Iridescence

24, Performance-obsessed systems engineer. Zig lover.
Author of CrossCraft

[Iridescence @iridescence_dev]

CrossCraft Classic on 3DS! (in-progress)

[Iridescence @iridescence_dev]

@bee_fumo @vaxryy Why wouldn't you? In something as critical as an init system you should know when a returned table is invalid / has typos way before a user ever runs it. Any functions the init system API provides also are easier to validate and you can validate a config without running it.

[Iridescence @iridescence_dev]

If we accept this premise, we should see that miraculously your phone runs faster. Games work on mediocre hardware. Memory usage decreases, user enjoyment increases. Any layman can create his custom app

The fact it isn't shows us not all code can be written by an AI or by anyone

[Ed Andersen @edandersen]

@cath_menarion Yeah but now I get zero dopamine from it now that I know it could have been done in a fraction of the time with AI by anyone

So it is not enjoyable

[Iridescence @iridescence_dev]

@muzudho1 @KDallas368 @valigo そう、通りです！

[Iridescence @iridescence_dev]

@saltyAom Close enough, lgtm.

[Iridescence @iridescence_dev]

@KDallas368 @valigo Pedantically: It's the ability to conditionally jump to arbitrary locations on the tape. A loop is just one type of jump.

[Iridescence @iridescence_dev]

@DaytonEllwanger @valigo They could, but it's probably a calculated tradeoff of what makes business sense over any genuine love for good software that an engineer like you or I may share.

[Iridescence @iridescence_dev]

As a whole organization, yes the best engineers are 100% in the ML and backend ML side. Here's why: Claude Code is a cash cow with huge inertia. People will use it as the default until it's bad enough to move. The amount that moves away because Claude Code sucks is insignificant to the total revenue it generates at this point in time. The biggest dumpster fire burning at Anthropic is compute costs for inference and training, with the majority being training. Frontier multi-trillion parameter models are hard to run and even harder to train. If they can spend $400k on engineering talent what are they going to spend it on: possibly making the cost of training go from $5B to $2.5B or making the TUI nicer to go from $1B to $1.1B if that?

[Iridescence @iridescence_dev]

@DaytonEllwanger @valigo Allocation. You can have all the best engineering talent, the core product is the model which you pay for, not necessarily the harness. The business, seeing this, allocates the best engineering talent to making the next models and making them better rather than improve the TUI.

[Iridescence @iridescence_dev]

@rfleury Lmao why even have an option if it just ignores it and does whatever it wants.

[Iridescence @iridescence_dev]

@kidtsang And it's not like we started with high amounts of "taste" from software developers with respect to performance, memory usage, responsiveness, etc. or even just the complete disconnect from what consumers use as hardware vs devs. And AI is only going to amplify these "tastes".

[Iridescence @iridescence_dev]

Imagine spending "10,000 hours" vibe coding and being confidently incorrect and belligerent to more experienced developers who try to help you just for this result. You would be a way better developer if you just spent 10,000 hours learning how to actually code then using AI.

[robot @alightinastorm]

if you ever wanted to know how long it takes to vibe code with codex + threejs a whole city editor with highways, streets, junctions, traffic lights, terrain, volumetric clouds, atmospheric effects, GTAO, FXAA, procedural buildings, props, traffic signs, a (semi) realistic car simulation, all from scratch, zero, nada except for threejs

it's 11 days, with 6 hr evenings each

now we've established this, stop pretending you can do this without vibes coding faster and better

thank you for your attention to this matter, making of video below in the comments and if you see this for the first time, it's worth going through the quote posts

[Iridescence @iridescence_dev]

@kidtsang Micro "helper" functions, comments which will definitely go stale, brittle tests, etc. It's more dangerous to use it closer to the start of a project or in this case, for the whole project, than it is to integrate with an existing large codebase.

[Iridescence @iridescence_dev]

And to be clear, I'm not even against using AI in its place: a tool for refactoring, information aggregation, and trivial mechanical generation. But I think you stop growing as both a programmer and a person if you stop inspecting what your agent is doing and fully "vibe" it.

[Iridescence @iridescence_dev]

A blade swung 10,000 times without purpose will never rival one guided by purpose.

You cannot cease entirely. Your brain is a muscle. It will atrophy and forget the texture and quality of those 30,000 hours of cultivation.

I fully understand why one would vibe code for a job that does not care about craft. But for your own work, you are simply doing yourself a disservice.

To care about software, both as an artifact and a craft is to think deeply. To find a new idea, a new rhythm, a new groove. If you never look at the code of your personal projects, what are you learning? Or are you merely a servile human agent for your mechanical god?

[robot @alightinastorm]

@iridescence_dev now hear me out, i have spent over 30,000 hours coding pre AI

i just don't do it anymore 🤯

[Iridescence @iridescence_dev]

I get that he made the game in 66 hours, but imagine how much you could learn as a human in 66 hours that is just wasted being a human REPL for his exo-brain.

[Iridescence @iridescence_dev]

Most of the "open source" crowd that is mad about how this open source project determines which patches to accept were never meaningfully able to or going to contribute. A small handful of contributors make the vast majority of impactful changes. Huge win for the core team.

[Ladybird @ladybirdbrowser]

Ladybird is moving into a new phase as we work toward our first alpha release.

We are tightening how code enters the project: going forward, code changes will only be introduced by project maintainers, and we will no longer accept public pull requests.

https://t.co/iauF4r9f3q