Oh the gangs are doing that already.
https://www.cbsnews.com/news/drug-cartels-more-bomb-dropping...
376 karma · joined April 3, 2024
https://www.cbsnews.com/news/drug-cartels-more-bomb-dropping...
All of them do at this point. It isn't hidden.
You can't buy a large FPGA without an ARM core in it. The ARM cores all have an opaque signed blob running in EL3 that you can't replace. This isn't a soft core on the fabric; it's dedicated silicon. And it has access to the ICAP (internal configuration access port) on Xilinx devices, and the equivalent on all the other manufacturers.
Also, it doesn't work:
error: access to absolute path '/home/user/.ssh/id_rsa' is forbidden in restricted mode
https://nixos.org/manual/nix/unstable/command-ref/conf-file....
Hope this helps.
I don't think it did. I'm not sure what it was supposed to help with.
Try being your own bank. Dump the fiat.
I dunno about you, but I definitely don't believe those two powerclowns.
I demand a duel. Tablets at twenty paces. No BitBlts.
this sounds scammy
I know that humans are doing more than interpolating, because at the rate we read and for the typical lifespan we have, we simply cannot ingest enough text to perform the sorts of tasks we perform by simple interpolation.
I also know that whatever our brains are doing, it isn't backpropagation, nor is it even remotely related to it. The inventor of backpropagation, Geoff Hinton, frequently points this out. Backpropagation is egregiously nonlocal.
... which is 100% blobless. Or the Cavium MIPS64 chips (100% blobless). Or if you want Arm64, the Rockchip RK3399 (100% blobless). You have plenty of choices.
One mitigation is to keep the device offline.
Here's an even better mitigation: don't buy junk like this chip.
Stop blobwashing junk hardware like this with deceptive misrepresentations.
Things like the Intel ME are what make this arrangement work. The fact that the ME exists is public knowledge; there is a huge team of engineers at Intel who are responsible for designing it. The flaw that lets you tickle it in just the right way can be added by and known to only one or two people.
Just this year a spectacular example of one of these bugdoors was revealed in Apple's mobile chips, where an EL1 (kernel) attacker could tickle an undocumented L2 cache debugging mechanism to elevate to hypervisor/bootloader (EL2, EL3) level, and completely bypass Pointer Authentication:
https://media.ccc.de/v/37c3-11859-operation_triangulation_wh...
This hardware bugdoor (CVE-2023-38606) was discovered only because it was used against security researchers who were good enough to capture a trace of exactly what it was doing. Otherwise we would never know. We got very lucky here; there's no telling how many more of these things go undiscovered.
EVs won't affect the noise problem, they'll just cause the assholes to change their tactics.
Ptychographic X-ray laminography has come a long way and can image even the latest chips at the transistor level:
https://spectrum.ieee.org/xray-tech-lays-chip-secrets-bare
True, it requires a synchrotron (one of the few things that actually costs more than a chip fab), but China has those:
https://en.wikipedia.org/wiki/Shanghai_Synchrotron_Radiation...
China might not be able to make bleeding-edge 2nm chips, but it has the equipment needed to expose any backdoors in them. Publicly, worldwide, with proof. That is an existential threat to any hardware manufacturer -- getting caught doing that is a bankruptcy-level event. Until recently they didn't have to be afraid of getting caught.
Being able to image chips at the gate level or better is a critical lever in fixing the balance of power between hardware manufacturers and their customers. It's why Bunnie Huang's work on IRIS is so important:
Dude, not even close. That repo is for an opaque blob:
https://github.com/SolidRun/cn913x_build/blob/master/binarie...
... which is mashed together-bits of other blobs, including the notorious will-never-be-open-source highly-radioactive "snps blob":
https://github.com/MarvellEmbeddedProcessors/mv-ddr-marvell/...
and its vile EULA which you have agreed to:
https://github.com/MarvellEmbeddedProcessors/mv-ddr-marvell/...
That shit contains an ARC core. Never heard of ARC? That's okay, most people haven't. It's an obscure niche architecture used for almost nothing except for ... drum roll ... the Intel Management Engine (up until 2019). Gee I wonder why that's in there.
Don't touch Marvell shit with a ten foot pole, it's blobs all the way down. They're just good at hiding it.
Tracking these sorts of "X means Y" mappings is precisely what the Q-K-V matrix of a transformer (or rather, Schmidhuber Fast Weight Programmer) does. This particular capability isn't even learned -- it's programmed in by the human who wrote the model evaluation code!
Whatever we call it there clearly is some amount of emergent abstraction in the models
I really, genuinely question this. I see extremely-high-dimensional interpolation over an extremely large dataset. Take away the dataset and what's left is gradient descent. And the token embedding, I guess. I'm not sure how you would "unlearn" something (like King-Man+Woman=Queen) from the embedding, or even what that would mean.
