Don't pay any attention to people giving you shit for using translation software. A lot of us sometimes forget that the whole world knows a little English and most of us native speakers have a ridiculous luxury in getting away with being two lazy to learn a few other languages.
I think it's good form to mention it as a little disclaimer, just so people don't take it the wrong day. Just write (this post was originally written by me but formatted and corrected with LLM since English is not my primary language).
From what I've seen, people generally do not like reading a generated content, but every time I've seen the author come back and say "I used it because it isn't my main language" the community always takes back the criticism. So I'd just be upfront about it and get ahead of it.
Part of the problems with using LLMs for translation is precisely that they alter the tone and structure of what you give it, writing using the LLM cliches and style, and it's unsurprising people see that and just assume completely generated slop. It's unfortunate, and I would probably try and use LLMs if English wasn't my first language, but I don't think it's as simple as "using translation software", I've not seen people called out in that way for dodgy Google Translate translations, for example, it's a problem specific to LLMs and the output they make having fundamental issues.
I wasn't annoyed about it, I just said it might be good to mention because people will notice anyway, and at this point there's enough AI slop around that it can make people automatically ignore it so it would be good to explain that. I'm surprised I got downvotes and pushback for this, I thought it was a common view that it's good to disclose this kind of thing and I thought I was polite about it
To be clear I think this has good information and I upvoted it, it’s just that as someone else said it’s good to get ahead of anyone who won’t like it by explaining why and also it can feel a little disingenuous otherwise (I don’t like getting other people to phrase things for me either for this reason but maybe that’s just me)
God forbid people actually learn the language they're trying to communicate in. I'd much rather read someone's earnest but broken English than LLM slop anyway.
It's disingenuous to call LLMs "translation software", and it's bad advice to say "don't pay attention those people".
Even if you don't agree with it, publishing AI generated content will exclude from ones audience the people who won't read AI generated content. It is a tradeoff one has to decide whether or not to make.
I'm sympathetic to someone who has to decide whether to publish in 'broken english' or to run it through the latest in grammar software. For my time, I far prefer the former (and have been consuming "broken english" for a long while, it's one of the beautiful things about the internet!)
I'd rather you write in broken English than filter it through an LLM. At least that way I know I'm reading the thoughts of a real human rather than something that may have its meaning slightly perturbed.
Hi HN,
I was evaluating IDEs for a personal project and decided to test Trae, ByteDance's fork of VSCode. I immediately noticed some significant performance and privacy issues that I felt were worth sharing. I've written up a full analysis with screenshots, network logs, and data payloads in the linked post.
Here are the key findings:
1. Extreme Resource Consumption:
Out of the box, Trae used 6.3x more RAM (~5.7 GB) and spawned 3.7x more processes (33 total) than a standard VSCode setup with the same project open. The team has since made improvements, but it's still significantly heavier.
2. Telemetry Opt-Out Doesn't Work (It Makes It Worse):
I found Trae was constantly sending data to ByteDance servers (byteoversea.com). I went into the settings and disabled all telemetry. To my surprise, this didn't stop the traffic. In fact, it increased the frequency of batch data collection. The telemetry "off" switch appears to be purely cosmetic.
3. What's Being Sent:
Even with telemetry "disabled," Trae sends detailed payloads including:
Hardware specs (CPU, memory, etc.)
Persistent user, device, and machine IDs
OS version, app language, user name
Granular usage data like time-on-ide, window focus state, and active file types.
4. Community Censorship:
When I tried to discuss these findings on their official Discord, my posts were deleted and my account was muted for 7 days. It seems words like "track" trigger an automated gag rule, which prevents any real discussion about privacy.
I believe developers should be aware of this behavior. The combination of resource drain, non-functional privacy settings, and censorship of technical feedback is a major red flag. The full, detailed analysis with all the evidence (process lists, Fiddler captures, JSON payloads, and screenshots of the Discord moderation) is available at the link. Happy to answer any questions.
I'm sure you didn't mean to, but you've crossed into being aggressive with another user. Please don't do that on HN—not with anyone, and least of all new users who deserve to be welcomed and treated charitably, not harrassed for not already knowing HN's arcane and rather primitive formatting rules*.
(First, thanks for this reply, which was nicer and more receptive than I was expecting!)
