I posted in the Mac apps forum about a browser I build, and I was told — reasonably enough — that it reads like it was written by a machine, and that the whole thing looks rigged. I answered that where it was asked. This is the part that did not belong in that thread.
Nobody asked me why I do this. I think the absence of an answer is part of what made it look wrong, so here it is, and it is short.
My grandfather was deaf and mute from birth. My grandmother was left disabled after typhus. This was the Soviet Union, where that meant a great deal more than it would now — not because they suffered more nobly, but because the state simply had no place for them, and the weight of that fell on the people around them. Mostly on my mother.
I want to be exact about what this does and does not give me. It gives me no insight into being blind. I have not lived it, and I am not going to claim a share of something that is not mine. Deafness is not blindness, and a grandson is not a witness.
What it gives me is the reason I do not stop. I grew up knowing that whether a person gets to take part in ordinary life is mostly decided by whether anyone bothered to build the thing properly. Usually nobody bothered.
That is all. I am not asking for anything here and there is nothing in this to buy. It was a fair question and it deserved a plain answer instead of silence.
Dmytro Liesniak
Comments
The issue isn't your intent - it's the generated prose
Tragic family history aside, the core criticism remains unaddressed.
You state you want plain, honest feedback, yet every update, bug response, and personal explanation is written in the exact same distinct, highly stylized LLM-generated prose. Using an AI to polish a personal post about your family creates a jarring disconnect between the claim of transparency and the medium being used.
Automated tools can fix markup, but when human-to-human communication is run through a text generator, it looks like an attempt to simulate empathy rather than offer it.
People are so ungrateful for hard work
And you wonder why decent suggestions and decent apps in these features don’t come because people are ungrateful when people try how about constructive positive feedback instead of being jerks to people
Tell Us: How Good Is Your Russian?
@DaniilGusev, how good is your Russian?
Are you offering to translate for all of us? If not, why complain?
If using AI allows him to communicate with us, isn't that a good thing overall? Near real time translation was something out of science fiction not too long ago. If he was a native English speaker using this tech out of laziness, then sure, I would probably pile on with criticism as well. But if he finds English difficult, why not use whatever tools we can to bridge the communication gap?
My take
I know it's a hot topic, using LLMs to write. Maybe for those of us who do need to use it, labelling it as such might be a good practice. Written with assistance from LLM, or something similar. I think people, quite rightly, get upset when others attempt to pass off generated text as their own work. The ideas are still their own, it's just that language is precious, and how we say something is almost as important as what we say. Automating this removes a portion of the credibility.
can't agree more
I am severely disabled myself on top of my blindness, and I am quite frankly sick of seeing posts where people find nothing better to do than to pile on someone's words just because it is llm generated.
Sometimes, but not always, I will resort to using llm to convey what I am trying to express in writing because I genuinely couldn't otherwise. This is not about laziness. This is not about struggling with a language I do not know. But in the end, the result is the same.
There is no need to treat us like we're less of a being than you are because we have to use llm to communicate correctly.
There is a big difference between translating and generating
My native language is Russian, and I use LLM tools to help translate and refine my messages into English as well. There's nothing wrong with using AI to bridge a language gap or improve grammar.
But using AI to translate your own thoughts is very different from posting raw, prompt-generated filler.
The issue here isn't English proficiency. It's the unedited, distinct LLM tropes and melodrama left completely unchecked. When someone claims total transparency while posting unedited bot-prose pretending to be a heartfelt confession, it just feels disingenuous.
My thoughts
I'm in the camp of either do it well or don't bother trying. That's probably why I abandon all my personal projects, that and because I have lots of ideas and the required experience to build almost anything that comes to my mind. I only work well in an environment where someone else accepts the responsibility of telling me to push forward because it's already good enough. I'm a perfectionist, not for fear of being judged, but because I am completely unsatisfied with the quality of software these days, especially cross-platform or vibe coded software.
I feel that if the user interface is such a huge chunk of your project that it requires a cross-platform framework, then your project should not exist as it lacks a useful business logic. Browser engines are somewhat exempt from this rule since browsers are all about user interfaces, but for other kinds of applications, there are no reasons to not go native. Rust is a native language that's perfectly suited for the business logic; it doesn't integrate well with the Swift Package Manager so it cannot be integrated properly with Swift libraries, but it does integrate with Xcode's support for external build systems so there's absolutely no reason to not write the cross-platform business logic in Rust and the UI in whatever happens to be the native language of the platform's frameworks for user-facing applications.
Then there's AI. I'm personally against using AI for writing code, not for fearing competition but because the code generated by AI is generally subpar compared to what I write. There are lots of good use cases for AI in development, like reviewing, debugging, and even learning, but generating code is not one of them. AI is terrible when it comes to structuring code, learning implicit conventions, reusing existing implementations in the codebase, and even deep thinking beyond what was in its training corpus, because it cannot do one thing that humans are good at which is to apply knowledge from one area into another, and that is because it lacks the deep pattern recognition abilities currently only present in the human brain. Even if AI was good at coding, current models are still totally unsustainable in terms of economical and power resources. I don't know if you used AI to generate any code for this project, but if you did that's one more reason for me to avoid it. The simple fact that you did not know what user agent string was being announced leads me to believe that you don't really know much about the codebase and thus its likely largely vibe coded.
Finally, a web browser is an extremely critical product that exposes a huge attack surface, so security should not be taken lightly. This means that I simply don't feel comfortable trusting anything other than a major battle tested browser. Modern browsers deal with way too much personal information and authentication to be taken lightly.
Regarding using LLMs for translation
I personally look down on anyone who uses LLMs to generate anything at all, and that includes translation or even just grammar. The existence of LLMs should not be an excuse to not learn proper English, which is not my native language either, but that doesn't stop me from constantly trying to improve my normative US writing and speech. Mental laziness is what bothers me about most people. If you don't want to put the effort into studying a subject, then don't try to project yourself as an expert when you are, in fact, on your toes.