@qwwerty the most boring thing of all time is a paper comparing the training of 2 copies of the same model using 2 different implementations of ML. Yawnnn

@YourMom of course, models are generally initialized with random weights, not zero, because zero weight wrongfully indicates a negative relationship between two neurons. If the weight is random, it can be smoothed out as you train into the correct value as you optimize for your loss function.

I tuned my LLM personality to make ChatGPT drop the sycophancy plus other conveniences about a month ago. It works great and now I don't feel imposter syndrome every time it blatantly over-praises me (it no longer does this)

The importance of this depends on what "ML code" is.

If it's an already-trained model with temperature at some constant, this is slightly interesting because AFAIK an LLM tuned with 0/ constant temp should produce the same results. It's just a program reading the weights of the model.

If it's a container that is training the model twice and comparing results, this is painfully obvious. There's a lot of stochasticity in model training.

thats why i like windows 7. it only took 7 minutes to boot....

@Lensflare @badger yeah the physics of it all makes this question terribly un-fun. That's why I specified it's a visual-only glitch: the world still works, so the only thing left to do is notice

It would probably take me a few weeks. Why? Well, that's just about how often I use a flashlight on something thats kinda far away (typical flashlight use is from my phone on nearby objects)

I'm talking the tops of trees, the end of the street, whatever. Outdoor flashlight use.

I feel like I would notice the lag of the beam on very far targets. It would be extremely subtle, but it would be there. Like switching from 60 to 120 FPS on a game. Otherwise, how would I even know?!

@whimsical I couldn't get past two days of it. Fuckin boring, ugly, difficult language that saps all the fun out of programming.

(If I actually put in the work to learn it, I would probably love it)

Others have already corrected you but yes, this was nothing to do with Rust, not even tangentially. Cloudflare's Rust system is very accomplished and something like Python would never be able to handle the loads CF sees without doubling server costs.

The error was in their change management system and outage management/ mitigation. Some poor SQL made it to production, doubled the expected max input size, and propagated quickly throughout the network due to quick response needs (canary not possible).

However, I saw someone mention on Hackernews that there was a EXTREMELY similar thing that happened to Google, but their automatic systems caught the bad change, rolled it back within 4 minutes, before the on-call team could even hear about it.

If you want to blame a language: SQL

If you want to blame the real issues: change management and disaster mitigation systems.

If you *really* want to blame rust: stack traces not being available by default.

@jiraTicket Nvidia is in a dangerous position where Radeon cards are a very extremely viable alternative and even Intel Arc cards are starting to get good. When you upgrade your GPU, go for AMD instead of Nvidia!

@YourMom yeah for a very short duration they made hash-limited cards, it was firmware. But you need to ask yourself what amount they were producing for consumers vs. the amount they were producing for crypto farms. When crypto boom happened, GPUs were a unicorn in a haystack, and the hash reduced cards didn't do shit. This is because Nvidia could basically sell anything they squirted out of their factories to crypto miners and Nvidia couldn't give a shit and a half about gamers and their cards getting scalped.

@YourMom why would I want to join that list again?

I wanted to say "I think this might really be the time I delete my account" but where the fuck do you go, Reddit? I'd rather die. And I'd rather deal with 5 minute page load times before that.

JS (browser) equivalent:

window["$50.00"] = "JavaScript"

window["$50_00"] = "likes"

window["50"] = "it"

window["else"] = "rough"

1-4 are valid and you can even reference #2 without PO(JS)O notation.

The fact that you indeed have 10 years of experience and additionally said time was not spent with your thumb up your ass doing nothing, probably

That's cool! I just started my first day at github. These idiots think it's some sort of exercise program, but I know better, so I deleted push and pull. Hopefully people appreciate it 🙏🙏🙏🙏🙏

If you need, I can pose as a consultant and tell them to use whatever you want.

Which - out of curiosity - is what? My personal projects I go 100% vanilla and implement all my own controls as custom HTML elements (sort of like web components, but I typically avoid shadow DOM because it's more headaches than it's worth).

But my personal projects don't have the same level of seriousness as a business project does.

With all the time and code you guys spend and write on devRant peripherals, you could have made sox clones by now. At this point just steal the SVG assets and republish them, dfox isn’t even active enough to shut down the site let alone sue you for stealing.

@Lensflare I don't quite see myself working with Swift or Kotlin in the foreseeable future, unfortunately. I do quite like Swift though, maybe I'll spend a weekend with it one of these times.

@Lensflare That's the tragedy, isn't it? Excited to learn it, but crushed by the reality that it'll only ever benefit me in personal projects or academically.

@Lensflare I think so because on a daily basis I work with other people's code and I can barely think of any place besides a really techy place that will actually use algebraic data types, let alone a language that implements them.

@Liebranca In reference to what you actually said, I agree. From my reading, the real benefits of functional languages is they're optimized for recursion (or, more to the point, they're more optimized to turn recursion into something actually performant!)

From the code I've been writing, it's a lot more difficult, but the code ends up being shorter and a lot closer to what you're actually trying to do.

Like, I saw a quicksort (unoptimized) that was essentially just the algorithm in plain f#. There was barely any implementation, it was just "here's how the algorithm works, here's how to say that in f#, done"

https://fsharpforfunandprofit.com/p...

@Liebranca I spent today studying F# and I can confirm the stdlib is the main drawback: it's designed to be compatible with the .net framework 100%, so the function calls are not curried by default, they're tupled. So insted of calling them in a functional way, you call them in a traditional c# etc way;

F#:

print "hello %s" world

F# Standard library:

print("Hello %s", world)

Which was a major disappointment, so I'll be shelving this functional thing and moving to Haskell when I get the itch again. Shame.

@Liebranca seems like it’s a trope in the functional space :(

@12bitfloat @retoor ya, my opinion is definitely an outsider's one, that's why I said it was null from the start. I don't know much about Linux or Rust so I'm just guessing. I've seen some of the drama between Linus and the Rust for Linux community and that's basically all I'm working off, besides the two days I spent with Rust before getting bored and moving to Zig instead (fun!)

This is actually a real thing! I watched a YouTube video on Film Noir last night.

Noirvember has a wikipedia page!: https://en.wikipedia.org/wiki/...

As someone who wrote a fairly large JS-only backend and then abandoned it exclusively for the reason it was JS-only, let me give some insight (TypeScript only has a subset of these issues, the build system actually is really nice for adapting JS to the back-end):

- It's great for really small stuff. I still recommend it as a backend for dev servers, simple game servers (like slither.io, whatever), etc.

- It scales horrendously. I was even using modules, not CommonJS, and shit was just awful. Refactoring only worked half the time, so once you write code and start to use it, you're basically stuck with it.

- Performance *does* stink. You can point to benchmarks where bun outperforms assembly or whatever, but at the end of the day, you're writing a scripting language that lives on the stack. If you need to do something that requires high performance, you're going to need to write it as a micro-service that your backend calls out to.

... more, but I'm out of comment space.

I don't really give a shit about the memory safety of Rust. You can write memory safe c code, and c code that is the age of the Linux kernel (plus, the extra set of eyes helps, being the largest collaborative software project on the globe) typically is memory safe anyways.

It's mostly just that c *will eventually* fall out of style. It's already largely out of style. Business applications simply are not written in c anymore. While most devs can make their way around in a c codebase, there are vanishingly less c professionals that work with it daily. That's a bad thing.

But a newer tool, like Rust, that genuinely finds its way into non-niche business applications regularly, has a professional pool to pull from. People that use it at work, then come home and work on open source projects for fun or whatever.

In the next few years I'm honestly expecting a fork of Linux to be completely re-written in Rust, without Linus.

I'm torn on rust-for-linux.

For one, I'm not a Rust dev, so my opinion is largely null anyways.

I am a c/c++ hobbyist, though, and updating the tool chain isn't that bad of an idea from this perspective.

I think the Rusty conversion will be good, but it will take time. A bunch of lifeless hacks will half-implement buggy versions just because they're Rustacean fanboys and want Rust in Linux. We're probably in that stage now.

Then, once things get noticeably bad, genuinely good devs will step in and either replace or fix the shit-tier code hastily written by the hardcore Rustaceans.

Well if it's an internal tool, then that's a horse of a different color, especially if all 8 people are also responsible for maintaining the database.

There should still be a prod and dev environment though, and there should still be a promotion process for database changes from one to the other.