Hacker News new | past | comments | ask | show | jobs | submit login
Is Grammarly a keylogger? What can you do about it? (kolide.com)
464 points by terracatta on Feb 25, 2022 | hide | past | favorite | 293 comments



Years ago I worked at a company where many people were using Grammarly. One of the top devs took a look at it, and saw that the text was sent to Grammarly's server unencrypted and warned everyone not to use it. Some still did.

At my previous engagement, a large number of staff spoke English as a second or third language, and Grammarly was prevalent. Even as a native English speaker, they wanted me to use it as a sort of proof reader. I'll admit that it caught some of my dumber mistakes, but I never felt comfortable using it. I could have proof-read my work better is all. Perhaps if I wasn't given mind-numbing work, the quality would have been better.


My previous role was at a public facing ecommerce site. One day I started noticing a lot of public traffic to internal administrative endpoints that were failing - likely bots, but also to URLs that bots would have never known existed. Urls that only someone internal to the company would even know existed, due to the complex way they were crafted. It was very concerning.

We spent a LOT of time tracking down, and finally realized that the "bot" traffic was coming about 30 minutes after one of our employees legitimated visited the site. We found that user was using grammarly. Once we deactivated grammarly, all of the bot traffic stopped.

As best as I could tell, every URL that particular person went to in their browsers, grammarly had a service about 30 minutes later that would try and hit the url directly and ascertain what was there.

Haven't been on the crusade against it ever since.


Did something like grammarly attempt to correct your post here?

>Haven't been on the crusade against it ever since.

You have been or you haven't been? It sounds like a contradictory statement from the rest of your comment.


Because of the proven URL visiting bot, they don't have to. Everyone understands it's not an idle threat that Grammarly snoops on everything you write.


Why is Grammarly keeping track of the urls you visit when it has nothing to do with checking your grammar?


Are you writing blog posts or message board comments? Are you on social media? Writing to one person or many? Are you writing for financial, health, tourist industries, or for your academic qualifications? To entertain, persuade or inform? A screenshot of the page can be reviewed and classified later.

All to improve the service, of course. You know, what they say in the privacy policy.

(Note: I have no insider information)


We can only speculate but I notice Grammarly has a feature for plagiarism detection[0].

> Ensure your work is fresh and original by checking it against 16 billion web pages.

How do they know what text is on 16B web pages? Presumably they have a web crawler of some sort.

[0] https://www.grammarly.com/plans


>Presumably they have a web crawler of some sort.

Can confirm. Caught one of their bots on my site and called them out about it on Twitter.

They did not respond.


> Ensure your work is fresh and original

The page linked doesn't contain the text.


On a desktop browser, click the "Plagiarism detected" benefit under one of the plans - the text will show up as a tool tip.

That alone doesn't indicate they collected the 16 billion documents themselves, of course.


I mostly wonder how? Is it an extension or program? I thought it was just a website where you can paste in a text-box for proof reading, and that sort of website shouldn't be able to track you everywhere afterwards, right?


It’s also a browser extension and extension for things like Microsoft Word.

I think it’s fine if you use the website with information you don’t mind sharing but their extensions are reading everything you write.


Perhaps another site that person visited frequently was stealing their Grammarly auth token with this bug? https://bugs.chromium.org/p/project-zero/issues/detail?id=15...


> "Haven't been on the crusade against it ever since."

haven't -> have, right?


I believe they mean that the abuse was so blatantly obvious that they didn't have to argue the case anymore.


> wanted me to use it as a sort of proof reader.

I can't imagine working in a place like this. I often write with some unusual but perfectly valid grammar constructions that are a result of being well read. Running what I've written, the codification of my thoughts, through a statistical homogenization machine is dystopian in a way I had never imagined. What kind of business was it?

Imagine running famous writers through this thing, even if they're just journalists. Gross. I'm gonna run Moby Dick through Grammarly later and see what it has to say.


While I don't use Grammarly, this doesn't strike me as a very fair description of what grammar checkers are meant to do. Just like spellcheckers, they only make the changes you let them make. If you blindly accept every suggestion they make, then yes, I suppose you're running your text "through a statistical homogenization machine," but that's even more foolish than blindly accepting every suggestion a spellchecker makes.

I do use a "grammar and style check" built into Ulysses fairly frequently, even on fiction. This is a server-side service like Grammarly, but without the keylogging-ish aspects. (It only works in Ulysses, and you have to explicitly go into "revision mode" and click the "Check Text" button.) And, sure, it's optimized for business writing and is going to make a lot of dumb suggestions. But not all the suggestions are dumb; it really does catch genuine mistakes and even make style suggestions worth considering (e.g., "the phrase here is so overused it's in our database as an overused phrase, so are you sure you want to use it"). And, of course, sometimes I really am writing something for business, and some of the suggestions that are superfluous in creative writing might actually be useful.

I don't like Grammarly's approach explicitly because of the "send everything to Grammarly in the background", but I think people write off style checkers as pointless a little too quickly. Even the stupidest grammar checker, like the one built into macOS (sorry, Apple), will catch things like inadvertently doubled words, which is something I'm prone to do when I pause in the middle of writing a sentence to collect my thoughts.


> I do use a "grammar and style check" built into Ulysses fairly frequently,

I parsed this sentence as you running Grammarly on the novel by James Joyce, which would be an interesting experiment given that Joyce deliberately changes his style and grammar to mimic famous authors of the day, sometimes mid-sentence.


I have a strong suspicion Ulysses would fail miserably checking Ulysses. :)


> I often write with some unusual but perfectly valid grammar constructions that are a result of being well read.

That's lovely for you. It sounds like you're very clever. However, most writing in the workplace is intended to communicate concrete ideas, and benefits more from clarity than cleverness.


That doesn't have to mean, and in fact should NOT mean, bland.


I think one of the important things you should be thinking about is whether (given your complexity is already tripping up automated tools) it will start tripping up your coworkers.


Yes. Bland is better than "unusual but perfectly valid grammar" any day of the week.


Grammar checkers flag more than "invalid" grammar. Historically lots of arbitrary rules have been improperly applied to English to make it seem like a classical language. I like to enable checks for passive voice but completely eliminating it creates a stilted result and isn't practical, especially in technical writing. These tools are guide. I am free to ignore its advice when I know better.


... and yet you used an expression that seems unnecessarily idiomatic and parochial (British right?). I can easily imagine someone in Asia struggling to fit that literal expression to your intended meaning.

>> any day of the week


Bland poetry is bad.

Bland biz comms are not only acceptable but preferred.


Bland is boring and boring documents tend not to be read.

Succinct writing with some humor or at least some character to it will actually get that functional spec read.

Joe Spolsky wrote a good piece on this topic: https://www.joelonsoftware.com/2000/10/02/painless-functiona...


Sense of humor?

1) humor? it's subjective

2) you better be spot on because humor can easily be misinterpreted.

3) along the same lines, it's likely not everyone reading will be of the same native language and/or culture. Humor will likely marginalize them.

4) Frank Luntz "Words That Work". Chapter 1 is comms evergreen gold.

https://www.amazon.com/Words-That-Work-What-People/dp/140130...


Bland is forgettable. You don't want important biz comms to be forgotten. Bland doesn't spark thought in the reader. Clarity is achievable without being bland


You're talking about copy writing. That's different than peer to peer biz comms.


I hear clever and I think of the scene from Fight Club. Clever is not something good.

https://m.youtube.com/watch?v=PXcEPSUl0uE


I hate to break it to you, but you’re not writing Moby Dick at your office job. You’re trying to communicate a (possibly complex) idea in a way that another person can understand it and make a decision.

> Running what I've written, the codification of my thoughts, through a statistical homogenization machine is dystopian in a way I had never imagined.

This is why Grammarly exists.


And if you can't do it without Grammarly, maybe you shouldn't have that job.


My dude. I have worked with quite a few really smart engineers whose grasp on the English language was imperfect, though quite capable of communicating most points well enough. I would prefer it if I could take a little less time in my code reviews on copy editing, doing things like fixing plurality disagreements and false cognates.


An algorithm is isn't going to be able to clear up miscommunication more accurately than the person who receives the message.


How long did you test Grammarly before coming to that conclusion?


Even if Grammarly was forwarding its data to humans for manual correction, they still wouldn't have the whole context or industry-specific knowledge the person on the receiving end has. And even GPT3 isn't at this level.

It's the classic case of removing all of the apparent errors which then results in the downstream system underestimating the actual quantity of errors in the source and thus not being vigilant for more critical errors. Sure, it makes people look better to their colleagues, but what happens when they need to communicate in person, or in cases like mistaken context for double negative? The receiver will be so impressed with the level of communication they receive that they won't expect basic errors might be made.


