This is revolting at so many levels.
Big difference between "AI, rewrite this passage to sound more like Hunter S Thompson" and "Grammarly-brand unauthorized digital agent Hunter S Thompson, provide a critique of my writing"
Let's see what company values informed this decision [0].
> At Grammarly, it all starts with our EAGER values: Ethical, Adaptable, Gritty, Empathetic, and Remarkable. These values are guiding lights that keep the Grammarly experience compassionate and our business competitive.
In other words an LLM can spit out a plausible "output of X", however it cannot encode the process that lead X to transform their inputs into their output.
One lesson they might draw from the negative press is to offer a more open-ended interface, like ChatGPT, where for years people have already been asking "Pretend you are X and review my writing". This interface design pattern gives the press nowhere to point their angry fingers
Does it add any value for writers?
We believed this was coming and that the best way to handle it was give the real person control over their persona to grow/edit/change and train it as they see fit.
I actually own the patent on building an expert persona based on the context of the prompt plus the real persons learned information manifold...
If it feels like Grammarly does not respect your right to digital sovereignty, it is because it does not.
For me, Grammarly gives me the same impression as Datadog, but I have no explanation for why I feel that way.
i can ask it to tell me how to write like a person X right now.
Isn't that what grammarly has always been, since long before the invention of the transformer? They give you a long list of suggestions, and unless you write a corporate press release half of them are best ignored. The skill is in choosing which half to ignore
Generative AI is a plague at this point. Everybody is adding to their wares to see what happens. It's almost like ricing a car. All noise, no go.
https://www.sciencedirect.com/science/article/pii/S0749596X2...
If you think “the tacit knowledge and conscious/subconscious reasoning mix that caused X to write like X” can be meaningfully captured by some 1-page “style guide” like llmtropes, I’m not sure what to tell you. Such a style description would be informed by a soup of reviewers that most certainly cannot write like X even with their stronger and more nuanced observations than what the LLM picked up.
It really feels so wrong to spare nobody, not even dead writer/people.
All it's gonna do is something similar to em-dashes where people who use it are now getting called LLM when it was their writing which would've trained LLM (the irony)
If this takes off, hypothetically, we will associate slop with the writing qualities similar to how Ghibli art is so good but it felt so sloppy afterwards and made us less appreciate the Ghibli artstyle seeing just about anyone make it.
The sad part is that most/some of these dead writers/artists were never appreciated by the people of their time and they struggled with so many feelings and writing/art was their way of expressing that. Van Gogh is an example which comes to my mind.[0] Many struggled from depression and other feelings too. To take that and expression of it and turn it into yet another product feels quite depressing for a company to do
[0]: https://en.wikipedia.org/wiki/Health_of_Vincent_van_Gogh
Words paint the picture, but the meaning of the picture is what matters.
Unrelated but surprising to me that I've found built-in grammar checking within JetBrains IDEs far more useful at catching grammar mistakes while not forcing me to rewrite entire sentences.
Seems pretty likely usage of Grammarly's core product has cratered in the past few years. Not totally hard to imagine one of the big AI labs paying their legal fees in exchange for putting this out there and kick starting the legal process on some of these issues.
If you built an LLM exclusively on the writings and letters of John Steinbeck, you could NOT tell the LLM to solve an integral for you amd expect it to be right.
Instead what you will receive is a text that follows a statistically derived most likely (in accordance to the perplexity tuning) response to such a question.
Sounds like something I'd expect to see on a banner in an elementary school classroom.
That train left at full steam when companies scraped the whole internet and claimed it was fair use. Now it's a slippery slope covered with slime.
I believe there'll be no slowing down from now on.
They are doing something amazing, will they ask for permission? /s.
[1] https://plugins.jetbrains.com/plugin/12175-natural-languages... [2] https://languagetool.org -- warning: is coated in somewhat-misleading AI keywords [3] https://github.com/languagetool-org/languagetool
The person in that room, looking up a dictionary with Chinese phrases and patterns, certainly follows a process, but it's easy to dismiss the notion that the person understands Chinese. But the question is if you zoom out, is the room itself intelligent because it is following a process, even if it's just a bunch of pattern recognition?
"The work is public, hence the name. It's well known, it's in the data. Who cares".
What will they do next? Create similar publications with domainsquatting and write all-AI articles with the "public" names?
