Hacker Newsnew | past | comments | ask | show | jobs | submit | more juitpykyk's commentslogin

You're using the wrong CPU.

Consumer AMD 7950X supports AVX-512, it's faster than M2 Ultra at half the cost.


I didn't see benchmarks that suggest the 7950X is faster than M2 Ultra. I only saw performance numbers for 7995WX which has 6x the cores and 6x the cache.

Either way, I think these comparisons are moot since an M2 Ultra comes with 2x M2 Max GPUs and an NPU and up to 192GB of unified memory running at 800GB/s. In other words, you wouldn't want to run your LLM on the CPU if you have an M2 Ultra.

The point of OP is to increase LLM performance when you don't have a capable GPU.


FBI could subpoena GitHub for IP addresses.


They probably have/will, though we are unlikely to find out what they get unless they manage to arrest/charge/try him.

Jia Tan probably used a vpn though - we know that they did for accessing IRC (source: https://boehs.org/node/everything-i-know-about-the-xz-backdo...)


Most (but not all ) VPN providers keep logs and payment info that are subpoenable. You could use something like Mulvad with Lightning Network payments, but I am not sure that even that is fully anonymous.

The Witopia VPN that he used for IRC [1] is US based: https://www.personalvpn.com/contact-us/ and they don't mention neither LN payments nor not keeping logs.

1. "~jiatan@185.128.24.163" https://boehs.org/node/everything-i-know-about-the-xz-backdo...


If he used a well known vpn, he probably used a fake id and stolen credit card to pay it.


I had a hard time choosing which comment to select to reply, so I chose yours since it's higher up. Apologies if it's irrelevant.

I don't know why most people assume that hackers even bother with stolen credit cards in the first place. I mean, they sure do, but those are your average Joes in the business of refund reshipping and other types of scams.

Those who want the maximum anonymity don't even bother with buying anything. It's as simple as going to one of the popular websites who leak databases, setting up OpenBullet software or spending anywhere from 1 to 5 hours writing custom mail:pass validators to spam requests to either API or login form through (once again) leaked proxies, etc. using leaked credentials. Or simply going into one of those threads titled 'x100 Mullvad accounts" which have already validated accounts with anywhere from 1m pre-paid to multiple years. And there's even a bonus of not being shown as a user of this account if you do not use official App and simply load configuration manually through ovpn, etc.

And then there's proxy-chaining if you're doing something truly nefarious. It's super easy to chain multiple VPNs with few socks proxies.

People behind XZ backdoor to me look much more smarter than myself, so I would bet they took care of this angle and will be untraceable.


But well-known VPNs mostly keep IP logs - I know from experience; in my company the FBI found a DDoSer this way.


Mullvad accepts cash in an envelope with no return address, good luck tracing that


Postal mail has non-zero metadata, e.g. origin can be traced at least to departure postal code.


There's... actually very little that stops you from sending mail from a non-local postal code.

I've occasionally sent packages postmarked as being from one zipcode from another; as long as it's in the same region, much of the postal processing doesn't care so much.

There's also remailers and forwarders.


How do you do that? Walk up to the postal counter, ask them to postmark it, then ask for it back, drive to another post office and slip it in their outgoing pile?


Semi-Presorted post can be acquired pretty easily. Places like Shippo and PirateShip offer it. As long as you put enough money into the postage paid, the post office doesn't really give a shit if it comes out of somewhere weird. There is no requirement that "return address" and "sent from" area are the same so long as they're within the same postal zone.

This is why AMZN packages have a return address in Vegas or similar sometimes.


You seem to assume Mullvad stores something that allows correlating a Mullvad account to a specific incoming envelope.

Or that they store something that allows correlating an IP+time to a Mullvad account.


Correlation is about combining metadata to incrementally narrow datasets.

When a VPN provider isn't cooperative, metadata can be sought upstream.

Mullvad deserves much credit for accepting non-digital payments, which increases the cost of deanonymization.


Departure postal code could be an area with like 1 million people. And if it is a nation state I am sure they could send the mail from another country.


Most populous US zip code is less than 150K people, https://worldpopulationreview.com/zips


I mean, even if I live in Manhattan, I can easily take a subway train to an area with a different zip code and mail it there. And that's almost the easiest thing you could do -- if you want to hide where the mail is from, it is trivial.