Safe to assume that because you dodged the question you spent 0 time testing Grammarly?


The question is irrelevant because OP's not about grammar accuracy or comprehensibility but about domain expertise. Grammarly has no idea what you're talking about, only how. It tries to get around this with the web-scraper activity mentioned elsewhere in comments but obviously that's a bad way to do it for many reasons.


It’s fairly clear they misunderstand the domain:

> An algorithm is isn't going to be able to clear up miscommunication more accurately than the person who receives the message.

The whole point isn’t accuracy. Grammarly improves the clarity and accuracy of the message so another person doesn’t have to. The recipient can focus on the message, not applying their superior ability to clear up miscommunications.

So, I tend to agree with you. Whether or not this person has tested Grammarly is somewhat irrelevant because they clearly don’t understand why it exists.


Yeah, I'm looking at this and perhaps it's useful to distinguish absolutely useless from of limited use.

My guess is -- the improvement one might get from using Grammarly might be equivalent to "a faster version of, let me read a BUNCH of examples of similar writings to spark an idea of a better way to say a thing?"

In other words, it feels like the thing that Grammarly might be useful for probably lean "syntactical" over "meaningful?" Something like that.


Miscommunication destroys productivity and in the worst cases ruins end products by corrupting the development cycle.

Ultimately an organizational problem but I gotta say it gets very peeving to write 5 different solutions to the same ask because the other person is careless with their words.


Odd, I thought I responded elsewhere with this point -- but yes, I'd refine my point by saying Grammarly is probably a bad idea for internal communication, but perhaps fine where writing is the product.


And if you can’t code without Google, maybe you shouldn’t have that job.

Tools exist to help us. Use them.


> I hate to break it to you, but you’re not writing Moby Dick at your office job.

What if he does? Do you like generalization?


Read his writing. Moby Dick isn’t coming out of his keyboard unless he’s retyping it.


You're missing the value-prop of Grammarly -- to help someone who doesn't know write more like a fluent native English speaker. Helping someone write "more average" is exactly the point until they're fluent enough to know when and how to break the rules.

Wait until you find out about the 5 paragraph essay that's taught in HS that follows this model.


> who doesn't know write

In my experience it doesn't.

It helps people who are already reasonable good at writing English to write slightly better. Through only if their writing is limited to "business" English.

For e.g. papers it's in my experience a catastrophe.

There are also multiple categories of errors it can't cope with and does bad recommendation for. It's the kinda of errors I'm doing a lot. Maybe due to having some dyslexia, maybe because it's my second language, or maybe because my brain thinks slightly different (not joking; Luckily it's just different, not worse.).


Ten years ago I went back to school and finished my BS. All of the instructors required us to run our work through a plagiarism checker and certify that it was our original work when turned in. The plagiarism checker was part of a suite of tools that included a grammar checker, and some instructors used the grammar checker themselves when grading student work.

I always viewed the grammar checker as a crutch and only used it when required, but some students I worked with on team projects were obsessed with getting no warnings from the grammar check. There were a few ESL students that produced almost unintelligible work that had passed grammar check, and they argued with me when I tried to help them.

My best experience was when one instructor ran my work through the grammar check and then sent me the graded paper (A) which included his long critique of the changes the grammar checker had recommended.


> You're missing the value-prop of Grammarly -- to help someone who doesn't know write more like a fluent native English speaker.

That's not a part of the value of Grammarly. The tool is going to recommend mistakes; if you can't recognize them, then it will hurt you a lot more than it helps.


   I can't imagine working in a place like this. 
I couldn't either. I don't work there anymore.


> Imagine running famous writers through this thing, even if they're just journalists. Gross.

Try comparing the text of a Great Illustrated Classic book to the text of the actual book it "adapts".


The goal is not to make famous writers write like journalists. It is to make people who struggle with writing write at the level of journalists.


I only know this company as the annoying YouTube pre-roll ad company. Any company that needs to advertise on YouTube that heavily raises red flags, IMO. Not surprising to hear they are cretins. For what purpose do they use the keylogging data. Doubtful they tell their customers all the exact details. "We use the data to improve the service." Gimme a break.


I'm starting to form a really negative impression of YouTube ads as well. Lots of scammy stuff all over the place. I'm kind of intrigued by all the "This young scientist invented ...", "This single father came up with a solution that his boss wouldn't give freedom to explore and so he broke out on his own to improve the world...". All these thin and exceptionally vague human interest stories filled with anonymous non-identifying biographical information leading up to some purported great product.

At first I was fooled by those productions since they had some non-zero amount of polish to them that looking back I realize is professional stock footage of some kind. Now I know it's junk faceless import products pretty much that are being peddled. There's a whole genre to them and it's funny now to see past them immediately when they appear, but irritating too.


Who still sees YouTube ads (or any ad for that matter) on todays web? I’m thinking anyone using the web without an adblocker is either a masochist or a fervent disciple of the “adtech” industry.


"Who still sees YouTube ads (or any ad for that matter) on today's web?"

Someone testing homegrown methods of blocking online advertising traffic on their own network.

I never see ads except when testing. I do not use an adblocker for downloading and watching YouTube videos. I use methods that work outside the browser. Adblockers require "modern", graphical web browsers.

"Modern", graphical web browsers are funded by online advertising.

I do not require a browser for accessing YouTUbe.


So, both of my phones have the following keyboard option:

"Suggest text corrections

Tap words or phrases underlined in green or tap the more menu icon when you see a green dot, to review grammar and writing suggestions.

Powered by Grammarly"

Does this mean that if I have text correction turned on while using the keyboard on my phone, because it is "powered by grammarly", it will be sending unencrypted information to Grammarly?

Personally, I couldn't care less if some giant company is reading my information, but I don't want that transmission being sent unencrypted.

Edit: I should mention that this is not an app that I have installed on my phone. This is literally just the text prediction for my keyboard by default.


Looks like Samsung is adding the Grammarly functionality to their built-in (still an app, technically) keyboard in an upcoming update, or they already have.

So now, very likely, both Samsung and Grammarly have access to everything you type on your phone.


Thankfully there's a few open source keyboards.

I've been checking out FlorisBoard, available on fdroid. Sadly, Google and Samsung keyboards are very good at swype/glide, FlorisBoard is coming along though, & I prefer to support it because the other two can't be trusted.


Both phones I was referencing were Samsung devices.


Do you know that grammarly keeps this information and doesn't toss it after it's AI looks for patterns for training?


I'm not aware of any actual AI system for NL using on-line training -- this unlabeled, after all! So "after it looks for patterns" is more "after the next training batch runs", 30 days if they're GDPR compliant.


Yes, all your text is being sent to Grammarly servers. It's done over HTTPS, so no third party can see what you type in, only you and Grammarly see your texts.


That's really bad actually. Especially if they are storing it away in a database associated with your unique ID for surveillance for 3rd party companies and the government.


Hmm, so if I understand things correctly: The means of transmitting that information is (supposed to be) secure, but the information within it - if the means of transmission (HTTPS) is compromised in any way - means a third party would then be able to read that information. Right?


Yes and grammarly obviously reads it, since it works over the Internet. Do they store it? Probably not but read their TOS and obviously you have to trust them as well.


Why wouldn’t they store it? That’s a good mine of information in many different ways (product improvement, upselling, ad targeting, etc)


…until there is a security breach at grammarly


I feel like I can't function without it. Grammar mistakes can be seen very badly by others. If only being able to understand the other person was sufficient.


Grammarly has become a crutch and I am having a hard time finding an alternative.

This is my method to write grammar sensitive content -

- Write initial draft in notepad

- Review the document

- Paste content into grammarly

- Review their suggestions

- Review the content one last time.

I just couldn't find anything that is non obtrusive and secure.

Edit: I read through the article. My method seems to be the most secure way to use grammarly. But I am deleting the extension and the app.


I've been using ProWritingAid instead of Grammarly. They have discount deals every once in a while, incl the Lifetime sub.


I have a relative with severe dyslexia who is a native English speaker. I wonder if grammarly could help him, even with things like spelling, too.


It does, it's not just that though, it also helps with picking the right words and phrases for your target audience and what type if a message you are writing.


Feedback loops? In my product copy?


What does it even offer that MS Word doesn't? Word has all sorts of grammar and style checking.


The obvious one: integration with every text box on your system.


Which is exactly the very very scary prospect of the thing. Giving Word this ability while using Word makes sense. Giving some other software access to 100% of everything that occurs on the system is very unnerving. <shudder>


For all the years that I used Word, its grammar checker used incorrect rules for when to use "which" and "that".

This issue may be fixed now, but I would never trust a grammar checker with such a lousy track record for basic issues like which/that.