It is far from clear that this presumption holds.
Ingesting more text than a human could read in several lifetimes produces the ability to interpolate answers to a surprisingly large range of questions. This is not intuitive to humans, because we've never met anybody who could possibly read this much raw text, so we mistake this level of interpolation for reasoning since we've never met anybody who could possibly memorize-and-interpolate so much. The only thing we've ever seen that can answer questions like this is a human using reasoning capabilities, so we assume that's what this thing on the other side of the screen must be doing. Like chimpanzees mistaking their own reflection in a mirror for another ape.
This is the other "bitter lesson" of ML.
But if you spend enough time playing with these models you start to figure out what questions to ask to make them look foolish. And that is totally fair game for the Turing Test. Remember, there is no time limit on the Turing Test. However there is a strict requirement that the machine under test cannot be serviced, modified, or updated while the test is underway -- and because of this, nothing that OpenAI has produced is capable of even taking the test. We know that OpenAI tweaks and tunes their models whenever they please, as many times a day as they like, and that they use discussions here on HN to feed that process. So stick to the models you can download.
It's uncontroversial to say that this sort of "massive interpolation using superhuman text ingestion" is exactly how modern machine translation models work, and they work extremely well. LLMs were created by taking a machine translation model, throwing away half of it, and then fiddling with the leftovers.
I like articles like the one above where you can see the author's original title, before the editors got their grubby mits on it. Much more informative:
Realpage antitrust lawsuits allege collusion among corporate landlords
Massively higher information content in that title. But of course, lower "conversion rate" so add moar clickbait.
The great thing about unix is that there are no "wars" over these things, because everybody gets to decide for themselves.
Well at least that's how it was before systemd -- and all of DJB's unix work long predates systemd. By the time systemd came around DJB had been focusing on ECC exclusively for almost a decade.
The way I remember it is that most people didn't understand DJB and just kinda ignored his work, while a bunch of other people recognized what he was on to and integrated his ideas into software with frendlier user interfaces. For example, runit, which is PID1 for Void Linux to this day, and s6, which is PID1 for both Liminix ("NixOS-on-your-wifi-AP") and Spectrum ("Qubes for Nix"). Indeed increasing numbers of NixOS users are ditching systemd for s6.
Anyways I don't remember anything close to a "holy war".
Well they kinda are. And most of them want to stay in the Bay Area, so you don't end up randomly losing a bunch of your potential "excellent match" candidates because of geographical preference. There are only two other schools in that tier; a lot of MIT grads want to stay on the East Coast.
Carnegie Mellon grads are the recruitment secret superpower though: they all want to get the hell out of yinzburgh as quickly as possible -- anywhere on earth is an improvement. Crazy smart too, especially the ones who do programming languages and compilers.
They use OCaml because it is explainable. Which it is. They use it like a non-lazy version of Haskell -- side effects are used rarely if ever. So there's no nonlocal behavior in the code, which makes it easy to reason about. And that kinda matters when a lot of money is at stake.
Incidentally, Standard Chartered has their own compiler for Haskell, without the laziness. The group is led by the guy famous for Cayenne (the first dependently typed Haskell dialect).
Mostly this.
DJB is a genius, responsible for all of the non-NSA asymmetric cryptosystems, symmetric cryptosystems, and authenticated encryption algorithms supported by TLS (curve25519, chacha20, Poly1305). He's also the one who got us off of the footgun-by-design, broken-random-number-generator-will-spray-your-privatekey-everywhere nondeterministic nonce signature schemes prior to Ed25519 (the first standardized signature scheme which required deterministic nonces). Oh yeah and the only post-quantum cryptosystem that OpenSSH was comfortable shipping.
And pottering gave us pulseaudio. The gift that keeps on giving.
I'm still seeking a source here.
Yes of course, internet appliances are capable of DNS-blocking whatever their operators tell them to block. This is not news to me, nor to anybody else reading Hacker News.
Also the vendor of this appliance does not classify things as "Extremism". The owner/operator of the appliance does those things. Claiming that "SonicWall is classifying it as Extremism" is like claiming that "Intel marked your email as spam" because your SMTP server uses an Intel CPU.
Those are the best ones; stop telling everybody about it or they'll get flooded with losers from that orange site I keep hearing about.
I disagree.
Moronic people in positions of power are often in this situation. So far this has worked out well for the moronic powerful people: as long as they can convince intelligent people to work for them, they get to be moronic, powerful, and successful. This strategy has been working for them (as a group) for several millenia.
I think this is why the "LLMs will evolve into God In My Pocket" story gets so much traction. To a moronic person in power this sounds a lot like everyday life, but with computers cleaning up their messes instead of humans. It is totally plausible to them.