Dense urban areas are often blanketed by a range of spectrum sensors for the purpose of retroactive correlation (e.g Palantir) with other metadata sources.


Considering what the account was up to I sincerely doubt it was being used without Tor or a VPN.


This article asserts some opsec failures - time zones switching when they shouldn’t etc.

It’s quite plausible that they didn’t manage perfect vpn usage every single time.


Expert level opsec - very rarely make deliberate "mistakes" by having an inconsistent timezone, or tunneling through someone's compromised home device instead of a VPN, to throw adversaries on wild goose chases.


They also committed with a different email address and a different name once or twice. This may have been intentional, though, to misdirect.


I'm surprised they didn't set up guards to prevent mistakes like commits with the "wrong" timezones...


Probably; but ask Ross Ulbricht how easy it is to screw that up.

The thing is you only need to make a mistake once.


The ssh public key might also be interesting, given the opsec failures, they might have used it for other accounts / at other providers


Not going to be useful if they consistently used a VPN to access GitHub. But people make mistakes sometimes.


history shows that it is generally very hard to not slip up here and there, especially if you are not expecting to be a huge target.


If a US gov. agency is "Jia Tan" then this might not happen.


But Sir The Calls Are Coming from Inside the House!


this [1] but with handcuffs

[1] https://i.redd.it/obvqaa0lhn841.jpg


I learned that AMD GPUs have layers upon layers of drivers, in user space, kernel space, and drivers running on the device itself.


And they didn't really open source anything. I thought they were better than Nvidia in that regard, but they're not.

They only open sourced the API. It's still closed firmware just like Nvidia.


Isn't HDCP/DRM implemented in ASIC hardware? There is nothing programmable there.


It's probably implemented mainly on the Platform Security Processor.


Neuron's DNA is not passed down, it would be quite logical for evolution to use neuronal DNA for weight storage.


Many nervous system behaviours do appear to be passed through genetics - for example, the ability to breathe, the reflex to avoid pain, etc.

I suspect in the future we might find mechanisms beyond simple natural selection that allowed those mechanisms to get encoded in genetics.


I think you're talking to the wrong point. These memories aren't being encoded in germ cells, they are after the fact changes to DNA in mature neurons which have completely differentiated. I would think it's very possible at that stage of development for them to add or remove segments of DNA in order to encode new information not related to the development of the cell as long as it didn't interfere too much with parts that are actively used for the ongoing upkeep of cell activity. It would need to alter how the cell functions a little bit for the changes to modify the neuron's ability to process signals though.


I should note that studies have demonstrated that bacteria who have been modified not to be able to consume lactose will develop mutations that allow them to consume lactose again much more quickly than would be expected given the number of bacteria, the rate of random mutations and the size of the genome. It has been hypothesized that there is a cellular mechanism to control which portions of DNA are easily mutable, possibly through a combination of chromatin structure, epigenetic modification and changes to the local chemical environment via metabolism.

This mechanism might exist in a scaled up form in humans.


Isn't that what happens in antibody germinal centers?

https://en.wikipedia.org/wiki/Somatic_hypermutation


bacterial plasmids are a common form of horizontal gene flow between individuals


Could they be transferred while in the womb?


>> it would be quite logical for evolution to use neuronal DNA for weight storage.

To pass that down you'd have to replicate the connectivity of the network for the weights to be relevant right?

Related: The article doesn't say which DNA areas are broken and repaired. Nor does it say if they are modified. It seems like encoding weights in DNA would make them more robust but harder to change. If so, there should be a particular region where this is happening. Maybe there's a mapping between certain DNA areas and each synapse. That'd be really interesting.


Independent of this breaking/fixing, it's already known that DNA near the synapse (not necessarily in the neurons Soma) is modified via epigenetics to sustain the synapse at the new level.

So yes, DNA epigenetic changes near the synapse are a key part of maintaining the "weight" or strength of that particular connection. ("key part" phrase because there is a lot of complexity and they haven't nailed it all down, there could be other "key parts").


>> DNA epigenetic changes near the synapse are a key part of maintaining the "weight" or strength of that particular connection.

What do you mean by "near the synapse"? Is there DNA outside the nucleus or something? Is there DNA that maps (corresponds to) the synaptic pattern of the neuron?


