65a

As a user, I like wayland. X11 was a security disaster. Wayland is much better about tearing.

What scares me though are all the responsibilities passed to compositors, because what ends up happening is that each compositor may reimplement what should be common functionality in annoying ways. This is especially true for input things, like key remapping. This ultimately fragments linux desktop experiences even harder than it was before.

I'd really like to get more clarification on offline mode and privacy. The github issues related to privacy did not leave a good feeling, despite being initially excited. Is offline mode a thing yet? I want to use this, but I don't want my code to leave my device.

You can sometimes find the serial lines if you are careful. Otherwise you can use the flashrom to store the output, and read it back out after each failure. It is much easier to just poke around and find the serial if you can, either from schematics (it seems the author has these) or by hand with a lot of patience or board scrying.

Possibly. Usually this is handled by the embedded controller, and not sure if that was reversed or not. You may be able to tristate the GPIO line that tells the CPU that a pin means PROCHOT, which would allow you to ignore the ECs attempts to do this.

In a signature context, you probably want someone else to know that "you" signed it (I can think of other cases, but that's the usual one). The way to do that requires them to know that the key which signed the data belongs to you. My only point is that this is actually the hard part, which any "replacement" crypto system needs to solve for, and that solving that is hard (none of the methods are particularly good).

Yes, not saying that web of trust ever worked. "Pre-established channel" are the other mechanisms I mentioned, like a central authority (https) or TOFU (just trust the first key you get). All of these have some issues, that any alternative must also solve for.

I certainly want to get rid of gpg from my life if I can

I see this sentiment a lot, but you later hint at the problem. Any "replacement" needs to solve for secure key distribution. Signing isn't hard, you can use a lot of different things other than gpg to sign something with a key securely. If that part of gpg is broken, it's a bug, it can/should be fixed.

The real challenge is distributing the key so someone else can verify the signature, and almost every way to do that is fundamentally flawed, introduces a risk of operational errors or is annoying (web of trust, trust on first use, central authority, in-person, etc). I'm not convinced the right answer here is "invent a new one and the ecosystem around it".

4800MHz single rank ok?

Right, I think those were the closest we got to the CHRP standard, as they moved the platform toward PC-style floppies, PS/2, ATX PSU and even more generic "platform" stuff than most clones. I'm fairly sure I had an ISA slot, I do remember trying to get a bargain bin NE2K card working in mine under linux (it didn't work). Definitely did nothing under OS 8/9.

The powercity models were interesting, because they came out after Apple revoked Motorola's clone license. A German company, ComJet, bought up the boards and sold unlicensed clones cheap. Case was slightly different, but otherwise they corresponded to StarMax models (fairly certain they were identical but may have been last revision boards).

There's a great postmortem here about what might have been a similar SEU (single event upset--bitflip) here: https://www.atsb.gov.au/sites/default/files/media/3532398/ao...

StarMax series (and the 4400) seemed to be about as close to CHRP as we got. My off-brand StarMax clone (PowerCity) had a PS/2 and an ISA port. Ran BeOS well, and had a quirk that I could hear a tight loop on the speaker.

I really cannot say Uber's use of Go is particularly idiomatic to me, having started writing Go more than a decade ago now. It just strikes me as overwrought, and I've worked on big services.

UEFI itself is way too complex, has way too much surface (I'm surprised this didn't abuse some poorly written SMI handler), and provides too little value to exist. Secure boot then goes on to treat that place as a root of trust, which is security architecture mistake, but works ok in this case. This all could be a lot better.

Electrolytics are usually nothing too fancy, but it is proprietary. Water and electrolytes, hence the name. PCBs are in the big transformers and what used to be called bathtub caps which looked like this https://i.ebayimg.com/images/g/VjwAAOSwfGJjYtHx/s-l400.jpg (think 1950s electronics stuff)

I assumed they had jammed a new DXE into the UEFI capsule, which would probably be able to subvert secure boot.

I don't see how enabling secure boot helps here, since UEFI is responsible for enforcing that and is compromised. I'm sure some might recommend more roots of trust and signing down and verification that starts at the chipset, but I'd recommend an alternative with less attack surface and better user control: a jumper.

Go is safe from the perspective of RCEs due to buffer overflow, which is what matters here. Happy to be enlightened otherwise, but "I broke your (poorly implemented, non-idiomatic, please use locks or channels ffs) state machine" is a lot better than "I am the return instruction pointer now"

There are several pure Go bittorrent libraries from a cursory search

Voting for new legislators, personally. I wish they'd do something about PG&E or housing instead of criminalizing software development of chatbots. Truly useless, and I wish we had more choice of non-insane candidates.

This was also my reaction almost immediately. Tattoos can have extensive correlation with social and lifestyle factors that could easily mean the difference between correlation and causation here.

I find it hard to believe that a Canadian company's model contained an undertrained token related to hockey (albeit in German). In all seriousness, this is pretty cool and am excited to see understanding of tokenization impacts on models improve. One notable finding is that a lot of the earlier open source models have issues with carriage returns, which are not that uncommonly introduced depending on where the data is coming from etc.

This exactly. I would be much happier if the regulation was "don't use GPT-4 to decide when to kick Grandma out of the hospital" or "don't use a Llama finetune to make policing decisions", which is where I see the most certain need of regulation in the near future.

It wastes taxpayer funds on enforcing a moat for Sam Altman, it establishes a fixed computational bound in a legal regulation, it tries to police a free speech activity because of possible harms (but not the harms directly), and it is likely to have negative national security implications as other (less regulated) regions deal with fewer lawyers as they advance the state of the art.

I don't think this is a hot take at all, it's matches my understanding. One of the reasons language itself is so difficult (miscommunication, etc) is we have a mostly similar but not identical "compression table" of ideas that words map to, and why we spend so much time aligning on terms, to ensure correct "decompression".

We need compression because internally cognition has a very "wide" set of inputs/outputs (basically meshed), but we only have a serial audio link with relatively narrow bandwidths, so the whole thing that allows us to even discuss this is basically a neat evolutionary hack.

I get the strong sense you understand neither ITAR nor this article. Good luck!

There are certainly types of data that are already prohibited for export and dissemination. In this case, I would argue no new law is needed, the existing laws cover the export or dissemination of dual use technologies. If the LLM becomes dual-use/export-restricted/etc because it was trained on export-restricted/sensitive/etc data, it is already illegal to disseminate it. Enforce the existing law, rather than use taxpayer money to ban and police private LLM training because this might happen.

Laws should be about the outcome, not about processes that may lead to an outcome. It is already illegal in California to produce your own nuclear weapon. Instead of outlawing books, because they allow research into building giant gundam robots, just outlaw giant gundam robots.

640kb should be enough for anyone!

Or maybe we are repeating the experiment of the 1990s/2000s and can't wait to see the results confirmed?

The base unit is actually a single bit, so why aren't we talking about decabits instead of bytes, following the SI argument?