Doesn't HTTPS encrypt your traffic to Grammarly's server?


Why would that matter? The traffic is encrypted in transit but still received as unencrypted plain text by Grammarly.


> the text was sent to Grammarly's server unencrypted

What does this mean exactly? That they were using HTTP instead of HTTPS, or some custom unencrypted format?


It may have nothing to do with the protocol. It can mean that everything Grammarly receives, no matter how transmitted, is unencrypted, in the same sense that this comment I sent to HN is. Of course this makes sense, but I think the difference is in the expectation of what is and isn't sent.


Years ago, you used something else instead of Grammarly.


I think worse than keyloggers is that people are learning how to make the yellow lines in Grammarly go away rather than learning to write better. The training of humans on AI which was trained to be a (dull) average of prior humans has unforeseen consequences. I've seen Google grammar suggestions getting worse with time.


This may be the case, but "this tool is a potentially massive privacy and security intrusion" is a drastically different conversation than "calculators mean students can't do arithmetic anymore".


I think "calculators mean students can't do arithmetic anymore" is perhaps an uncharitable take because it ignores that calculators aren't also trying to learn arithmetic from examples of people using them. Eventually grammar correction algorithms will injest text written with them or other grammar correction algorithms as ground truth in their efforts to improve and adapt to new idioms- this may already be the case.


NLP researchers are tearing their hair out about this right now, since people are posting mountains of GPT/etc.-generated text online with no easy way to distinguish whether it's of human or other origin.


Reminds me of a scifi book where the Internet-analogue is so corrupted with junk deliberately injected by filtering services so that they can sell you the filters that it's impossible to use "naked".

I think it's either Neal Stephenson or maybe Stephen Baxter, but I'm not sure which book it was an aside in (it's not Fall, I haven't read that yet, though that appears to have a similar idea).


I guess we're entering a new phase of language evolution, then; whether we want it or not.


Maybe we should create a neural network model that could label arbitrary text as -- nah!


That's a Gan but with more steps right?


Well... you can read it.

I might be wrong, but so far I think it's been pretty easy to tell if text came from a human or a deep-learning system.

Granted, that probably doesn't scale well.


That has been true up until very recently, but lately there has emerged an uncanny valley that has confused the distinction between "underpaid freelance (ESL) writer farm" and "shoddy but roughly convincing neural language model" so that you may be convinced the blogspam you may happen upon across the internet may just as easily be computer-generated as anything else.


If GPT-3 et al. are producing results on par with "underpaid freelance (ESL) writer farm" then wouldn’t the next step be to focus only on better writing? i.e. books, newspapers, magazines, etc.

Obviously the corpus (perhaps 1 trillion to to 10 trillion words?) will be exhausted by a sufficiently large model so there’s an upper bound.


Okay, interesting.

Thanks for sharing!


Most, not all.


This sounds like a case of "garbage in garbage out". Machine learning needs to account for that regardless.


Writing and arithmetic are a lot different. Activities involving math are very rarely done away from a phone, and computers are for all intents and purposes 100% reliable at it.

The best mathematician in the world will probably get the same answer I will on a calculator. At most maybe I'll get an irrelevant rounding error.

If there was only one good way to state any idea, no matter the context or intent, it would be different.

I could choose not to use grammarly if I had something to hide, with little trouble. It doesn't do anything most can't do themselves, or anything a local FOSS checker can't do once browsers start integrating them as they probably will someday.

The real danger would be a world in which I can't legally choose not to use it or for some other reason it is no longer a free choice.

Is journalism not part of how we prevent that?


This is comparable to the argument that programmers are learning how to eliminate compiler/linter warnings rather than learning how to program better. Immediate feedback is the best learning tool and one leads to the other. Of course, natural languages are far more complex and have no universal standard. While it is true that no easy way to check for style, be it in programming or writing, we should encourage the use of such tools (once they are secure and privacy-friendly) as writing assistants.


No, it’s comparable to your compiler/linter suggesting fixes when it spots issues. Pointing out an error is one level of knowledge, suggesting a fix is not in the same class. Take away the grammar tools’ ability to make suggestions, and you have something equivalent. You still have to know how to math/program with a calculator/compiler.


I couldn't agree more. Every now and then I let gmail auto complete for me, but it feels like my writing is just becoming generic.


Yet, I've used the same closing line for my emails for years. And Gmail still doesn't suggest it as completion.

"AI powered".


English is not my first language. Grammarly greatly improved my English, even when I'm not using it.


One of the beat ways to tell really peoficient foreign speaker of, in my case, German isn't the accent (as some people don't have one anymore) but a too perfect Grammar usage, either in writing (less obvious) or speaking (clear as day). I yried one of the German grammar tests of my son, and failed miserable, despite being a native speaker. My wife isn't, and she's so much better in German grammar then I am.

The gist: Don't worry too mich, just use the language. Most people are than delighted to meet a foreigner trying to speak their language.


"Perfect grammar usage" is only a negative when you have "perfect grammar knowledge", whatever that means. I'm actually knowledgeable enough to to be more spontaneous if I want. In reality I only use Grammarly on the desktop, which is only a tiny fraction of my usage of the English language. Things are not as black and white as they seem. I'm neither a child nor a robot.


It's the same with English. Maybe it's confirmation bias, but I've noticed many non-native speakers either write too formally, or choose the wrong synonym for the context. Everything is grammatically correct, but it just feels wrong.


Non native speakers are generally more formal regardless of their usage of formal. I'm autistic so I'm highly formal in my native language as well!


My wife told me that when she was studying Russian in college, some of the professors in the advanced courses would refer to certain phrasings as the "spy's variant."


You might be worrying a bit too much about what others think. I am English and if someone is, say, Russian, and writes to me in English, I don't care if their grammar is a little poor. The great thing about English is you can understand even when most of the words are in the wrong order :-)

Also, most English people don't know much about grammar anyway and lots of people still confuse things like there/their/they're; are/our; its/it's etc.


I think you skip the case when both communicating parties are not native speakers which is probably much more often case than when at least one side knows English well. And risk related to a misunderstanding caused by breaking grammar rules is much higher.


Also if you're not native speaker how do you know if the computer's suggestion is an improvement or a regression?


It is possible to have relevant knowledge in languages that are not your own.


Totally agree, we already have a generation of folks who don't know how to spell without the crutch of a spell-checker (I include myself as a victim), does Grammarly produce even worse outcomes?

The one caveat here (which I try to cover in this post) is there are definitely people who suffer from things like dyslexia who heavily rely on these types of tools to be able to communicate confidently. In that way, they are very useful.


Even worse: entirely new language primitives based on 10wpm mobile keyboards.


Despite the moral panic over the OG mobile keyboards in the noughties (remember the breathless "kids these days write txt spk, m8 in their English exams" headlines every year?), the pandemic of inability to "write good" never actually materialised among Millenials.

Indeed, it has been suggested that reductions like "wait" to "w8" even represent the same kind of phonological awareness of language that's correlated with better spelling.


The same thing happened in the 19th century when coded abbreviations to save money on telegrams became a fad and infected everyday writing. Newspapers and periodicals both celebrated and decried them. Civilization didn't collapse. TTFN.


Well WWI happened towards the very end of the telegraph age, so civilisation did collapse.


What could be more civilised than murdering thousands at a time with science (e.g. gas attacks and aviation)?


I mean "txt" speak is something only my parents still use. There are some shortenings like omw, ngl, ikr but they're not inventing new words.

I mean hell if this is the straw than the internet ruined that before I was born -- tldr, imho, afaik, mfw, /s.


Heh, now it's just emojis and memes for communication.


does it actually matter that people don't all the grammar rules or how to spell things? an english sentence contains a lot of parity data. when you do know all the rules, it can be painful to see the mistakes people make, but confusing "principle" with "principal" or using "who" when you should have used "whom" doesn't really obscure the meaning of the sentence.

when it comes to formal correspondence, spellcheck is always there to help. or better yet, get a copywriter or technical writer to help you and go back to your main responsibilities.


> does it actually matter that people don't all the grammar rules or how to spell things?

It does if you hold the view that teaching someone to speak and write is teaching them them think.

If that is correct than if someone does not know how to speak and write in a grammatically correct fashion then the implication is that they do not know how to think properly.

Even if there isn't a perfect 1to1 mapping between grammatically correct writing and thinking skills, I still think it's a good proxy for measuring a persons ability to think because in general the more you read, the better you get at writing, and the more you read, the more you know.*

* the traditional caveats apply with garbage in, garbage out.


> It does if you hold the view that teaching someone to speak and write is teaching them *them think*.

> If that is correct *than* if someone does not know how to speak and write in a grammatically correct fashion then the implication is that they do not know how to think properly.