Yes there is DNA outside the nucleus. The DNA near each synapse gets modified (epigenetic) based on activity in that physical area so it can produce the proper proteins to preserve the state of that synapse over the long term.

Edit: the DNA is in Synaptic Vesicles


Neurons are not on the germ line, whatever happens to their DNA is not passed down to your children.

There was another article in the recent years about neurons using RNA or DNA for storing information related to their activation patterns.


The baby is connected to the mother's placenta for months, maybe information could be transmitted then. I've never heard anything to support that idea, though!


This always seemed like one of those little biological details, like the well known example of that nerve which loops all the way down a giraffe's neck and back again in order to connect two regions only a few inches apart, that shows that nature doesn't refactor.

Because it seems like such a waste of the opportunity afforded by extended physical secueity and direct connection between mother and developing child, that some means of transferring a portion of the mother's learned knowledge, or at least some coarse grained abstraction of it, to the fetus, has never developed.

The lazy dismissal of this question is just to say, if nature needed it, it would have evolved it, but this doesn't seem to hold in every case [0]. It seems rather that there was no way for such a capability to be built out of extending existing mechanisms, with the major barrier being the absence of nerve tissue in the umbilical cord, where higher level CNS connectivity might have evolved from as a foothold

[0] and certainly doesn't account for what may happen in the future unless nature is completely done developing everything that could be developed. Nor does it incorporate the idea that human manipulation of our own biology is not itself also part of nature.


Wellcome. Sometimes it may happen that familiar stem cells cross maternal-fetal barrier in placenta, persist somehow and start to function regardless, where stem cells are needed - usually in younger sibling coming from the older, in place of original cells, even in the brain - forming part of it as of another person (more or less) - interconnected but not the same..

The Most Mysterious Cells in Our Bodies Don't Belong to Us https://www.theatlantic.com/science/archive/2024/01/fetal-ma... ( https://news.ycombinator.com/item?id=38861497 )


In addition, most parts of the first cell of what will become a baby, come from the mother. This includes all DNA in mitochondria and another organelle that I don't remember the name.


epigenetic inheritance is real


I wonder if that means that each neuron could act as a mini turning machine


Check out Michael Levin's work, who's done some experiments with skin cell and has shown that they learn to do stuff.


stop wondering and look deeper, youve bumped into the begining of an incredible journey. even individual protiens, exhibit rule paring.


My equivocation was to avoid downvote brigades that hit me last time I posited this same idea on HN.


For scale, Microsoft requires 40 TOPS to certify a (consumer) computer as Copilot capable.

https://www.techpowerup.com/320933/microsoft-copilot-to-run-...


I've once got 000000. If you do the math it takes about 1 year to cycle through all million of them, assuming no duplicates.

Which makes me wonder, what's the expected repeated distribution?


This is the birthday paradox. The rough approximation is you need sqrt(n) values to get to a 50% chance of having a duplicate. Sqrt of a million is a thousand, and if they're every 30 seconds, that's ~8 hours or so. So you probably get a duplicate or 2 every day.

There's _much_ better approximations than the sqrt one, but I don't know them and the actual math is too hard.


Correct, the article actually mentions the birthday problem somewhere. Even with 10k codes (2 days) you get dozens of duplicates, so I'd say it's potentially even more common.

Sextuples are 1 in 100,000, so something like every 50 days (per account).


> Even with 10k codes (2 days) you get dozens of duplicates, so I'd say it's potentially even more common.

Yeah I screwed up here, at the end:

> Sqrt of a million is a thousand, and if they're every 30 seconds, that's ~8 hours or so. So you probably get a duplicate or 2 every day.

You'd get 1 or 2 every day (approx.) if you're only, at any point, looking at the collection of codes generated in the past 8 hours. But of course that wasn't the question, and the odds go way up once the period we're looking at goes larger and larger over time.

I shouldn't have tried to go beyond "ballpark if you wait 8 hours you have a coinflip of having at least one duplicate", anything more than that requires different math.


There was an article quoting someone close saying he wants to be President and world first trillionaire.

Control of AGI could certainly give you that.


> for purposes of providing the Services to you

This seems the key part


I am not a lawyer but I have seen startups distort/rationalize legal language as their tech services evolve to grandfather new situations into old language.

