HNHacker News
TopNewBestAskShowJobs

SilasX

17,815 karma · joined December 6, 2011

blog: http://blog.tyrannyofthemouse.com

Github: http://github.com/SilasX/

email: sbarta@gmail.com

Location: Austin, TX (left the Bay Area August '18).

Studying infosec to make a new career out of it.

[ my public key: https://keybase.io/silasbarta; my proof: https://keybase.io/silasbarta/sigs/EIsQUp3Ln8z5fOgzWFj9alDI8Ci9Vl4JSgP5OZQcmjU ]

submissionscomments
SilasX··on Robert Hooke's "Cyberpunk” Letter to Gottfried Leibniz
>It turns out that, aside from their common interest in antagonising Isaac Newton, Hooke and Leibniz also shared an interest in mechanising scientific reasoning through the invention of a universal language for science. Leibniz called his project the "Characteristica Universalis".

I'm surprised it doesn't mention Leibniz's famous "Let us calculate" quote:

>>In a 1679 letter to one of his patrons, Johann Friedrich, he described his project of the universal language as “the great instrument of reason, which will carry the forces of the mind further than the microscope has carried those of sight”. Later he wrote:

>>>The only way to rectify our reasonings is to make them as tangible as those of the Mathematicians, so that we can find our error at a glance, and when there are disputes among persons, we can simply say: Let us calculate, without further ado, to see who is right.[1]

I'm also kind of surprised that Hooke wrote the letter in English, since I assumed all academic communication across language barriers back then would have been in Latin. But ChatGPT tells me Leibniz was unusually multi-lingual.

[1] Sorry, ad-heavy site but I wanted one that gave context: https://publicdomainreview.org/essay/let-us-calculate-leibni...

SilasX··on Rouille – Rust Programming, in French
>On the other hand, ".gouv.fr" is something used in France. gouv[ernement] is completely different than go[u]v[ernement]

How are they different? Contractions and abbreviations drop letters. That's the point. .gov would have been perfectly fine and matched other countries. It's a clear example of being different for the sake of it.

SilasX··on Rouille – Rust Programming, in French
For another example of this, there was BritCSS someone made that lets you use British spellings in CSS:

https://github.com/DeclanChidlow/BritCSS

SilasX··on Rouille – Rust Programming, in French
Even so, it parallels a real thing that happens in non-joke contexts, where they avoid valid French when it looks "too English":

https://news.ycombinator.com/item?id=45773624

SilasX··on Rouille – Rust Programming, in French
Haha there are other cases where there is valid French that isn't accepted in French speaking areas because it looks too similar to English.

1) Quebec wanted "arrêt" instead of "stop" on stop signs, even though the latter is accepted as valid French and used in France.

https://en.wikipedia.org/wiki/Stop_sign?utm_source=chatgpt.c...

2) The use of the TLD .gouv.fr instead of .gov.fr, even though "gov" is a recognizable contraction of the intended French word "gouvernement".

(No, it's not a valid defense that "'gov' would be pronounced differently from 'gouv'": the English TLD .com is a contraction of "commercial", even though the "com" in "dot-com" is pronounced differently from the "com" in "commercial".)

SilasX··on Using AI to negotiate a $195k hospital bill down to $33k
Oh wow, I had independently formulated that as the duality/parallel between corruption and cheat codes: "The more out-of-band information you need in order to be treated normally/fairly, the more corrupt the system is."

https://news.ycombinator.com/item?id=9933801

SilasX··on Palisades Fire suspect's ChatGPT history to be used as evidence
Right, for the OpenAI case to be analogous, they would have to switch to a system where your chats are homomorphically encrypted -- i.e. OpenAI does all its operations without knowing either the input or output plaintext. In that case, they'd only have encrypted chats to begin with, and would have to somehow get your key to comply with a warrant for the plaintext.

And note: the above scenario is not likely anywhere in the near future, because homomorphic encryption has something like a million times overhead, and requires you to hit the entire database on every request, when state-of-the-art LLM systems are already pushing the limits of computation.

