Don’t know why you are getting downvoted, it’s absolutely true. Raw specs these days mean relatively little. With smart frequency boosts that vary with thermals, CPU and GPU on the same package, different workloads stressing different components differently, RAM bandwidth playing different roles for CPU and GPU applications, and many other factors, just stating that the M4 has so and so many cores is practically useless.

The only real way to gauge performance differences is via benchmarks and measuring sustained workloads.

Yeah that’s what we did last time. I implemented a basic framework on top of a very widespread system in our codebase, which would allow a number of requested minor features to be implemented similarly, with the minimal amount of required boilerplate, and leaving the bulk of the work to implementing the actual meat of the requests.

These requests were completely independent and so could be parallelized easily. The “framework” I implemented was also incredibly thin (basically just a helper function and an human instruction in the shape of “do this for this usecase”) over a system that is preexisting knowledge. My expectation was to have to bring someone up to speed on certain things and then let them loose on this collection of tasks, maybe having to answer some question a couple times a day.

Instead, since the assigned colleague is basically just a copilot frontend, I had to spend 80% or more of my days explaining exactly what needed to be done (I would always start with the whys od things since the whats are derived from them, but this particular colleague seems uninterested in that).

So I was basically spending my time programming a set of features by proxy, while I was ostensibly working on a different set of features.

So yeah, splitting work only works if you also have people capable of doing it in the first place. Of course I couldn’t not help this colleague either, that’s a bad mark on performance review you know. Even when the colleagues have no intention of learning or being productive in any way (I live in a country with strong employee regulations so almost nobody can be fired for anything concerning actual work performance, and this particular colleague doesn’t hide that they don’t care about actually doing a good job, except to managers so they still get pay raises for “improving”).

Yeah, you can tell I’m unhappy

who is actually stopping them from dealing with it?

Management. Someone in management sets idiotic deadlines, then someone tells you “do X”, you estimate and come up with “it will take T amount of time” and production simply tells you “that’s too long, do it faster”

they don’t care about the details or maintenance

They don’t, they care about time. If there are 6 weeks to implement a feature that requires reworking half the product, they don’t care to know half the product needs to be reworked. They only care to hear you say that you’ll get it done in 6 weeks. And if you say that’s impossible, they tell you to do it anyway

you have to include the cost of managing technical debt

I do, and when I get asked why my time estimations are so long compared to those of other colleagues I say I include known costs that are required to develop the feature, as well as a buffer for known unknowns and unknown unknowns which, historically, has been necessary 100% of the time and never included causing us development difficulties and us running over cost and over time causing delays and quality issues that caused internal unhappiness, sometimes mandatory overtime, and usually a crappy product that the customers are unhappy with. That’s me doing a good job right? Except I got told to ignore all of that and only include the minimum time to get all of the dozens of tiny pieces working. We went over time, over cost, and each tiny piece “works” when taken in isolation but doesn’t really mix with everything else because there was no integration time and so each feature kinda just exists there on its own.

Then we do retrospectives in which we highlight all the process mistakes that we ran into only to do them all again next time. And I get blamed come performance review time because I was stressed and I wasn’t at the top of my game in the last year due to being chronically overburdened, overworked, and underpaid.

I see this as political discourse. I can now look at whoever started and starred the fork and unquestionably categorize them as people not worth interacting with.

What “it” is configurable? If the code is indented with 4 spaces, it is indented with 4 spaces. You can configure your editor to indent with 1 space if you want, but then your code is not going to respect the 4 spaces of indentation used by the rest of the code.

I repeat, the only accessible indentation option is using tabs. This is not an opinion because every other option forces extra painful steps for those with vision issues (including, but not limited to, having to reformat the source files to tabs so they can work on them and then reformat them back to using spaces in order to commit them)

deleted by creator

Code example looks wrong? null is passed as updater to a. As far as I understand, what is passed as callback to a is supposed to be the updater instead (especially since the ctor errors on both val and updater being null). Also, would be more clear what the difference and usage of updater and callback are if they were called something like calcVal and onValChanged respectively

deleted by creator

Looks to me like the ruling is saying that the output of a model trained on copyrighted data is not copyrighted in itself.

By that logic, if I train a model on marvel movies and get something that is exactly the same as an existing movie, that output is not copyrighted.

It’s a stretch, for sure, and the judge did say that he didn’t consider the output to be similar enough to the source copyrighted material, but it’s unclear what “close enough” is.

What if my model is trained on star wars and outputs a story that is novel, with different characters with different voices. That’s not copyrighted then, despite the model being trained exclusively on copyrighted data?

deleted by creator

What is this mysterious device that requires specific libraries and kernel modules? So I can state the fuck away from the device and the brand

if you’re using windows and expect any privacy at all […] throw that notion out the window

Correct. And the same is true even if you are using linux, macOS, android, or a butterfly to manipulate bits to send a message through the internet.

Because if your message ends up on the screen of a windows user, it’s also going to be eaten by AI.

And forget the notion of “anything you post on the internet is forever”, this is also true for private and encrypted comms now. At least as long as they can be decrypted by your recipient, if they use windows.

You want privacy and use linux? Well, that’s no longer enough. You now also need to make sure that none of your communications include a (current or future) windows user as they get spyware by default in their system.

Well maybe not quite by default, yet

The “or later” is optional, the FSF specifically doesn’t have the power to update the terms of every GPL-licensed software because the wrote the clause in such a way that they don’t.

If I give you software licensed under the GPL3, and a GPL3.1 comes out, it doesn’t apply to your copy of the software. Likewise the copyright holder of the work is also not forced to relicense their software under the GPL3.1. And even if they did, copies of the software distributed under the GPL3 would still be licensed under the GPL3.

The “or later” clause simply means that if I received a copy of a GPL3 software, I can redistribute it under the GPL3.1 if I so wish (where “I” in the previous sentence is everyone with a copy of the work, as the GPL gives everyone with a copy redistribution rights)

deleted by creator

deleted by creator

At work we use lld as opposed ld. With ld, the project I work on links in 60-something seconds, with lld it links in less than 6.

Mold is faster than lld. It is absolutely worth it

Vampire survivors. I swear it’s one of the very few games that I found actually fun in the last 5 years

deleted by creator