And you immediately defeat yourself with your own argument.

Nobody ever has, nor should strive to, follow every grammatical rule perfectly. Errors only matter if they create actual ambiguity; if you understand my intent, then I have used language effectively.

As human thought evolves, language should too. Poets are always breaking rules in the name of art, and many of those changes get codified as new rules. Shakespeare simply made up dozens of words that we use today without a second thought. Over the last decade, modern poets steeped in the culture of sarcasm have given the word "literally" a new meaning. Those deviations and inventions made sense, both to the speakers and listeners, so new words, rules, and understandings were created. The only people left confused are the prescriptionists who cling to outdated rules that describe how people used to talk.

As long as humans can turn these AI assistants off and share their imperfect creativity, I'm not too concerned.


> Nobody ever has, nor should strive to, follow every grammatical rule perfectly.

Actually, this is untrue; people automatically follow the rules of their language with extreme accuracy. They don't "strive" only in that following the rules doesn't involve any conscious effort.

This is behind the principle articulated for HTML of "be strict in what you emit and liberal in what you accept". That is how people behave when speaking. But it's not something they're capable of doing in HTML, and trying to apply it there was a mistake.

> Errors only matter if they create actual ambiguity; if you understand my intent, then I have used language effectively.

This isn't right either, in a couple of more serious ways:

First, ambiguity is always present regardless of whether any errors have been made or not. A fun example of ambiguity I noticed recently is "Flemish is more influenced by French than Dutch". Perfect grammar, but the two obvious parses have opposite meanings. Ambiguity arising from a grammatical error is no more serious than ambiguity arising from a lack of errors.

Second, the major problem with failing to follow the grammar of a language is that it prevents your audience from knowing whether they understand you or not. One thing I can say if you speak to me in broken English is that you aren't aware of the rules of English. Given that, even your apparently flawless sentences might have been intended to mean something quite different from what they would mean if spoken by an English speaker.


> Actually, this is untrue; people automatically follow the rules of their language with extreme accuracy. They don't "strive" only in that following the rules doesn't involve any conscious effort.

You can't possibly be serious! If this view is based on your real-life experience then I am amazed at the rarefied circles to which you belong.

https://www.thinkimpact.com/literacy-statistics/

- Nationwide, on average, 79% of U.S. adults are literate in 2022.

- 21% of adults in the US are illiterate in 2022.

- 54% of adults have a literacy below 6th grade level.

- 34% of adults who lack proficiency in literacy were born outside the US.

Based on these stats, you can rest assured that at least half of the adults in the United States do not have perfect grammar. And although "proficiency" is not precisely defined, I think it's safe to assume that a majority of these low-literacy adults are native speakers.


> If this view is based on your real-life experience then I am amazed

How much effort did it take you to select the form 'am' in 'I am amazed' while selecting 'is' in 'this view is based'?

When's the last time you heard anyone make a mistake in conjugating be?

That's what a grammatical rule looks like.


People always be conjugating "be" incorrectly. It's actually hallmark of a very common English dialect, which people speak and understand all the time without issue:

https://abagond.wordpress.com/2008/09/09/ebonics-time-and-te...


> if you understand my intent, then I have used language effectively.

I'll agree as far as it goes, but it ignores the burden of understanding on the the part of your interlocutor.

Have you ever been in a conversation where either you or the other party didn't speak the language well? It is a strain on all parties, even if sufficient information was transmitted and understood. I might consider this merely sufficient or barely "effective". I'd hope we aim for clearly expressing oneself without causing strain on your conversation partner or reader.

A good deal of what people quibble over regarding "grammar" has more to do with good style (whatever that means to you -- proper use of less/fewer, who/whom, capitalization and punctuation, etc) and achieving a certain verbal register.


You're not even wrong.


> It does if you hold the view that teaching someone to speak and write is teaching them them think.

I don't hold that view. at least, I don't think effective communication requires mastering the formal rules of a natural language. I'm not even sure communication itself is that closely tied to complex thought. I've known a few brilliant engineers whose english was barely sufficient for work. perhaps they were quite elegant in their native tongues; I wouldn't be able to say.


> I'm not even sure communication itself is that closely tied to complex thought.

I think it does in some respect; our language's grammar and vocabulary litteraly govern the way we see the world, after all[1]. As a personal anecdote, I'm ESL and write as a hobby, both in my native language and English, and I do feel the differences (and limitations) in the way my mind works depending on language.

I know that my inadequacies in English mean that my style is different. Descriptions and reasonings are simpler. In my native language, I can easily use word plays, ambiguities, be bolder and "more clever" (or less stupid :p). In English, I unconsciously keep things less complex. I'm a different author, and so the worlds I describe in my texts are simpler, too.

English is also far more "pragmatic" than my birth language. I can't word things the same way. Thus, my thinking as an "English writer" is far more rational and cause-to-effect-ish.

It's like any tool: you can absolutely use a programming language through tutorial and google searches, but you have to truly master it to really understand its concepts, to really use it to do something cleverly and intelligently -- and you won't plan a complex task the very same way in two different programming languages.

[1] https://www.sciencedaily.com/releases/2011/03/110314132531.h...


does it actually matter that people don't all the grammar rules or how to spell things?

Does it matter if someone does not write with perfect grammar and spelling? Probably not.

However at some point you have drifted so far from normal conventions that you are no longer communicating effectively. You can make reasonable arguments about both prescriptivism and descriptivism but if your spelling and grammar are so bad that someone can't understand you then you're not a descriptivist, you're just wrong.

I have encountered this more often than I'd like in professional settings. I work in software development, a field where precision is important, so if I'm looking at your job application and it's full of basic language errors then I absolutely will judge you for that.


Yes, clarity and grammar matter. For example, with your lack of capitalization, I wouldn't know if you were trying to help your Uncle Jack off a horse or your uncle jack off a horse.


Though no one has ever written either of those sentences for any reason other than to put up a strawman to tear down in their defense of grammar.

I'm a fan of learning grammar well, to be clear, and wish mine were better, but this is not a good argument for grammar mattering.


okay, you can also come up with plenty of examples of ambiguous parses that don't require violating any grammar rules. fortunately humans are a bit more intelligent than compilers and can use their knowledge of context to settle on the more reasonable interpretation.


But why make other humans do extra work to parse you? It’s arrogant in the extreme to say ‘that’s good enough you figure it out’.

Even scanning text is jarring when the rules aren’t followed, takes extra time to comprehend and adds to the minor irritants of daily life and will have people judging you, even subconsciously, as unintelligent.


> does it actually matter that people don't all the grammar rules

No.

*ducks*


ironic, but it actually supports my point so I'll leave it that way :)


I for one blame the Grammar-Nazis and their anti-semantic ways

ok, i'll let myself out.


Good writing is probably like dressing up. It's not just what people think of you, you think of yourself better too.


I know, we should also get rid of Calculators because people should just do all math in there head

Hell let get rid of computers as well, people should just do everything manually, if you need to communicate drive to the person an talk to them, if you need to write something get out the hammer, chisel and rock...

</endsarcasm>

No Computers, Spell Check, nor grammar check has not ruined civilization or made people stop learning things...


we use calculators because you can't do the kind of arithmetic in your head that calculators are made to perform, and because random arithmetic isn't conducive to understanding math. If you can't add two single digit numbers any more I'd start to be concerned.

Having a proper understanding of grammar and spelling is relevant even in verbal communication. Tools should be used to augment human capacity, not used as an excuse to justify atrophy of basic skills.


You can see that's false by the existence of languages with simpler grammar than English whose users are still able to function to as high or a higher level than English speakers. Some of the grammar in English is redundant. We shouldn't waste our brainpower on it any more than on mental arithmetic/etc.


I don't necessarily disagree but that doesn't have much to do with my point. Simple or complicated grammar, you as a person having command of it rather than outsourcing it to a tool is a skill you need often.

It's irritating how often people even confuse "your" and "you're" nowadays. Some simplifications are harmless but plenty of them can change meaning. And it seems like texting and autocompletion makes people care less and less.


I use it, but only as a suggestion.


Having been subjected to a thrashing early on by a really good editor, they will stress that their isn’t always to correct and re-write your work - they’re often there to mark areas that need clarity or don’t fit the norms of editorial standards.

If they just fixed everything for you, you wouldn’t improve at all.

This experience made me much more careful about the way I write, and who I write for.


my grammar actually improved from proof-editing my essays on grammarly


In Québec we have Antidote. It's a good "old" piece of software you install on your computer and it integrates with office and other programs to provide _explanations_ of why what you write, looks wrong. No keylogger, no getting dumber. It essentially let's you either learn why you're wrong or decide the software is wrong. It used to be French only but they added English too a couple years back. I have no affiliation with Antidote, I just use it everyday.