SilasX··on Why do LLMs freak out over the seahorse emoji?
That’s just hacking.
SilasX··on US government shuts down after Senate fails to pass last-ditch funding plan
A true government shutdown, yes. But here, all the nation-state-defining stuff (military, border control, police) stays up, and only parts of it shut down.
SilasX··on Boeing has started working on a 737 MAX replacement
But the intended insight isn't stupid, and "the exception proves the rule" is a natural, easily-inferred contraction of "the rarity of finding an exception proves the general validity of the rule".
SilasX··on Boeing has started working on a 737 MAX replacement
Yes, but in common usage it has come to also mean "the [rarity of finding an] exception proves the [general validity of] the rule", and it was clear from context which one the parent meant.
SilasX··on Hackers strike Harrods in latest UK cyberattack
Interestingly, it's common in the Midwest to add "'s" where it doesn't exist when referring to a brand e.g. calling Costco "Costco's".

https://styleblueprint.com/everyday/why-do-people-add-s-to-t...

SilasX··on Help us raise $200k to free JavaScript from Oracle
>I think the order here is reversed: If you ask the public for money, it's fair to perform a public service. If you just do something you wanted to do anyways, and probably would have done anyways, then it might be viewed as less-than-charitable to ask others for money to help you achieve your goal for yourself (even if other people might benefit somewhat too). Especially when you are far richer (like 100+ times richer) than the people you're asking for money.

I get the where you're coming from, but it's this exact attitude that ends up with critical infra like OpenSSL being maintained ad hoc by some devoted geek for a pittance, who inevitably can't keep up with critical patches.

SilasX··on Wind turbine blade transportation challenges
1) I was curious why they can't just attach two partial blades onsite to make a longer one, and the article makes some attempt to address it, so, to save you from reading the whole thing:

>Shipping them in multiple pieces and reassembling them on-site won’t work because the joints would create weak spots. Junctions would also add too much weight compared with that of blades made from single pieces of polymer, says Doug Arent, executive director at the National Renewable Energy Laboratory Foundation and emeritus NREL researcher.

>“It comes down to the stress engineering of the components,” Arent says. Blades could one day be 3D-printed on-site, which could negate the need for an airplane, but that research is still in early stages, he says. (Lundstrom says 3D-printed blades will never happen, since it would require a large, sophisticated manufacturing facility to be built at every wind farm.)

2) I'm also curious if anyone has done the numbers on how long it takes these large turbines to pay back the energy cost of flying them there? You would have to a) find out how much more energy they make from the same footprint compared to smaller wind turbines, and b) how much more energy it takes to fly them there compared to transporting the smaller ones (and I'd be curious about a smaller plane vs ones that can be transported on the ground).

SilasX··on Anthropic judge rejects $1.5B AI copyright settlement
I think you're focusing too much on whether LLMs are "really human-like" or not. That's a distraction, and I shouldn't have made reference to it. Let me zoom out to a broader point:

It has never been a part of copyright to include the right to be influenced by the copyright work. Period. It's been the diametric opposite. Copyright as always existed, and been justified, as a way to get good, new works out into the public, so that later works can be influenced by them. The fact that one work was influenced by another has never, by itself, been a reason to consider it infringement. Not until 2022, when AIs actually got good at it.

When you argue for AI training as copyright infringement, you're saying that "the fact that your work was influenced by previous works, means you owe license fees". This is wholly without precedent[1], and was widely rejected until the moment some activists realized it could be a legal tool against Bad People. It's a Pandora's Box no one really wants (except perhaps very large media companies who will be able to secure general "learning licenses" for mass libraries of works). That was the the point emphasized in my original comment: If Anthropic is infringing because they base new works on old ones, so are you. You too owe licensing fees for every work you observed that fed into how you create. If that feels like an expansion of what copyright is supposed to cover ... that's the point.

