent.

nickysielicki··on Deno has made its PyPI distribution official
For better or worse, pypi is the executable distribution mechanism of the future.

Other cool tools you can install from pypi:

1. https://pypi.org/project/cmake/

2. https://pypi.org/project/ninja/

3. an entire c/c++/zig toolchain: https://pypi.org/project/ziglang/

4. the nvcc cuda compiler: https://pypi.org/project/nvidia-cuda-nvcc/

nickysielicki··on When compilers surprise you
Yikes, gross. That’s like an option of last resort IMO. I’d rather maintain the clean loop-based code unless I had evidence that the compiler was doing the wrong thing and it was in my critical path.
← PreviousPage 4 of 34Next →