Another older piece of software is the Unix Writer’s Workbench, which legend says put all other grammar checkers to shame. Unfortunately it seems to have been lost to history and I am too young to have used it.

https://en.m.wikipedia.org/wiki/Writer's_Workbench


Looks like a good floss project to contribute to:

> The GNU operating system contains free software implementations of several wwb utilities, such as spell, style and diction.[9] As of early 2019, the look utility had not yet been ported to GNU, but its implementation from 4.4BSD-Lite is available as free software, for example via Debian.


Thank you for this recommendation. I'll probably pick this up for French writing practice!


Well, it doesn't seem to be working:

> It essentially lets you [...]


I use it everyday, but I don't use it for everything.


I'm on the team at Sapling Intelligence, a deep-learning AI Writing Assistant. A lot of privacy and security conscious folks don't like the idea of a keylogger, so we have self-hosted/on-premise/cloud-premise options for businesses. We have a list of available offerings here: https://sapling.ai/comparison/onprem. Sapling deployments can also be configured for no data retention, sacrificing some model customization.

Cost-wise, it doesn't make sense for individuals to host a neural-network based grammar checker, though some of the rule-based options may work. There's a future where if we can maintain some sort of Moore's law scaling we will be able to run these language models on individual computers as opposed to the cloud.


> Cost-wise, it doesn't make sense for individuals to host a neural-network based grammar checker

Why?

We already do the same in the photography world in the form of apps like Topaz Labs' denoise/sharpen/gigapixel, as well as video enhance. Why would I care how many gigs of disk space and even a GPU might be required for an NN grammar checker if it literally makes back the money by improving the writing that influences my career? Hell, I can expense what is needed to run this if the payoff to my company is "the quality of work is better, and more secure".


Well, I think you'd first have to know the resource requirements, and it's reasonable that so few people would be willing/capable of running it that it doesn't make much business sense to focus on that as an option.

I'm certainly curious to know.


I'm not an expert and would appreciate being corrected if I'm wrong, but I'm under the impression using a neural network after it has been trained typically requires relatively little computation and data. It's training it that takes the big compute and requires lots of data.

I think many services like Grammarly would be perfectly possible to implement without sending your data off device. There are just massive incentives not to.


Not really, that's an assumption based on a misunderstanding about how NN actually work.

While the training (e.g. the part necessary in order to even have a product) is incredibly resource intensive, and will yield a model that's typically hundreds of megabytes, actually applying the resultant model to new data takes milliseconds at most for plain text analysis.


fixed that for you:

A neural-network based grammar checker doesn't make sense.


While I don’t use it for the sake of privacy, folks saying “learn to write” are missing the point of grammarly. It’s an editing tool. Editing is remarkably difficult to do on your own writing. Ask any published author.


It is especially funny in a community with as many software engineers as HN. Imagine saying a developer who uses a linter or has to make changes to their code as part of a PR review should just "learn to code".


Honestly, that's a thing. The Go "readability" process at Google involved static analysis commenting on your code, and you got "readability" (the ability to approve other people's PRs) when it didn't have anything to say for several changes (among other more human criteria). You couldn't run this linter yourself before it was auto-run, so it did act as a check against making the same mistake over and over again on your own. (I think I figured out how to run it, though.)

This sounds like it's a bad story, but I thought it was a decent system. (I was a readability reviewer for Python, and I wish I was as consistent as the bot.)


I regularly use such tools (usually when writing formal stuff) as I usually mess up my grammar, tense, punctuations and some other stuff.


I used Grammarly in the past and I stopped because the privacy issues were concerning. I switched to https://www.antidote.info/en, which works entirely on-device, without sending your data to a cloud service. They now do offer a fairly minimal web application that can be used if you have their subscription, but they offer a one-time purchase for the desktop application.

Similar to Grammarly, the growing use of AI-based pair programming tools, like Github Copilot and similar, poses similar serious privacy risks. While the intelligent autocomplete is helpful, it uploads large parts (or all) of your source code; which most companies should be very concerned about.



Yeah, it still absolutely blows my mind folks allow Grammarly anywhere. It's horrifying from a privacy and security standpoint. I get requests to install it at work from time to time, and then have to basically explain that it would be illegal for me to allow it.

I would argue if you're subject to any sort of data security compliance policies, you can't allow Grammarly on your systems.


I would never use it but this feels like a common issue in general which is trading convenience for privacy and security.

This happens with a lot of choices. Smartphone vs flip phone, using an Alexa or similar device vs not, etc..

A fully offline Grammarly would be a really nice app to have. I'd pay something like $49 a year to have access to that if it worked really well and was kept up to date. The hard part is "really well", it's so much more than detecting spelling errors with a Vim plugin. At the same time I have a feeling the "goodness" of a tool like this is only because of how much data it can harvest to train its algorithms and models which makes me think it won't happen until accurate models are in the public domain.


Here is my OG Grammarly (offline):

https://www.antidote.info/en?

Have been using them for over 10 years.

Cannot recommend enough.


LOL. This trojan is 5 years old.


What makes it a trojan?


Hah, I use an old DOS grammar checker called Grammatik.

It works well enough for me, I use it with mutt instead of ispell. Naturally, it's 100% offline.

I made a (terrible?) unboxing video a while ago: https://www.youtube.com/watch?v=6DMlaJ-ROXc


One can run language tool [1] also on premise (or directly on one's machine).

[1] https://github.com/languagetool-org/languagetool


I like to use write-good[0] - it takes a glob and prints suggestions to stdout.

[0]: https://github.com/btford/write-good


I saw an ad for Grammarly once, maybe a year ago, on YouTube. I was intrigued and clicked on it to learn about it and concluded that it was pretty awesome... until I learned about how it works and NOPE’d out immediately.

Since then, though, they aggressively bid to put ads on my browser. It’s a win-win for me, the ad networks get to feel like they’re delivering highly relevant personalized ads to a potential buyer and I get the peace of mind of not seeing other ads and the feeling of doing a good deed by baiting Grammarly into wasting ad spend. Selling a key logger as a service—- it’s an abomination.


You may have stumbled on a new form of activism: ad baiting. Wonder who pays the most for ad placement that's also deserving of being drained (in a small way) of funds.


I think they just had campaign. I never clicked on it and had tons of grammerly ads at around same time.


Isn't it plainly obvious that they take all your inputs and feed it to their models?

Isn't that literally the point of Grammarly?


I think it's more a case where many users only see one side of a dual-use technology...they don't necessarily even know how it works, just that they can install it and it checks grammar for them. It seems a bit like visiting the local dump for the first time as a kid: the sheer amount of stuff being collected in one place that came from everyone's homes can be really impressive, even though you knew all along that it had to go somewhere every time you took out the trash.


For some of us it is something we might think is likely.

I don't think it is that obvious / people would know that outright.


It's not obvious (to me) that all the "thinking" is done in the cloud. I would've assumed it was all local to your machine.


Is it a native app, an Electron app, or a webapp? I don't even know because I refuse to get anywhere near it.

If either of the last two, it seems inevitable that their engineers would push as much of the compute server-side as possible to retain fidelity at the keystroke level.

We really should be doing a better job about demonstrating the distinctions between locally-run and -hosted tools, and cloud tools, for novices and laypeople.

Back when Facebook started pushing everything server-side it was a pretty obvious change if you were on a bad connection. When you'd type in a form on a bad connection it would often re-populate text that you had just deleted, or move the cursor around to unexpected places, presumably to match back up between their server-side caches and the local state. In the early days my adblock was interfering with this, too, which made the issue more apparent and caused some very strange behavior.


If you don't want to get closer, you can always check it in a third party resource like Wikipedia. There it says that it's cloud-based, and that it's both an app for platform and a browser extension. No need to wear ignorance as a badge.

https://en.wikipedia.org/wiki/Grammarly


Is there a list of Grammarly servers, so it can be blocked, for example, at a corporate firewall?


We had a ticket where someone complained about "latency" when cutting and pasting into Microsoft Excel.

After some quick tests yielded the expected results, the person selected an entire paragraph of text and pasted it into a cell.

Sure enough, there was about a 3-4 second delay as the Grammarly add-in sent it in for analyzing.

Imagine if you had confidential data in your documents.


Yes. Don’t use it, and require that contractors not use it either.

I’ve found attorneys using it - automatic dq for me.


The article explicitly calls out that kind of behavior under the subtitle "Don't Just Ban Grammarly and Expect Folks To Listen". I know, the whole article is sort of promotional, but the point still stands.


People say stuff like this all of the time. It’s just noise unless your relationship with the provider addresses whatever risks you have.