Is it still fair use, then?
So IMO they are just flinging things at the wall trying to find a way back.
this shows that you have very less idea on how llm's work.
LLM that is trained only on john steinbeck will not work at all. it simply does not have the generalised reasoning ability. it necessarily needs inputs from every source possible including programming and maths.
You have completely ignored that LLMs have _generalised_ reasoning ability that it derives from disparate sources.
Most importantly, negative but unused signals might not be available if the text does not mention it.
Isn't this obvious? There is not enough latent knowledge of math there to enable current LLMs to approximate anything resembling an integral.
can you give a specific example of what an llm can't do? be specific so we can test it.
I probably did not. Then I would have written that. They are fucking over the dead. They are clearly not communicating with the dead.
https://www.chiffandfipple.com/t/kenny-g-as-necrophile-long-...
You don't bring the dead virtually back to life to perform tricks for you.
This isn't 2023 anymore
i give the LLM my codebase and it indeed learns about it and can answer questions.
It reminds me of winzip.
This is not the same thing as reasoning.
LLMs are pattern matchers. If you trained an llm only to map some input to the output of John Steinbeck, then by golly that's what it'll be able to do. If you give it some input that isn't suitably like any of the input you gave it during training, then you'll get some unpredictable nonsense as output.
In school we would have a test with various questions to show you understand the concept of addition, for example. But while my calculator can perfectly add any numbers up to its memory limit, it has no understanding of addition.
Its obvious to you.
It isnt obvious to the person I am responding to, and it isnt obvious to majority of individuals I speak with on the matter (which is why AI, personally, is in the bucket of religion amd politics for polite conversation to simply avoid)
Do you have fond memories of being a teacher’s pet? Wish you could still get notes from your favorite college professor? Dream about some implacable voice of authority correcting your every word choice and punctuation mark? Well, great news: A certain software company has engineered a way to simulate criticism not just from bestselling authors and famous academics of our time, but also many who died decades ago—and the company evidently didn’t need permission from anybody to do it.
Once relied upon only to proofread for correct grammar and spelling, the writing tool Grammarly has added a host of generative AI features over the past several years. In October, CEO Shishir Mehrotra announced that the overall company was rebranding as Superhuman to reflect a new suite of AI-powered products. However, the AI writing “partner” remains called Grammarly. “When technology works everywhere, it starts to feel ordinary,” Mehrotra wrote in his press release. “And that usually means something extraordinary is happening under the hood.”
The expanded Grammarly platform now offers an AI solution for every imaginable need—and some you’ve probably never had. There’s an AI chatbot that will answer specific questions as you compose a draft, a “paraphraser” feature that suggests changes in style, a “humanizer” that revises according to a selected voice, an AI grader that predicts how your document would score as college coursework, and even tools for flagging and tweaking phrases commonly produced by large language models. (Sure, you’re using AI to do everything here, but you don’t want it to sound like that.)
Perhaps most insidiously, however, Grammarly now has an “expert review” option that, instead of producing what looks like a generic critique from a nameless LLM, lists a number of real academics and authors available to weigh in on your text. To be clear: Those people have nothing to do with this process. As a disclaimer clarifies: “References to experts in this product are for informational purposes only and do not indicate any affiliation with Grammarly or endorsement by those individuals or entities.”
As advertised on a support page, Grammarly users can solicit tips from virtual versions of living writers and scholars such as Stephen King and Neil deGrasse Tyson (neither of whom responded to a request for comment) as well as the deceased, like the editor William Zinsser and astronomer Carl Sagan. Presumably, these different AI agents are trained on the oeuvres of the people they are meant to imitate, though the legality of this content-harvesting remains murky at best, and the subject of many, many copyright lawsuits.
“Our Expert Review agent examines the writing a user is working on, whether it's a marketing brief or a student project on biodiversity, and leverages our underlying LLM to surface expert content that can help the document's author shape their work,” says Jen Dakin, senior communications manager at Superhuman. “The suggested experts depend on the substance of the writing being evaluated. The Expert Review agent doesn’t claim endorsement or direct participation from those experts; it provides suggestions inspired by works of experts and points users toward influential voices whose scholarship they can then explore more deeply.”