I don’t know if vultr language is worse than others, but my concern would be that someone selling you out can squeeze a lot in that clause for a long time, particularly if you never find out. Arguably that’s in bad faith, but…

Say that to provide the Services to you, vultr has to supplement its income by (old school) selling your videos to a dvd publishing company, or (newer) creating their own streaming tv channel, or providing them to an AI model training company, or providing them to an “affiliate” advertising-serving broker who slurps your created content and slaps one or more segmentation labels about your content (“kink”, “religion(X)”, “gamer”) tied to your email which it then resells to world+dog?

Ie is selling you out part of what vultr needs to do to provides the Services to you?

I find it very hard to trust companies based solely on their legal language when that language is viewed from an adversarial position. But I am not lawyer to know what kinds of “misreadings” are “beyond the pale”/not legally defensible.


"an adversarial position" is the only position you should assume when interpreting legal texts. After all, if push comes to shove, your the actual adversary. And in any other case the legal text is not needed.


Assume that immediately after executing any contract, the counterparty will be purchased by the Devil. Draft accordingly within reason.


They want permission to commercialise my content "for purposes of providing the Services to" me?

First time I hear such requirement


They offer services like store fronts. This might require them to sell your stuff and accept credit cards on your behalf.

https://www.vultr.com/marketplace/apps/woocommerce/


Ah I was not aware of this, I guess it makes sense then? They could make this section of the ToS specific to their marketplace/store front products


That’s an easy one: the company exists to provide The Services, revenue from commercialising the content supports the company.


That's the basis for any business model where you are the product.


> That's the basis for any business model where you are the product.

True but most times, when you are the product, the service is free. In this case you pay for the service.


They do offer relatively inexpensive solutions, though. And LinkedIn is a good example of a business whose revenues are largely made from sharing and harvesting data from both paid and free users for the benefit of some of those paying users, and some third parties, too.

Vultr is just even cheekier than LinkedIn.

Who's to say if they'll actually act on this, but them setting themselves up to legally do this is all a bit gross.


Then look at the TOS of Whatsapp, Facebook, and Instagram.


Yeah sorry I meant in the context of cloud providers


This verbiage is standard for almost all internet/service providers, it's language to allow them to display your content on their behalf.


Does it change anything? Paraphrasing:

"Vultr [will own] [all of your] User Content [and do whatever Vultr wants with] the User Content [...] for the purposes of providing the Services to you."

You could read that as: "if you want to work with us we will own all of your user content".


The question is: how would your lawyer read it.


They don't need a licence, and a very wide one, to do this.


That's probably not a legal opinion they paid for.


Knowing lawyers, it was probably sloppily copied from another ToS they (or not even them) wrote.


So they have a perpetual right to your contend, for as long as they are providing you a service.

Either way you read it, it seems like poor wording.


Perpetual means continuing forever, so why would it end at the end of the services?


If you ask your parents if you can stay up late to finish your homework essay, it should follow that you only gain that right until the essay is finished.

If you ask if you can stay up late for the rest of your life, it should follow that you gain that right for the rest of your life.

If you ask for both at the same time, in the same sentence, you might grow up to write TOS for vultr.


"Selling copies of your genome to partners enables us to affordably provide the service, therefore you agreed to it."


Russia foreign minister (Zaharova) continuously asks for Assange to be released, mentioning "freedom of speech".

Assange has building size graffiti murals in Moscow.


Zaharova is director of the Information and Press Department of the Ministry of Foreign Affairs. Lavrov is foreign minister.


What does this have anything to do with the extradition case?


Assange is a russian agent. He's not a journalist.


>"Assange is a russian agent."

Wake me up when the court renders decision that he is.

And even if he is it should have no relation to this particular case.


It's a bit rich to be lectured by Russia on freedom of speech.

Word to Russia: "Why do you see the speck that is in your brother’s eye, but do not notice the log that is in your own eye? Or how can you say to your brother, ‘Let me take the speck out of your eye,’ when there is the log in your own eye? You hypocrite, first take the log out of your own eye, and then you will see clearly to take the speck out of your brother’s eye." (Matthew 7:3-5, English Standard Version.)

I mean, look, the US should be judged for how it upholds its own values, and it does so less than perfectly. It justly deserves condemnation for when and where it falls short of its own stated values. But Russia... does Russia really want countries to be judged by freedom of speech?


Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: