ckwalsh

If a new part of this arrest is that the robots are considered to be pedestrians, I wonder what the police would do if the professor replayed the exact same circumstances with a friend in place of the robot?

Decide one of them is delivering something, run into each other at the exact same spot, and call the police to resolve it?

I think you are missing the point.

> who tf uses prompting to "pretty please don't cheat on this"?

People who don't understand how LLM's work. Kids, teachers, your next door neighbor. CEOs and government officials.

I am not trying to argue that the author doesn't understand how LLM's work; they clearly do. Their prompt was written in a way that leaves those assumptions implicit, in a similar manner to those non-technical folks using LLMs every day.

> the best practices for ages...

We cannot expect the entirety of humanity to understand / use LLM best practices. We need to figure out how to ensure LLMs operate as the majority of humanity expects despite how they are implemented, and until we can do that, do our best to prevent them from being used in an unsafe manner by the nontechnical population of humanity.

“With funding from private investors,…”

Im guessing those private investors specifically requested not to be named.

Blackhole is the name of one of the services used in display-time malicious content filtering.

I’m guessing the urls in that db were either generating a ton of backend load, so they were pushed to devices, or perhaps are customized on a per user basis for some reason

I'm in the "Pro-Rust" camp (not fanboy level "everything must be rewritten in rust", but "the world would be a better place if more stuff used Rust"), and I love this post.

They saw some benefits to Rust, tried it, and continued to measure. They identified the Typescript/Rust language boundary was slow, and noticed an effect on their contributions. After further research, they realized there was a faster way that didn't need the Rust dependency.

Good stuff, good explanation!

> I am also wondering if Meta still uses PHP or Hack

Meta’s WWW codebase is its oldest, and still written in Hacklang. Lots of internal tooling is also written in Hack.

I wouldn’t be surprised if there are more committed lines of Hacklang than any other language (although, a good chunk of that is codegen, vs other core languages where codegen is part of the build process).

The "Memory Safety" of Rust is not that memory leaks are not possible, rather its how memory is accessed. Specifically, using "safe" Rust, it should be impossible to:

1. Access memory that hasn't been explicitly allocated to your program 2. Access memory that was was previously allocated and freed by your program 3. Access memory that could be modified by a separate thread / function of your program, outside of some locking mechanism.

I'm sure there's a more academic definition, and stronger guarantees, but those are the basics.

Now, these guarantees impose restrictions on how code is written, and runtime overhead to enforce. There are many cases where a human can write optimizations and ensure those guarantees in a way the compiler cannot verify. This is why Rust has an "escape" of writing "unsafe Rust", where the compiler does validate those guarantees. Unsafe Rust includes expanded primitives, such as "Box::leak()", that enable writing these optimizations, and may explicitly leak memory without maintaining a reference

If a program is written using purely safe Rust, it should not be possible for a malicious input / request to corrupt internal data structures, read memory that shouldn't be read, or all the other "interesting" things that get highlighted by Heartbleed and other big security vulnerabilities.

Note that I say "should not be possible" - the Rust compiler can and has had bugs that break those memory safety guarantees under certain conditions. They aren't common, but it's unfair to say "Rust is perfect and 100% safe"

I strongly recommend everyone learn Rust, at least to the point where you "grok" the borrow checker and what it's trying to enforce, even if you have no intention with regularly using Rust. After learning Rust, I found the patterns the borrow checker enforces incredibly useful when writing C++, and ensuring I don't make silly memory mistakes that may result in a crash or vulnerability.

Ah that makes sense. For sufficiently small values of N, a hardcoded allowlist isn't a problem.

You're probably right that RSA 4096 "just worked", and some library in their stack doesn't have elliptic curve support. And again, if N is small, the verification performance doesn't matter that much.

Nice find and writeup!

> the legitimate ones I’d initially generated still worked

This spooks me. I take this to mean either:

- They are still using the compromised key for validation, meaning if you have access to any old token, you can still mutate that, maybe needing to play around with the issuing times

- They built an allowlist of all permitted tokens, and check that list first. In which case, might as well use random session ids instead of JWTs, and at the same point where the allowlist is being checked, mutate the request to inject a JWT that the backend can use.

Also, kind of curious why the switch to RSA4096 instead of elliptic curves, since they are generally faster / smaller.

+1 to this. I run Ubuntu 22.04 with microk8s. Keep all my yaml files in a local git repo, figured out how to hook up my NAS to provide storage via nfs.

It's definitely gone down a few times, but I've learned a TON tinkering with it. super easy to spin up a new hobby project, a nice web UI for seeing what the heck is going on.

I've completely borked it a couple times and survived one micro pc migration. Can't recommend it more

Do you happen to know where I can read about how ET and Mosh each establish their connections?

I have used Mosh for years and recently heard of ET, but when I tried it I experienced noticeable hangs that I don’t get with Mosh, and I went back.

I heard from several people that “ET is the new Mosh”, but it won’t be for me unless I can figure out/resolve those hangs

The 5 hours doesn’t surprise me.

I was on the jury of a federal fraud trial with 2 defendants with 15 charges, ~30 million in losses.

We were thorough and went through each count separately, including reviewing some of the evidence, and were done in maybe 8 hours spread across 2 days.

We ended up with a mixed verdict: one count not guilty for both, another not guilty for one. I fully believe they were aware and committed fraud for the not guilty counts, but the prosecutor wasn’t able to cross the “reasonable doubt” threshold in our minds for those specific instances.

Only thing we weren’t super careful about was the first requirement for Mail/Wire fraud, which is “Mail and wires” were used.

It was amusing that the prosecutors brought in a bank IT guy to explain that “the internet uses wires”, but not really something we questioned.

Is there any way to tie an expectation of long term security support with legal protection of the product against competitors/reverse engineers/other parties that manufacturers may not want looking too closely?

I’m not suggesting granting additional protections to manufacturers, but codify an expectation of “if you abandon it, other people can come in and potentially salvage it”