You don’t just ban it, you attach sanctions to it. Third party disclosure of legal work product is not so good for your legal career.

Would you be happy if the people administrating your health benefits, medical records or taxes were using some random free SaaS that accessed and processed your data without accountability?

How about your attorney? If you were facing criminal prosecution, would you want the information and metadata (how documents were revised) about your case to be subpoenaed and available ton the prosecutors?


The whole article doesn't gloss over those points; it is rather critical of Grammerly except for the last part. But also it, in my opinion, correctly points out that the security measure can't be separated from the social measure in that last part.


Does dq mean dequeue? Ironically, Grammarly would automatically expand your in-house abbreviations to make your writing more clear.


Sweet.

I’m sure the subject of a confidential investigation that could result in losing custody of their children would appreciate that Grammarly simplified that jargon in some attorney’s report. No problem at all that it’s stored in the cloud somewhere, monetized, or reviewed by some contractor somewhere.


DQ = Disqualification, a no go.


>Does dq mean dequeue? Ironically, Grammarly would automatically expand your in-house abbreviations to make your writing more clear.

I thought it meant "Dairy Queen[0]." Yummy[1], yummy Dairy Queen!

Using shortened/abbreviated forms without explanation is a recipe for misunderstanding.

Generally, writing out whatever will be abbreviated/shortened and noting the abbreviation/shortening is preferred, as it removes the ambiguity of just using an abbreviation.

For example, If I say that I'm going to "The Met," that could mean I'm going to the Metropolitan Museum of Art[2], the Metropolitan Opera[3] or The Metropolitan Police[4]. Which institution is implied may be clear from the context, but could be confusing.

Many years ago, I shared a taxi with a couple on my way to work at American Express. When I said I was going to "Amex," they assumed I was going to the American Stock Exchange and since their offices were quite close to that organization's offices, we decided to travel together.

Fortunately, the offices of American Express and the American Stock Exchange were both in the World Financial Center[5] (albeit at opposite ends), there wasn't an issue.

That said, clearly identifying shortened forms is extremely important, especially in writing, as you can't correct a reader's assumptions as to what you're going on about.

This applies to "dq" above, as well as many other acronyms and abbreviations. This article[6] discusses the issue at some length:

   Most abbreviations do require a definition. In
   any case where readers might be confused by TLAs 
   three letter abbreviations), it's important that 
   they be defined in the text or in a Table of 
   Abbreviations. A research proposal or article 
   with complicated and confusing abbreviations is 
   one that will quickly face rejection. The problem 
   is that authors become so familiar with 
   abbreviations that they forget to define them.
And since I (or anyone else) can't know what's in your mind unless you tell me, I may take an abbreviation[6] to mean something (like Dairy Queen) that the writer didn't intend.

[0] https://www.dairyqueen.com

[1] https://www.dairyqueen.com/en-us/menu/treats/classic-treats/

[2] https://www.metmuseum.org/

[3] https://www.metopera.org/

[4] https://www.met.police.uk/

[5] https://en.wikipedia.org/wiki/Brookfield_Place_(New_York_Cit...

[6] I'm using "abbreviations" here to mean abbreviations (e.g., Amex for American Express), acronyms (e.g. NAFTA for North Amerian Free Trade Agreement) and initialisms (e.g., TCP for Transmission Control Protocol, FBI for Federal Bureau of Investigation).


[flagged]


>This looks like forum sliding to me. Everybody knows what was meant by DQ, the context made it obvious.

I didn't know what "forum sliding" was either.

Apparently, it's this[0]:

"Technique #1 - 'FORUM SLIDING'

If a very sensitive posting of a critical nature has been posted on a forum - it can be quickly removed from public view by 'forum sliding.' In this technique a number of unrelated posts are quietly prepositioned on the forum and allowed to 'age.' Each of these misdirectional forum postings can then be called upon at will to trigger a 'forum slide.' The second requirement is that several fake accounts exist, which can be called upon, to ensure that this technique is not exposed to the public. To trigger a 'forum slide' and 'flush' the critical post out of public view it is simply a matter of logging into each account both real and fake and then 'replying' to prepositined postings with a simple 1 or 2 line comment. This brings the unrelated postings to the top of the forum list, and the critical posting 'slides' down the front page, and quickly out of public view. Although it is difficult or impossible to censor the posting it is now lost in a sea of unrelated and unuseful postings. By this means it becomes effective to keep the readers of the forum reading unrelated and non-issue items."

===

I'm not really sure how that applies here.

What, exactly, do you suppose I'm trying to distract others from seeing?

Please do elucidate, especially since it ought to increase the visibility of whatever it is you think I'm trying to obfuscate.

Thanks!

[0] https://news.ycombinator.com/item?id=21165439


>especially since it ought to increase the visibility

No, writing paragraphs of irrelevance (you didn't genuinely interpret DQ to mean Dairy Queen so the large comment you wrote about that misinterpretation was just sheer noise) doesn't increase the visibility of what you're trying to obfuscate, it shifts the signal/noise ratio downwards. It pollutes and derails the discussion.


>No, writing paragraphs of irrelevance (you didn't genuinely interpret DQ to mean Dairy Queen so the large comment you wrote about that misinterpretation was just sheer noise) doesn't increase the visibility of what you're trying to obfuscate, it shifts the signal/noise ratio downwards. It pollutes and derails the discussion.

Didn't I? How do you know what's in my mind?

Since Grammarly is a tool to improve writing, talking about improving one's writing (in this case, defining the abbreviations you're using) is certainly appropriate to the discussion.

It's unfortunate that you disagree. But I'm really not sure what you're disagreeing with me about. From what exactly do you think I'm trying to distract/deflect/obfuscate?

I'd really like to know what you think. Really.

Let's talk all about exactly what you want to highlight from this discussion. What might that be? You haven't been clear about anything except that you don't like what I have to say and, seemingly, more the way I said it.

You have claimed that I'm some sort of shill attempting to derail the conversation, but you haven't stated what you think it might be that I'm trying to derail/deflect/obfuscate.

So what's your gripe? What's your take on Grammarly? What do you think is important and should be the focus/foci of this discussion.

Obviously, you're incredibly wise and your opinion is much valued the world over, so please do tell.

And a bonus question for you: Do you have to work at being a jackass, or does it come naturally?


>How do you know what's in my mind?

I'm telepathic.

>What's your take on Grammarly?

I've commented elsewhere in the thread on that.

What's your take on Grammarly? I noticed you didn't contribute anything to the debate but the dairy queen noise.


>I've commented elsewhere in the thread on that.

Not that I've seen. Not going to go back through all those posts to find yours, either. Link to it perhaps?

>What's your take on Grammarly?

I have no "take" on Grammarly, except that as a "cloud-based" (read: someone else's servers) application, I won't ever use it.

Oh, and yes I did ignore your obnoxious blathering this time.

I'll ignore you completely moving forward, as you apparently don't have constructive stuff to say.

Toodles!


>This looks like forum sliding to me. Everybody knows what was meant by DQ, the context made it obvious.

I guess I'm not a member of the group "Everybody" then.

Does that mean I'm stupid?

Or does it mean that folks shouldn't assume that others know everything they do. IIRC, that usually comes around age four or five.


Grammarly seems insidious to me. Not only does it intercept the final versions, but all drafts of what users write. I know they sell a plagiarism product to schools[0]:

> Grammarly's integrated plagiarism checker instantly catches plagiarism from over 16 billion websites and ProQuest's proprietary databases.

So it's pretty clear that collecting and processing lots of semi-private writing is part and parcel to their business, which seems like a recipe for trouble sooner or later. To be clear, I have similar reservations about grammar check in e.g. Google Docs, so this is not limited to just Grammarly at all.

[0]: https://www.grammarly.com/edu


Right, imagine in the context of education, that a student is drafting some essay or whatever. In the process, that student might copy/paste some text from another article and then will go on to rewrite the text into their own language.

By normal definition, this wouldn't be plagiarism, so long as the student extrapolates and restates the original text. And ideally, the student would cite that source, but it probably doesn't happen.

Grammarly might be able to catch this "mistake" - because it would see the copy/pasted text in the first revision and then potentially flag the final outcome.

I'm not saying that plagiarism detection is all that bad of a thing. Teachers need some level of support to help keep their students honest. But there's too much information, in my opinion, being sent when you use any sort of keylogger tool or online editor.


I disagree about your example not counting as plagiarism if the original source is not cited.


Agreed—if that paraphrased or rewritten text isn't cited, it's absolutely plagiarism, by any academic standard.


By the normal definition, that is plagiarism. It is even plagiarism if the person was the original author of the pasted paragraph and it came from another work.


I never knew self plagiarism was a thing until recently. Is it a formality, or is there some kind of reason it's dishonest in the way normal plagiarism is?

Is it because it messes up the ability to figure out what analysis was done when, and what you had access to at the time, and makes it look like older work has the same trustworthiness as newer work that might be done with different equipment or access to sources or new experience?


> By the normal definition, that is plagiarism.

Depends on how much they changed it.

> It is even plagiarism if the person was the original author of the pasted paragraph and it came from another work.

Nah that concept can go screw itself.


what's wrong with the grammar check in google docs?


I believe they're referring to the part where they send all your writing to a HugeCo with a voracious appetite for information about its users and a weaselly, mutable privacy policy.

(Anyone remember when FB was caught sending itself unsubmitted data from web forms? Seems almost quaint now.)


I've seen people promote Language Tool as an alternative: https://languagetool.org It appears to be open source and you can host your own server!


Yes! I looked into languagetool last week to see if I could host it locally on my Synology. They have an interesting page about how, after setting up the initial 200MB install, one can enable n-gram checks[0], which requires an additional 8GB of storage and an SSD. I haven't tackled that yet, but it's on my list!

[0]: https://dev.languagetool.org/finding-errors-using-n-gram-dat...


I run the local server works well as a spelling/grammar checking when using the ngrams. About ten gig in size total. The browser extension is just so-so, I don't love the UI even now but considering the privacy and quality it is hard to be disappointed, there's nothing else out there.

If anyone wants to get set up fast on Windows:

- Server Doc/Download Link: https://dev.languagetool.org/http-server

- ngram Doc/Download Link: https://dev.languagetool.org/finding-errors-using-n-gram-dat...

I've had success with Amazon Corretto jre8 64x for my Java[0][1] and I use this bat file to launch it upon reboot (put it in the same folder as languagetool-server.jar):

     SET PATH=C:\program files\Amazon Corretto\jre8\bin\
     start javaw -cp languagetool-server.jar org.languagetool.server.HTTPServer --port 8081 --allow-origin "\*" -l en-US --languageModel "C:\LanguageTool\LanguageTool-5.4\ngram"
Set path just adds jre8 to bat's path context (there are multiple other ways to accomplish this). "start" just hides the black cmd window while the server is running. --languageModel must be the fully qualified path to the ngram.

You can automate executing this script either using Windows' scheduled tasks or just putting a shortcut to it in the Startup Folder (Win+R enter shell:common startup).

And don't forget to re-configure the browser extension. The setting is in Experimental Settings -> LanguageTool API Server URL -> Local Server

[0] https://corretto.aws/downloads/latest/amazon-corretto-8-x64-... [1] https://docs.aws.amazon.com/corretto/latest/corretto-8-ug/do...


This is really awesome, love that they don't mystify the tech and make it accessible to folks who want to have control of the data.


Unfortunately, Language Tool seems limited to very basic mistakes.

It's Spanish demo only highlights some mechanical punctuation, gender, and plurality errors though it has one useful suggestion for a word that is too specialized.

Useful, for sure, but Grammarly is much more helpful, pointing out things like run-on sentences, clauses that lack parallel verbs, and things like that. Maybe it's better in English, but hopefully Language Tool can get to that level one day.


You can also run it in a Docker container and connect the LanguageTool VS Code extension to it. Works great for Markdown documents!


I use LanguageTool daily since some years and it’s been an absolute pleasure with both English and German. Add to that, the bugs I reported for both language were fixed in mere hours.


How limited is the open source version? Or put another way, what besides convenience would drive me to pay instead of maintain an open source instance?


heiliger strohsack! this is a good suggestion but the site itself seem to be very much like Grammarly or am I mistaken here?

The golden chalice appears to be a self-hosted solution!


>you can host your own server!

If a person needs a server to figure out grammar, it's time to go back to school or crack open an English book.


You do know that there are programmers all around the world whose first language is not English and usually need to communicate in English with English-speaking programmers.

Eg, Log4j vulnerability was found by, Chen Zhaojun of Alibaba Cloud Security Team. His first language is probably Mandarin, while English is 2nd or 3rd Language. He then mailed the Apache team via mail. He would have used such tools for this important mail to properly deliver and explain what exactly he found.


>He would have used such tools for this important mail to properly deliver and explain what exactly he found.

This highlights perfectly the danger of tools which send data to a cloud service and why they should never be used in such a context.

An email coming from a respected source that has 'warning urgent vulnerability' will be given due consideration regardless of the grammar used.


In Grammerly’s Privacy Policy[0], it states as part of information they collect:

“User Content. This consists of all text, documents, or other content or information uploaded, entered, or otherwise transmitted by you in connection with your use of the Services and/or Software.”

and yet they don’t define this as a keylogger. I do understand keyloggers record everything a user types and Grammerly claims to not read “sensitive fields”.

[0] https://www.grammarly.com/privacy-policy


The Japanese government did disallow an IME from Baidu (a software that converts typed keys-strokes to Japanese kana and kanji), because it ran inference on a server.

- https://web.archive.org/web/20140119231002/http://www.techre...


I have disabled any auto-correct, suggest, or any so-called aid in any app that I use. I found most often they are a hindrance not a help. For example suggesting wont instead of won't or its when 99.9% of the time I wanted it's.

My spelling has become terrible and all my life I have been great at spelling. My grammar is OK I thought it was great until I went back to college and felt like I was illiterate.

In college I did try Grammarly mainly for its plagiarism tool. But Grammarly like a virus it's very difficult to uninstall. I caught many mistakes in grammar like "for free" and "off of". And Grammarly plan was supposed to be monthly $20/month then it jumped to $300 US dollars one year-plan automatically charged to my credit card. I didn't notice and after a month they said it was impossible to refund my money. Pure greed, scam, spammy junk.

I also realize I'm tempting Skitt's Law just by mentioning all this.


I feel like their slogan should be lifted from Mr Lee's Greater Hong Kong:

"Whether seriously in business or on a fun-loving hijink, make yourself totally homely in this meager environment. If any aspect is not utterly harmonious, gratefully bring it to my notice and I shall strive to earn your satisfaction."


This article misses the way I use it, which is much safer. I am security minded but also a terrible writer.

I have Grammarly as a browser extension that is OFF BY DEFAULT, except, when I am writing on Medium, and a few times when I click to enable it temporarily.

Problem solved! I feel this article is not serious about a the "What you can do about it". I am fairly confident I have sensitive information controlled, yet I do get the very real benefits when I write a blog post.

I also copy and paste markdown into the standalone web app occasionally because it can correct markdown without getting tripped up by syntax! I am very happy with the quality of grammarly corrections and I do think it is possible to use it safely, just not with its default settings.


I never considered this way of using it, but I can easily see how for a careful user it's a much better experience than using their standalone web app. I'll be sure to add a note about this to the article for future readers.

Thanks for commenting.


I used to use it in highschool. What I'd do is just copy/paste my assessment text into their web interface rather than installing their addon or desktop client.

I'm not sure if they.offer.a web interface outside of their plugins/client apps these days.


The main problem is that Grammarly doesn't want their models/rules/etc. to end up out of their control, hence they do the checking on their server.

But this means it MUST BE a key logger, how else could it work?

But tbh. what irritates me the most, is how bad their product is.

At least with this type of errors I do (some dyslexia, English being the second language, and me having some uh bad past with English in school).

Like the "corrections" they recommend (which go beyond what a "dictionary" spell checker is able to do) are often wrong and will result in another wrong text.

It's pretty obsessed with writing in one specific style.

It seems to have some major problems with listings.

It also seems to want to change anything with some subtle undertone to a version without it.

I would say maybe for people already somewhat good at English which do not make the kind of errors I often do, writing soulless "business" English, it might be good.

If it wouldn't be a major risk to confidentiality.

I do not trust a company like Grammarly (or most companies) to be cable of defending their IT infrastructure against professional attackers, and subtle backdooring Grammarly seems quite useful (for certain actors).

Btw. same for 1Password it's a supper juicy target, especially if it adds a crypto wallet (as they plan to do).

Also I'm pretty sure the usage of Grammarly for writing letters to customers is in conflict with more then just the GDPR (if they contain sensitive information, in more then one way).


It is the biggest botnet ever. How do you think they are paying for ads everywhere. The money have to come from somewhere.


The subscription price is high ($30/month, last time I checked). If that isn't turning a profit, I don't know what will.


Why would anyone pay for that?


Spellchecking by humans is more expensive and not instant. Grammarly is something between the classic software spell checkers and professional human spellcheckers.

Producing well-written documents is essential if you want to appear competent and professional. Using tools is helpful, especially when you are not a native speaker.


Turn employees from noticeably not able to pull off the job to sort of pulling it off.