Someone like King may see the advance of AI as unstoppable, and there may be nobody left to defend Zinsser’s 1976 handbook On Writing Well from the big tech vultures, but what of the countless other luminaries who still want to keep their material from being compressed into an algorithm? Vanessa Heggie, an associate professor of the history of science and medicine at the University of Birmingham, recently took to LinkedIn to share an especially grim example of how the feature works, accusing Superhuman of “creating little LLMs” based on the “scraped work” of the living and dead alike, trading on “their names and reputations.” The screenshot she posted showed the availability of analysis from an AI agent modeled on David Abulafia, an English historian of the medieval and Renaissance periods who died in January. “Obscene,” Heggie wrote.
An independent review of the Expert Review tool by WIRED reproduced the recommendations for feedback from the Abulafia bot, as well as from models based on the living cognitive scientists Steven Pinker and Gary Marcus. (Neither returned a request for comment.) As the software processed the sample text, it noted that it was taking “inspiration” from Elements of Style author William Strunk Jr. and the sociologist Pierre Bourdieu while applying “ideas” from Gone With the Wind author Margaret Mitchell and using “concepts” from writer and professor Virginia Tufte—all of whom are dead, with Tufte dying most recently, in March 2020. The guidance from her AI agent read: “Replace repetition with vivid, varied sentence patterns.”
C.E. Aubin, a historian and postdoctoral fellow at Yale University who shared Heggie’s LinkedIn post on Bluesky, tells WIRED that Grammarly’s “expert” system “seems to validate the profound mistrust so many scholars in the humanities have for AI and its seemingly constant use in fundamentally unethical ways.”
“These are not expert reviews, because there are no ‘experts’ involved in producing them,” Aubin says. “And it's pretty insulting to see scholarship used this way when the academic humanities are currently under attack from every possible angle—as though the actual people who do the thinking and produce the scholarship are reducible to their work itself and can be removed entirely from the equation.” She says this elimination of personhood is “awful” enough on its own, apart from “the issue of ‘reanimating’ the dead so cynically.”
Beyond the dubious ethics, there’s the question of whether these proliferating AI widgets are even effective or helpful. Grammarly’s plagiarism detector, for instance, didn’t catch a direct quotation I used from a scene in The Simpsons where Bart improvises a geography presentation he hasn’t prepared for, leading to an empty summation: “In conclusion, Libya is a land of contrasts.” (Grammarly did warn, however, that “a land of contrasts” is a sequence of words often generated by LLMs.)
Over the past several years, teachers and professors have struggled through a deluge of AI-written essays, finding it difficult to wean their pupils off of this self-defeating shortcut. And even before Grammarly had its “experts,” those who relied on it to proofread their papers were occasionally accused of cheating after the material was flagged by AI detection services. Giving these users the impression that they can have their work evaluated by leading thinkers before they turn it in may contribute to their sense that they are only double-checking their text, not violating any academic code of conduct.
But at least students can enjoy having their homework assessed by illusory mentors instead of their actual instructors, which may or may not be a slippery slope toward eliminating school faculty altogether. Shouldn’t take long to find out!
Not sure why you need a concrete example to "test", but just think about the fact that the LLM has no idea how a writer brainstorms, re-iterates on their work, or even comes up with the ideas in the first place.
> Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes.
I suggest „randomly adjusting parameters while trying to make things better“ as that accurately reflects the „precision“ that goes into stuffing LLMs with more data.
This Grammarly thing seems to be a bastardized form of that not even sparing the dead.
I'd say that there was some incentive by the AI companies to muddle up the water here.
It's very enlightening, if you ask me.
Unless you are actually fine tuning models, in which case sure, learning is taking place.
This isn't true in general, and not even true in many specific cases, because a great deal of writers have described the process of writing in detail and all of that is in their training data. Claude and chatgpt very much know how novels are written, and you can go into claude code and tell it you want to write a novel and it'll walk you through quite a lot of it -- worldbuilding, characters, plotting, timelines, etc.
It's very true that LLMs are not good at "ideas" to begin with, though.
LLMs can reason about integrals as well as in a literature context. You suggested that if it’s not trained on literature then it can’t reason about it. But why does that matter?
"my calculator can perfectly add any numbers up to its memory limit" This kind of anthropomorphic language is misleading in these conversations. Your calculator isn't an agent so it should not be expected to be capable of any cognition.