We don't want LLM-generated content any more than you do, and I'm confident that the vast majority of the community agrees. The problem is that there are lots of nuances yet to be worked out, so we shouldn't be heavy-handed.
Is it ok, for example, for a non-native speaker to use an LLM to fix up their English? I'd say it's clear that HN would be better off with what they originally wrote; non-native speakers are totally welcome, nearly always do just fine, and we want to hear people in their own voice, not passed through a mechanical filter. But someone who doesn't know HN very well and is nervous about their English couldn't know that.
People need to be treated gently. The cost of being hostile to newcomers, instead of focusing on what's interesting about their work, drowns out the benefit of enforcing conventions. It drives away people we ought to embrace. The community's zeal for protecting HN is admirable, but can easily turn unhelpful. The biggest threat to this site is not that its quality will collapse (it has been relatively stable for years now*), but that it will die due to lack of new users.
Like I said, I'm sure you didn't mean to have that effect. The problem is that most of us underestimate such effects in our own comments, so we end up propagating it without meaning to. The fact that several users replied indicating that they felt this way (edit: I mean that they felt your original post was too hostile) is a strong indicator.
* not that it's all that great, but these things are relative
Like a lot of the threads where you come in on my comments with "actually let's be a dick about this, I'll start" the rest of the thread seems to agree that some bad formatting in no way undermines an important bit of signal, that some language barrier adjacent friction is a tithe of noise to pay for it, and that erring on the side of tolerance around non-native speakers trying to navigate an English -dominated forum and industry is the better move.
I don't know how I got on your shitlist shortlist (I'll assume you don't do this with everyone), but give it a rest.
Hi HN, I was evaluating IDEs for a personal project and decided to test Trae, ByteDance's fork of VSCode. I immediately noticed some significant performance and privacy issues that I felt were worth sharing. I've written up a full analysis with screenshots, network logs, and data payloads in the linked post. Here are the key findings:
1. Extreme Resource Consumption: Out of the box, Trae used 6.3x more RAM (~5.7 GB) and spawned 3.7x more processes (33 total) than a standard VSCode setup with the same project open. The team has since made improvements, but it's still significantly heavier.
2. Telemetry Opt-Out Doesn't Work (It Makes It Worse): I found Trae was constantly sending data to ByteDance servers (byteoversea.com). I went into the settings and disabled all telemetry. To my surprise, this didn't stop the traffic. In fact, it increased the frequency of batch data collection. The telemetry "off" switch appears to be purely cosmetic.
3. What's Being Sent: Even with telemetry "disabled," Trae sends detailed payloads including: Hardware specs (CPU, memory, etc.) Persistent user, device, and machine IDs OS version, app language, user name Granular usage data like time-on-ide, window focus state, and active file types.
4. Community Censorship: When I tried to discuss these findings on their official Discord, my posts were deleted and my account was muted for 7 days. It seems words like "track" trigger an automated gag rule, which prevents any real discussion about privacy.
I believe developers should be aware of this behavior. The combination of resource drain, non-functional privacy settings, and censorship of technical feedback is a major red flag. The full, detailed analysis with all the evidence (process lists, Fiddler captures, JSON payloads, and screenshots of the Discord moderation) is available at the link. Happy to answer any questions.
By coincidence, I was just doing the same thing on the original comment. Since it's more or less identical now to what you posted, I'm going to move this subthread underneath https://news.ycombinator.com/item?id=44703481, where it will probably make more sense.
(Incidentally, mostly all we both did was add newlines, and it's on my list to see if I can get the software to do this automatically without screwing up other posts.)
I don't think the post you're replying to deserves what feels like an aggressive dressing-down. The reality is that post formatting on HN is really janky and it's hard to get it to do what you want. I don't blame someone for failing to master HN formatting on one of their first tries.
I've been here since like 2008 and I still screw it up sometimes. I'd have a not do it but HN comments are one of my "zero AI" zones. To echo sibling, its time to support markdown, its a standard now and even in arc, a reasonable subset is a pittance on anyone's time budget. It doesn't need to match GH bug for bug, that's a real project, but code formatting `backticks` is table stakes.
I wonder if its healthy to end up at aggressive dressing down is when we review peoples submitted work and point out the second half is LLM-generated 4 item lists.