For every single work of literature, you can go back and say "aha, this is clearly influenced by X, Y, Z". Precisely zero people were going out insisting that the author therefore owed fees to all those other creators, because the idea is absurd. Or was, until 2022, when some people needed a pretense for a conclusion they long supported for unrelated reasons ("Facebook must suffer"). So I think my second paragraph is justified.

"If you read 100 horror novels and write a new one based on everything you've noticed in them, you don't owe the authors jack squat. But if you have a machine help you compile the insights, suddenly, you've infringed the authors' rights." Yeah, you do need to justify that.

[1] I agree there are going to be cases where it, say, was captured too closely, but not for the general case, and it's further weakened when it's "imitating" a thousand styles at once.

SilasX··on Anthropic judge rejects $1.5B AI copyright settlement
Because the onus is on you to show the substantive difference. Learning from copyrighted works has always been accepted as free and unrestricted, up until 2022. Before that, nobody (to a rounding error) thought that simply being influenced by a previous copyright work meant you owed a license fee. If anything, people would have been livid about restrictions on their right to learn.

Only when big-corp critics needed another pretense to support a conclusion they long agreed with for other reasons, did they decide that the right to learn from your exposure to a copyright work was infringement.

If you're interested in the similarities and genuinely curious, you could look at the article linked above, which shows how both LLMs and humans store a high-level understanding of their training set. It's a way deeper parallel than "it's all learning" -- but you have to be willing to engage with the other side rather than just strawman it.

SilasX··on Anthropic judge rejects $1.5B AI copyright settlement
Right — laws can be arbitrary, and ignore constraints like consistency! It’s just something sane people try to avoid.
SilasX··on Anthropic judge rejects $1.5B AI copyright settlement
What is that replying to? I don’t see the relevance to my comment.
SilasX··on Anthropic judge rejects $1.5B AI copyright settlement
Be careful what you wish for.

While I'm sure it feels good and validating to have this called copyright infringement, and be compensated, it's a mixed blessing at best. Remember, this also means that your works will owe compensation to anyone you "trained" off of. Once we accept that simply "learning from previous copyrighted works to make new ones" is "infringement", then the onus is on you to establish a clean creation chain, because you'll be vulnerable to the exact same argument, and you will owe compensation to anyone whose work you looked at in learning your craft.

This point was made earlier in this blog post:

https://blog.giovanh.com/blog/2025/04/03/why-training-ai-can...

HN discussion of the post: https://news.ycombinator.com/item?id=43663941

SilasX··on I'm absolutely right
Sounds like a kind of Schelling point:

https://en.wikipedia.org/wiki/Focal_point_(game_theory)?uses...

SilasX··on Meta accessed women's health data from Flo app without consent, says court
> The correct answer is to look at the data and verify it's legal to use.

So repeal Section 230 and require every site to manually evaluate all content uploaded for legality before doing anything with it? If it’s not reasonable to ask sites to do that, it’s not reasonable to ask FB to do the same for data you send them.

Your position seems to vary based on how big/sympathetic the company in question is, which is not very even-handed and implicitly recognizes the burden of this kind of ask.

SilasX··on Meta accessed women's health data from Flo app without consent, says court
>I don’t like the analogy because “hosting an event” is a fuzzy thing. If you are hosting an event with friends you might be able to rely on the shared values of your friends and the informal nature of the thing to enforce this sort of norm.

You can't, though -- not perfectly, anyway. Whatever the informal norms, there are going to be people who violate them, and so the fault shouldn't pass on to you when you don't know someone is doing that. If anything, the analogy understates how unreasonable it is to FB, since they had an explicit contractual agreement for the other party not to send them sensitive data.

And as it stands now, websites aren't expected to pre-filter for some heuristic on "non-consensual user-uploaded photographs" (which would require an authentication chain), just to take them down when informed they're illegal ... which FB did (the analog of) here.

>If you are a business that host events and your business model involves photos of the event, you should have a professional approach to knowing if people consented to have their photos shared, depending on the nature of the venue.

I'm not sure that's the standard you want to base this argument on, because in most cases, the "professional approach" amounts to "if you come here at all, you're consenting to be photographed for publication, take it or leave it lol". FB had a stronger standard than this.

SilasX··on Meta accessed women's health data from Flo app without consent, says court
Yeah, I'm not sure if I'm missing something, and I don't like to defend FB, but ...

AIUI, they have a system for using data they receive to target ads. They tell people not to put sensitive data in it. Someone does anyway, and it gets automatically picked up to target ads. What are they supposed to do on their end? Even if they apply heuristics for "probably sensitive data we shouldn't use"[1], some stuff is still going to get through. The fault should still lie with the entity that passed on the sensitive data.

An analogy might be that you want to share photos of an event you hosted, and you tell people to send in their pics, while enforcing the norm, "oh make sure to ask before taking someone's photo", and someone insists that what they sent in was compliant with that rule, when it wasn't. And then you share them.

[1] Edit: per your other comment, they indeed had such heuristics: https://news.ycombinator.com/item?id=44901198

SilasX··on Monero appears to be in the midst of a successful 51% attack
But states generally like having a financial system, and don't like (or are at least annoyed and worried by) cryptocurrencies, so the incentives aren't the same.
SilasX··on Monero appears to be in the midst of a successful 51% attack
Reminder: if you want to bet on an asset's demise (i.e. short it), you don't need a derivatives market, you just need to be able to borrow the asset and sell it. So you could accomplish the goal there by borrowing Monero and converting it to USD. A lot of smartcontract platforms let you do this -- including on other chains, where they hold a token convertible into the original chain's native unit.

I bring this up because people are always asking what platforms are allowing me to short cryptocurrencies, which seems to miss that it's enough to just have a debt denominated in what you want to bet against.

SilasX··on Let's properly analyze an AI article for once
I would also include:

5. Debug why the API's don't behave as documented and figure out a workaround.

SilasX··on GPT-5
Heh, I had this dream/nightmare where I was typing on a laptop at a cafe and someone came up to me and said, "Oh neat, you're going real old-school. I like it!" and got an info dump about how everyone just uses AI voice transcription now.

And I was like, "But that's not a complete replacement, right? What about the times when you don't want to broadcast what you're writing to the entire room?"

And then there was a big reveal that AI has mastered lip-reading, so even then, people would just put their lips up to the camera and mouth out what they wanted to write.

With that said, as the owner of tyrannyofthemouse.com, I agree with the importance of the keyboard as a UI device.

SilasX··on GPT-5
I just got the same thing in the US too. (Am on the $20/month subscription.)
SilasX··on The Math Is Haunted
>Let's say our overall goal is to get to work showered, wearing clean clothes, and having brushed our teeth. After we showered and put on clean clothes, wouldn't you say that the remaining goal is to brush our teeth and get to work?

Yes, I agree! My issue is not with the reality of having reduced the problem to a different one, but with the "goal" metaphor, which is counterintuitive to how that term is normally used, especially since it connotes that each step is some brand new thing, rather than progress.

And sure, like any technical jargon, you can learn to accept some specialized term-of-art usage, but I'd prefer it be conducive to developing a correct ontology of what is going on in an assisted Lean proof.

While "subgoal" is better, I think the issue even goes back to calling the original theorem a "goal". Really, the goal is to prove the theorem. So a better term, more conducive to intuiting the Lean ontology, would be something like "[remaining] unresolved proof state", even though it's more verbose. (Perhaps you could get the best of both worlds with some kind of footnote for new users that clarifies that this is what "goal" means here.)

With that said, I appreciate the pushback, as your comment forced me to more carefully think about what I was objecting to.

SilasX··on “No tax on tips” is an industry plant
I often see the claim that it would be above board/legit/not a circumvention of employment laws if they ran a bidding system like that. But ... that's exactly how Sidecar worked (Uber competitor, shut down Dec 2014), and I never saw anyone distinguish them from Uber in their criticisms (until people forgot that model existed).
← PreviousPage 5 of 34Next →