When the “how many ‘r’ in ‘strawberry’” question was all the rage, you could definitely get LLMs to explain the steps of counting, too. It was still wrong.
> If you trained an llm only to map some input to the output of John Steinbeck
this is literally not possible because the llm does not get generalised reasoning ability. this is not a useful hypothetical because such an llm will simply not work. why do you think you have never seen a domain specific model ever?
if you wanted to falsify this claim: "llm's cant reason" how would one do that? can you come up with some examples that shows that it can't reason? what if we come up with a new board game with some rules and see if it can beat a human at it. just feed the rules of the game to it and nothing else.
here is gpt-5.4 solving never before seen mathematics problems: https://epochai.substack.com/p/gpt-54-set-a-new-record-on-fr...
you could again say its just pattern matching but then i would argue that its the same thing we are doing.
ex: i read a lot of shakespeare, understand patterns, understand where he came from, his biography and i will be able to write like him. why is it different for an LLM?
i again don't get what the point is?
>> Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes.
Which part was snarky, excessively hostile or unprofessional?
if i showed a human a codebase and asked them questions with good answers - yes i would say the human learned it. the analogy breaks at a point because of limited context but learning is a good enough word.
It's certainly possible to mimic many aspects of a notable writer's published style. ("Bad Hemingway" contests have been a jokey delight for decades.) But on the sliding scale of ingenious-to-obnoxious uses for AI, this Grammarly/Superhuman idea feels uniquely misguided.
Imagine a interviewing a particularly diligent new grad. They've memorized every textbook and best practices book they can find. Will that alone make them a senior+ developer, or do they need a few years learning all the ways reality is more complicated than the curriculum?
LLMs aren't even to that level yet.
And that's often inaccurate - just as much as asking startup founders how they came to be.
Part of it is forgot, part of it is don't know how to describe it and part of it is don't want to tell you so.
To really recreate his writing style, you would need the notes he started with for himself, the drafts that never even made it to his editor, the drafts that did make to the editor, all the edits made, and the final product, all properly sequenced and encoded as data.
In theory, one could munge this data and train an LLM and it would probably get significantly better at writing terse prose where there are actually coherent, deep things going on in the underlying story (more generally, this is complicated by the fact that many authors intentionally destroy notes so their work can stand on its own--and this gives them another reason to do so). But until that's done, you're going to get LLMs replicating style without the deep cohesion that makes such writing rewarding to read.
Seems like there could be others that are better.
I get that you're into AI products and ok, fine. But no you have not "studied [Shakespeare] greatly" nor are you "able to write like [Shakespeare]." That's the one historical entity that you should not have chosen for this conversation.
This bot is likely just regurgitating bits from the non-fiction writing of authors like an animatronic robot in the Hall of Presidents. Literally nobody would know if the LLM was doing even a passable job of Truman Capote-ing its way through their half-written attempt at NaNoWriMo
The point is that you dont become Jimi Hendrix or Eric Clapton even if you spend 20 years playing on a cover band. You can play the style, sound like but you wont create their next album.
Not being Jimi Hendrix or Eric Clapton is the context you are missing. LLMs are Cover Bands...
As another example, I can write a story about hobbits and elves in a LotR world with a style that approximates Tolkien. But it won't be colored by my first-hand WW1 experiences, and won't be written with the intention of creating a world that gives my conlangs cultural context, or the intention of making a bedtime story for my kids. I will never be able to write what Tolkien would have written because I'm not Tolkien, and do not see the world as Tolkien saw it. I don't even like designing languages
The LLM does not model text at this meta-level. It can only use those texts as examples, it cannot apply what is written there to it's generation process.
They absolutely do not. If you "ask it how it came up with the process in natural language" with some input, it will produce an output that follows, because of the statistics encoded in the model. That output may or may not be helpful, but it is likely to be stylistically plausible. An LLM does not think or understand; it is merely a statistical model (that's what the M stands for!)
I can do it at the moment with Shakespeare an LLMs.
But authors have not done this work alone. Grammarly is not going to sell "get advice from the editorial team at Vintage" or "Grammarly requires your wife to type the thing out first, though"
I'll also note that no human would probably want advice from the living versions of the author themselves.
https://github.com/theJayTea/WritingTools/blob/main/Windows_...
why do you think that's the case? lets start from here.
the real answer is that you get benefits from having data from many sources that add up expontentially for intelligence.
> LLMs are pattern matchers
but lets try to falsify this. can you come up with a prompt that clearly shows that LLM's can't reason?
can you provide a _single_ example where LLM might fail? lets test this now.
I do have a number of examples to give you, but I no longer share those online so they aren’t caught and gamed. Now I share them strictly in person.
You see, the comment that I replied to made an assumption, that assumption is embedded in the word 'probably'. The person that wrote that presumes to know what I intend. I corrected that. Clarified it and moved on. If that seems hostile and snarky to you then I'm happy to be educated. For myself, I think the comment I replied to could have been phrased as a question rather than a statement.
that's why we have really good fake van gogh's for which a person can't tell the difference.
of course you can't do the same as the original person but you get close enough many times and as humans we do this frequently.
in the context of this post i think it is for sure possible to mimic a dead author and give steps to achieve writing that would sound like them using an LLM - just like a human.
i can prove that it does have understanding because it behaves exactly like a human with understanding does. if i ask it to solve an integral and ask it questions about it - it replies exactly as if it has understood.
give me a specific example so that we can stress test this argument.
for example: what if we come up with a new board game with a completely new set of rules and see if it can reason about it and beat humans (or come close)?
You need to show me an LLM applying writing techniques do not have examples in its corpus.
You would have to use some relatively unknown author, I can suggest Iida Turpeinen. There will be interviews of her describing her writing technique, but no examples that aren't from Elolliset (Beasts of the sea).
Find an interview where Turpeinen describes her method for writing Beasts of the Sea, e.g.: https://suffolkcommunitylibraries.co.uk/meet-the-author-iida...
Now ask it to produce a short story about a topic unrelated to Beasts of the Sea, let's say a book about the moonlanding.
A human doing this exercise will produce a text with the same feel as Beasts of the Sea, but an LLM-produced text will have nothing in common with it.
LLMs can't consistently win at chess https://www.nicowesterdale.com/blog/why-llms-cant-play-chess
Now, some of the best chess engines in the world are Neural Networks, but general purpose LLMs are consistently bad at chess.
As far as "LLM's don't have understanding", that is axiomatically true by the nature of how they're implemented. A bunch of matrix multiplies resulting in a high-dimensional array of tokens does not think; this has been written about extensively. They are really good for generating language that looks plausible; some of that plausable-looking language happens to be true.
why are you bringing this constraint?
https://maxim-saplin.github.io/llm_chess/
ets not cherry pick and actually see benchmarks please. i would say even ~1000 elo means that it can reason better than the average human.
If someone has already done the work of giving an example of how to produce text according to a process, we have no way of knowing if the LLM has followed the process or copied the existing example.
And my point of course is that copying examples is the only way that LLMs can produce text. If you use an author who has been so analyzed to death that there are hundreds of examples of how to write like them, say, Hemingway, then that would not prove anything, because the LLM will just copy some existing "exercise in writing like Hemingway".
Editing is one of these things. There can be lots of different processes, informed by lots of different things, and getting similar output is no guarantee of a similar process.
this is not true, any examples?
If we are talking about human artifacts, you never have reproducibility. The same person will behave differently from one moment to the next, one environment to another. But I assume you will call that natural variation. Can you say that models can't approximate the artifacts within that natural variation?
If I trained (or, more likely, fine-tuned) an LLM to generate code like what's found in an individual's GitHub repositories, could you comfortably say it writes code the same way as that individual? Sure, it will capture style and conventions, but what about our limitations? What do you think happens if you fine-tune a model to write code like a frontend developer and ask it to write a simple operating system kernel? It's realistically not in their (individual) data but the response still depends on the individual's thought process.
Look, I don't think you understand how LLM's work. Its not about fine tuning. Its about generalised reasoning. The key word is "generalised" which can only happen if it has been trained on literally everything.
> It's relevant for data it hasn't been trained on
LLM's absolutely can reason on and conceptualise on things it has not been trained on, because of the generalised reasoning ability.
Of course, but reasonable behavior across all humans is not the same as what one specific human would do. An individual, depending on the scenario, might stick to a specific choice because of their personality etc. which is not always explained, and heavily summarized if it is.