HackerTrans
TopNewTrendsCommentsPastAskShowJobs

panstromek

no profile record

Submissions

The key insight for programming with LLMs effectively

yoyo-code.com
2 points·by panstromek·há 4 meses·0 comments

comments

panstromek
·mês passado·discuss
Casey also got less aggressive when talking about Stroupstrup lately, especially after his last talk at Better software conference, where he mentioned him multiple times with a lot more historical context.
panstromek
·há 2 meses·discuss
This is good diagnosis of the problem. I found that often the right solution is to ditch the hierarchy and use flat structure. I wrote about this some time ago: https://yoyo-code.com/embrace-flatness/

Big insight in that article is also from https://matklad.github.io/2021/08/22/large-rust-workspaces.h... about structuring large rust workspaces as a flat list.
panstromek
·há 3 meses·discuss
Yea, when there's a match, our app stops playing videos in Spain and we get some bad reviews. It's pretty annoying.
panstromek
·há 3 meses·discuss
I will join the others and say you should just leave twitter: https://yoyo-code.com/you-should-delete-twitter/
panstromek
·há 3 meses·discuss
That makes sense, because JWT is base64 encoded, and those base64 tokens are bigger and more expensive. JWT has 3 parts, so it's 3x more expensive, obviously.
panstromek
·há 3 meses·discuss
Also, the game is much faster on desktop, looks like it's maybe not framerate independent?
panstromek
·há 3 meses·discuss
The game is pretty fun. Page height is a bit wrong on mobile, you probably need some dvh height?
panstromek
·há 3 meses·discuss
Depends on how many JSON tokens you need to format. I recommend getting JSON ForMAX+ with 200k tokens and 100k sign in bonus.
panstromek
·há 3 meses·discuss
Actually, from the OP, the timeline is:

> March 31, 00:21 UTC: [email protected] published with [email protected] injected

> March 31, around 01:00 UTC: [email protected] published with the same payload

> March 31, around 01:00 UTC: first external detections

> March 31, around 01:00 UTC: community members file issues reporting the compromise. The attacker deletes them using the compromised account.

So it was found out almost immediately.
panstromek
·há 3 meses·discuss
Well, the hack didn't survive more than 2-3 hours if I'm not mistaken. I don't think that counts as "nobody acted on it."
panstromek
·há 3 meses·discuss
As far as I understood it, it only talks about electricity, so that doesn't seem like a contradiction to me. I think some electrification of heating is expected in 2030, but not that much bigger than it is now.
panstromek
·há 3 meses·discuss
The parent comments references real world data from Google: https://security.googleblog.com/2024/09/eliminating-memory-s...
panstromek
·há 3 meses·discuss
The talk "Black-Hat LLMs" just came out a few days ago:

https://www.youtube.com/watch?v=1sd26pWhfmg

Looks like LLMs are getting good at finding and exploiting these.
panstromek
·há 4 meses·discuss
Yea, passing `-movflags +faststart` to ffmpeg when processing the file should be enough.
panstromek
·há 4 meses·discuss
If you don't need to switch versions at runtime (ABR), you don't even need to chunk it manully. Your server has to support range requests and then the browser does the reasonable thing automatically.

The simplest option is to use some basic object storage service and it'll usually work well out of the box (I use DO Spaces with built-in CDN, that's basically it).
panstromek
·há 4 meses·discuss
That's good and arguably the right default for most websites.
panstromek
·há 4 meses·discuss
Yea, honestly you probably just don't understand. FE frameworks solve a specific problem and they don't make sense unless you understand that problem. That TSoding video is a prime example of that - it chooses a trivial instance of that problem and then acts like the whole problem space is trivial.

To be fair, React is especially wasteful way to solve that problem. If you want to look at the state od the art, something like Solid makes a lot more sense.

It's much easier to appreciate that problem if you actually try to build complex interactive UI with vanilla JS (or something like jQuery). Once you have complex state dependency graph and DOM state to preserve between rerenders, it becomes pretty clear.
panstromek
·há 4 meses·discuss
> Most integration tests are not thread safe and make assumptions about running against an empty database. Which if you think about it, is exactly how no user except your first user will ever use your system.

Yea, cypress has this in their anti-patterns:

https://docs.cypress.io/app/core-concepts/best-practices#Usi...

Dangling state is useful for debugging when the test fails, you don't want to clean that up.

This has been super useful practice in my experience. I really like to be able to run tests regardless of my application state. It's faster and over time it helps you hit and fixup various issues that you only encounter after you fill the database with enough data.
panstromek
·há 4 meses·discuss
> It feels a little tricky to square these up sometimes.

In my experience, this heavily depends on the task, and there's a massive chasm between tasks where it's a good and bad fit. I can definitely imagine people working only on one side of this chasm and being perplexed by the other side.
panstromek
·há 5 meses·discuss
> There must be a really good reason for this, such as Rust doesn’t interop well with C++

Yea, I'd bet it's that. Ideally, you'd want to stop writing C++ and continue with Rust on all new code, but Rust has stricter semantics, so the interop is somewhat "easy" in one direction, but very hairy in the other direction.

This means that in practice, you want to start porting from leaf components and slowly grow closer to the root, which stays in C++ for quite some time and just calls into Rust through C API (or something close to it).

If you're curious about the topic, there's a interop library called Zngur (https://hkalbasi.github.io/zngur/), which is built on this assumption. They have a pretty good explanation of the concrete problems on the homepage.