They have annual subscriptions that get sharply discounted on a regular basis, where it comes out closer to $70 a year. I pay for that, since I usually do need the editing help, but the linked article certainly gives me pause.


Enterprise. That’s a cost that just doesn’t matter.


Does most of the HN react the same way I did when I first saw an ad for grammarly?

I remember just thinking that it would definitely be logging my text and recording my internal endpoints.

Is that a normal thought or would most of HN not expect this ?


Generally, Grammarly feels like a practical example of something humans want to believe AI will be very good at, but will eventually turn out not to be.


I am very proud I have always refused to use this kind of software, the question is "Is Grammarly a keylogger" is hard to answer per se, but it has always potential to collect enormous data about its users, it doesn't matter what they claim in their T&C, Facebook also started as a "just" social network, they ended up being one the biggest data collector of the world.


It's not a hard answer, Grammarly IS a keylogger, vast amounts of sensitive data is being actively sent to them. Whether they abuse it, or get hacked so a third party can abuse it, is the question. Currently not that we know of. But the software itself facilitates it and that they provide value doesn't change this.


What can you do about it?

Learn to write. Don't use Grammerly. That's the article. Instead, we keep trying to find little tricks to keep the utility without surrendering privacy.

High school and college essays are already full of enough mindless fluff and tropes. Why put everything you write into something that then makes you sound like a bot? Your essays will all end up with YouTube Face.


How about non-native English speakers that are using it as an way to take tips and improve their English writing?


I'd argue that's even worse for non native speakers. It's alluring, really I see it. This would be massive amounts of utility if I were trying to learn French.

But the cost is it's not actually natural, it's what the AI says is natural.


There are plenty of ways to improve your English without grammarly.


Yes but if people are gravitating towards Grammerly it's because it's offering something those other ways aren't.

I mean I guess you can learn an instrument by reading sheet music and sheer force of will but I think most people prefer lessons.


> Yes but if people are gravitating towards Grammerly it's because it's offering something those other ways aren't.

I disagree. Grammarly has bombarded the internet and YouTube with so many ads, there are ads that make fun of Grammarly ads.

Nevertheless, people don't use it because it offers something that others don't. They use it because they don't know any better.


I don't use it, but people I know who use it claim it helped them a lot. So, clearly, they found a lot of value in what it is doing.

Even in this discussion, most suggestions for "better" is completely useless advice of "learn to write".


Yeah. Grammarly greatly improved my English.


> Learn to write. Don't use Grammerly.

*Grammarly.


Oh dear! I walked right into that one didn't I? :'D


Well, it's Grammarly, not Vocabarly or Spellarly


I know how to write, but trimming down content the way grammerly does(Which is presumably what many want to see, since it's everywhere) , takes time.

I've never been to school, and I probably wouldn't use it for anything like an essay in real life, but it seems good when brevity is the main critical factor and sounding like a bot is expected.

I rarely ever see writing outside of actual books, and pre-2008ish news articles that specifically impresses me though, so maybe it's not the best for anything that matters, bot style writing does get pretty painful to read for hours.


Sounding like a bot is improvement for most high school essays. And the fact is, without someone or something correcting you, your learning to write will be much slower.


There ain't no reason I wouldn't never consider using Grammarly once, at least especially not now or whatnot.


Could you use Grammarly keylogs to train AIs? It could have infinitely more content than the web has.



I’m the only one that saw Loom having access to the webcam in the article?

Loom can open your webcam and record you audio if they want…

So grammarly is offering this service thanks to the cloud and thanks to the data transmitted over internet. We cannot do nothing else that trust this companies. We should maybe ask to our government to check and validate how our data are treated by companies and if the privacy is respected.


Second question: Is TextExpander a key logger as well?


Does it send the data anywhere?


There is a cloud component, but in this case it’s syncing text expansions you saved to it making it more like 1Password than Grammarly.


I guess I try to follow least permissions principle everywhere; no matter what it gets and where it sends - but good point.


Not a point, honest query.

My personal rule would be if it sends data anywhere yes, if it logs the key press data locally yes too (also why) otherwise probably not but never say never


The article suggests how could it all of the sudden read the Note after being turned on without new key presses — uhh duh the accessibility APIs it discussed?

Seems all very overblown to me. It’s known to begin with that it sends things you type to the server - that’s the model (or rather, how they run their models). That does not automatically mean evil things are happening.


I have long suspected Grammarly as a massive undercover FSB operation to monitor the west. The amount of marketing push behind the product never made sense otherwise, and their corporate HQ is in Ukraine.

Even if you don't buy the conspiracy theory, the cold hard truth now is that those servers will be under Russian jurisdiction within a month.


I have thought that one of the most effective spyware tools would be a really good open source library that most people will use without really knowing what it does and it sends your secrets to some bad government somewhere or at least gives them a master kill switch if they want to DDoS everyone.


The more people there are using an open source library the more likely it is that someone is going to take the time to see exactly what it does and any unexpected network traffic (sending data to some bad government) will sound the alarm much faster. The most effective spyware tools are things like cell phones, facebook, Windows, and Google. They encourage people to spill their own secrets and make it difficult if not impossible to use them while protecting your privacy in any meaningful way.


Their corporate HQ is in the US, and they are hosted on AWS.

They have a development office in Ukraine, too, but they have evacuated. Even if Russia takes over they won't get access to your writing.


It absolutely is an intelligence carve out.


Is anyone gonna bring up GitHub Autopilot?


I use Grammarly and Github Copilot to write technical documents in VsCode (with vim mode). It's actually great. But the documents are very far from being confidential.


It also injects html/junk into email editors.

Our staff use it. Having typos in political emails is bad and super duper stressful. On the whole I think worth it for our use case.

But when you look at the actual draft email html there is almost always some grammarly fragment left like pseudo html elements and stuff


Yes, grammarly is an excellent data exfiltrator, don't ever use it while typing anything you don't want published externally.

The easiest way to do that (instead of being constantly aware) is to just never use it (or anything like it that sends all your private typing to somewhere else).


For German I want the Duden spell checker tool back.

German has more strict rules for a lot of it's grammar, which can be checked rather well without needing any AI.

It ran local.

Or maybe I'm just nostalgic.

But as far as I remember it was the best spell checker I ever had (for any Language).

But then I'm hardly writing German now-days.


Related blog post I made:

http://trevcan.duckdns.org/blog/rant-writing-tools.html

It talks more about morality and not legality.


I tried that once just to see what it's all about, and it kept regularly sending irksome e-mails ("We're not seeing any activity from last week") for, I think, well over a year.


Not saying good or bad, but they do have a page on their "trust":

https://www.grammarly.com/trust


I'm sad to read this. As someone who is dyslexic, it's a necessary evil I have to use. Unless there is alternatives?

Infact I use Hemingway App and then Grammarly to write content.


Hmm it seems curious that this attack on a successful Ukrainian startup is happening at this time.

Could it be a Russian smear campaign? It seems like the sort of thing that the St Petersburg disinformation teams would attempt, in very subtle ways...


I doubt this has anything to do with Russia. Sometimes, all it takes is a little common sense and curiosity to even ask about these things.

Or maybe the guy just read the TOS. Some people actually do that!


It looks like Russians want to a) spread doubt, b) force users to look for alternatives, so they will find and install their trojans, to try them.


It would have been good if he asked how long grammar checked text is retained to grammarly / an engineer that works on it.


You wouldn't get an accurate answer to that if there were any sort of shenanigans going on. It's a pointless question.

Also, considering how unlikely it is that any particular sentence a user has typed is unique, would it count as 'storing that text in a database' if the text you typed remained in their systems in a tokenized fashion, i.e. incremented a sentence use counter?


All AI based systems that use your data to improve itself is incentivised to collect everything. Especially if it's free


Incidentally, using Grammarly is verboten, internally at a more famous company whose name also happens to start with a G.


As I was wondering where they are located: https://www.google.com/search?client=firefox-b-d&q=where+is+...

    San Francisco, California
    
    The software is produced by Grammarly Inc, which is headquartered in San Francisco, California, with offices in Kyiv, New York City, and Vancouver.


let me get this straight.. it is listening to every key input i make on my mac regardless of the app i am in? then sends this data to its cloud? im assuming this includes passwords?

if so, how is this app still legally available to the US consumer?


I installed its plugin on Microsoft Word and I feel a bit safe now.


How long until OpenAI comes with an open replacement?


Why is different than Gmail auto complete?


Try out outwrite.com

If you need a grammar correction.


@dang, This is Russian attack on Ukrainian Unicorn. Can you shadowban awakened bots?


...What? Is this incomprehensible to anyone else?


Yes? Not use it?


Don't use it and all concerns resolved.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: