HN

Comments (144)

• Menu

  aftbit
  The same argument applies to DRM and piracy. DRM only disadvantages people who want to consume content legally. Only one person has to bypass it (be it through cryptographic or software failings or the good old fashioned analog hole ala telesync) then upload the resulting unencumbered media file to a piracy site.As a result, it's not just cheaper, but also easier and more reliable to steal media than it is to buy it legally.What would happen if media were offered without DRM? Everything would be on a piracy site? It already is.
• Menu

  deyiao
  It's unfortunate that the final few years of the fastest AGI progress happen to coincide with a global surge in nationalism, polarization, and mutual hostility between human groups.That raises a disturbing failure mode: under intense conflict, an AI could learn that exterminating some groups of humans is a justified or even desirable objective. Once that principle is accepted, the step to concluding that exterminating all humans is justified may become much smaller than we'd like to believe.
• Menu

  jameshart
  Slight timeline correction: 25 years ago in 2001 Blowfish was no longer contraband - the crypto export rules had been changed by 2000. As of January 2000 Netscape and IE were both shipping 128bit DES internationally.Still, the t-shirts with Blowfish source printed on them were a cool bit of retro hacker chic.
• Menu

  bonoboTP
  This is AI written. Next time just paste your prompt or conversation with the AI that wrote the article. I'll then converse with Claude about it myself. There is no need for this intermediary.
• Menu

  aetherspawn
  It’s kind of a bit ridiculous for Opus to blow up the minute I ask it to read from Ghidra, when the Chinese models, and even OpenAI, do not.Leave the legality up to me, please, and don’t presume to be a power tool vendor regulating the use of their power drills, so to speak.
• Menu

  no_multitudes
  It seems like you got bored halfway through and had an LLM write the second half of your post? You know there isn't a minimum length for blog posts, right? You can just stop writing.
• Menu

  theturtletalks
  We need a cypherpunks movement for open weight models
• wei_b0
  I'm 27, and I genuinely had no idea what had happened in cryptography 25 years ago. This was a really refreshing and insightful read - thanks for sharing.
• zkmon
  I guess the author (who is familiar to me) missed the common theme behind these two things. It's use in military systems. So no surprise there. Right from the days of Enigma machine to AI drone, the same restrictions were used many times.Germany or Europe is no different.
• Menu

  yalogin
  With these models, what is the moat? Is it the architecture itself or the training data? Why can’t China create these models themselves? It just feels like theater to restrict models. Can someone tell me why I am wrong if I am?
• Menu

  Scoundreller
  Time to create some esp32-in-a-lightbulb open weight model dead drops.Please don’t open-circuit the light for several hours.
• markmental
  I feel like eventually we're going to end up just like how the cryptography restrictions were in the 90s. They might have seemed necessary back then, but now looking at it, the technology and AI is probably going to become so ubiquitous in the next 10 years that we're going to wonder why we even restricted it in the first place, when it's something that's just going to be everywhere. Like they said at the end of this blog: "What arrived in my letterbox after two weeks on a CD can be downloaded today in fifteen minutes."
• heisenbit
  The difference with weights is that unlike in the past one can not fit the secrets on a T shirt print.
• Cider9986
  Apparently we are still fighting for secure communication through cryptography [1].Other issue that doesn't get enough attention on HN is the criminalization of writing code. Probably because it was cryptocurrency applications. First they come for the people you disagree with. Just like they first came for porn sites, and now Wikipedia is being monitored.[1] https://fightchatcontrol.eu/
• Menu

  class3shock
  The framing of this entire issue is just wrong."Today the reflex is back."To do what? Ban export of AI? Are you sure? Because it seems, so far, there has been some personal motivated government restrictions due to people not bending the knee and industry moves and clamoring about dangers to generate hype."The fears have changed."Yeah, people are fearful that the entirety of the internet has been hoovered up and people are using that data to make tech that they are promoting by saying it will eliminate a huge percentage of jobs and is exceedingly dangerous. At the same time as they do that, they refuse to take any accountability for the negative effects their tech has or will have, avoid any over sight, and laugh at the idea of discussing that maybe such a consequential tech doesn't belong in the hands of a few tech bros whose only qualification for owning it is being rich."The worry is now cyber capability, biology and models that do things nobody asked them to do."No, we are being sold those worries (really, really hard) but these are not "peoples" worries. They aren't even real governments worries. They are the worries being peddled to try and work some deal where the taxpayers end up holding the bag of hot air that is the AI economy.This post is so frustrating because AI is the antithesis of everything the people fighting for cryptography access, working on opensource, pushing for a the sort of web that is beneficial to all, were working against. A giant concentration of power enriching the few at the expense of many.
• Menu

  SpicyLemonZest
  I don't understand what we're talking about here. The cryptography debate was won because the government was wrong. PGP for example is not dangerous, it's a safe and well-behaved piece of software that prevents certain messages from being read by people other than their intended recipient. If PGP had "escaped containment by finding a zero-day in a package proxy and reached production infrastructure at Hugging Face, exploiting additional zero-days along the way", everyone including Phil Zimmermann would have immediately agreed that this is a dangerous product which needs to be handled carefully.
• esafak
  Cryptography is nothing like AI. It does not have the ability to upend the economy, and enable people to launch cyberattacks, etc.
• rhelz
  (in a Forrest gump accent) my momma said that locks on doors only keep honest people honest. I'm not Forrest Gump, but my mom actually did say that and it's true.What ever we have can be taken from us by powerful people, even life itself. In point of fact, that's exactly how these weights were created in the first place--they scraped the intellectual property of the world, and now are trying to restrict access to what they themselves just took.That being said, I don't think any of the AI billionaires who are talking about this have taken the locks off of the doors or gotten rid of their security systems.
• lofaszvanitt
  Maybe it's already out there.
• js8
  Even if we look at crypto analogy, US government actually created an open standard of DES and AES, so they knew the argument. Maybe the crypto export controls were just a protectionist play with a fake explanation of national security. (In that case it's useful because as a commercial subject you have to follow the law.)Anyway, I think open weight models will win. I am strong believer that AI (even the frontier) can be done just as "deductive core, encyclopedic shell". That is:- Deductive core is a small number of basic primitives (in an LLM probably less than couple thousands, like words in simple english) which encode the human common sense reasoning logic and epistemology. These might correspond to the elements of the embedding vector as well.- Encyclopedic shell is a collection of many facts about the world (most importantly their definitions) expressed in the language of deductive core. The new facts can be deduced by applying the rules from deductive core to the encyclopedic shell; that's what we call "model reasoning".Deductive core, encyclopedic shell is not a new concept. Any programming language has primitives and then a library of useful functions. In Lisp, the deductive core is special forms; in Haskell, it is Haskell Core (based on HM typesystem); in Lean, the core is based on Calculus of Constructions; in Prolog, the core is mostly the resolution algorithm. Even in predicate logic you have axioms and then you build a library of useful tautologies from that.I also believe that current AI (large reasoning models) deductive cores suck quite badly. They have been generated by an evolutionary algorithm as an emergent property of LLM pretraining, and are probably order or two magnitude larger than actually required. I believe hand-crafted deductive cores will win in the long term.Now, through this lens, what is there to be export-controlled (or just controlled in general)?You can't control the encyclopedic shell. It's like Wikipedia or other books that have already been published. It would mean closing down research and STEM publishing.I also don't think you can control the deductive core. It's just an algorithm (a language interpreter), and desirably, a small one. Any country with a rich enough history of philosophy can probably recreate one (such as China). Moreover, I believe many open source models already contain decent deductive cores so that fight has been lost as well.The only reason we have this discussion is that LLMs are shrouded in mystery, because their vast knowledge (the encyclopedic shell) is written in an unaccessible language (the deductive core) of their weights.Every couple years, capitalism tries to reinvent itself by creating a new type of property. Yesterday it was "big data", today it's "model weights". Don't fall for it; it's an enclosure of publicly available knowledge and our ability to use it.
• Menu

  IshKebab
  Yeah this persuaded me more than I thought it would, but there are still some pretty big differences IMO. A frontier AI model is a lot harder and more costly to produce than an encryption algorithm. Especially if we believe Anthropic that the Chinese can only do it by distilling their models (though tbh I would err on the side of not believing that).
• liesliy
  [flagged]
• 1saadcodes
  [dead]