Sasha Isaienko

I craft software

[Sasha Isaienko @sonerik]

Hot take:

macbook-only > macbook+display in terms of productivity

[Sasha Isaienko @sonerik]

@shiweidu Finalizer doesn’t guarantee any timelines on when it will be invoked. I some cases the sequence of old.dispose() -> new.create() is very important. Finalizer-based disposal breaks that sequence guarantees.

[Sasha Isaienko @sonerik]

All custom iOS glass effect implementations are screwed now, again?

[Sasha Isaienko @sonerik]

https://t.co/8FBjDn0AqP

[Sasha Isaienko @sonerik]

official Odin example, lol

[Sasha Isaienko @sonerik]

first you grind
then
you die
early

^ this is the usual outcome

[terminally onλine εngineer @tekbog]

first you grind
then
you shine

[Sasha Isaienko @sonerik]

Just built myself a functional custom ProtobufEnum switch exhaustiveness checking lint and proto <-> client <-> server navigation vscode plugin in an hour or so using LLM.

This kind of software was really out of reach before. Now it’s trivial. The single best use of LLMs.

[Sasha Isaienko @sonerik]

They desperately need an excuse for not delivering on their bullshit promises due to getting close to the tech's ceiling.

"Global slowdown in the name of safety" is the perfect one.

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

HOLY SHIT LET'S FUCKING GOO

[Sasha Isaienko @sonerik]

@eiranoirx That's the whole point of reputation.

[Sasha Isaienko @sonerik]

@Zhuinden @KhalidWarsa I worked with both and some other options, which led me to basically make a Provider, but with less boilerplate.

Feel free to try it out if you are open to something like this:

https://t.co/J0yTiGQ4WD

[Sasha Isaienko @sonerik]

@neogoose_btw My bet is $1100+. Even at that price, I’d still consider it as a replacement for my Series S, though.

[Sasha Isaienko @sonerik]

@hasen_95dx Cursor+Composer

[Sasha Isaienko @sonerik]

@sean_from_earth That's most of Flutter :)

And it's usually fine this way.

[Sasha Isaienko @sonerik]

@APEALED It's a decent low hanging fruit approach, though, IMO

[Sasha Isaienko @sonerik]

@alightinastorm Not a flex, man.

I'm using LLMs daily (most of us do), just not fully vibe-coding stuff. I just make sure that things actually work as intended, not just "plausibly". And that gap is very far from being closed by current LLMs.

[Sasha Isaienko @sonerik]

@alightinastorm Here's my vibe-coded JSON viewer, for example: https://t.co/0KNHEOdEtx

It "does the job", but changing anything in it is a real pain due to regression cycles. And it only gets worse with the project size.

[Sasha Isaienko @sonerik]

@alightinastorm Quite a lot, actually. I'm trying this once every ~6 months for the past couple years to see if there's any decent progress in that space.

It definitely got much better in the last two years, but the issues are fundamentally the same and I doubt this will ever change with LLMs.

[Sasha Isaienko @sonerik]

@twerske tvOS maybe?

[Sasha Isaienko @sonerik]

@mhrsntrk @fatih @cursor_ai Worked initially, but at some point during Composer 2.0 era it broke for me, original answers just got followed by a “caveman section”. With Composer 2.5 it seems the caveman didn’t make much difference in terms of verbosity, only diff is smaller synonyms and less punctuation.