A question is if you use these AI used tools will the AI detection for assessments and for submitted articles be triggered by just the grammar corrections, even if there was no generation of content?
I just checked out Lex with an off-hand observation about a NYT house profile.
I was absolutely horrified by the suggestions for improvement. It turned a pungent, nay, piquant 
brief observation about house design into utter pablum with ridiculous suggestions. And the grammatical suggestions were wrong. 
Here is the original:
Not bad! Not bad! Very much a Portland climate house. It’s done thoughtfully for older persons. I want that bathroom! But, you know? Even though I would enjoy staying there for a visit, there’s something not-right about the spaces: they are not balanced. Look at the smaller Frank Lloyd Wright houses and how the spaces gently sing. And $900/sq.ft. and you’re using plywood? What!? Where’d the money go? Core-Ten sheathing isn’t that expensive! And “Before moving, the owners purged 35 artworks from their collection. They hired a specialist to hang the remainder.” Yes, they were exhausted but that’s nice stuff they’ve got there and you’re going to pay some art-dweeb to hang them? WTF?
And here is the summary response:
The main idea here seems to be a critical analysis of a specific house design, focusing on issues like the unbalanced spaces, the use of plywood despite the high cost, and the owners’ decision to hire someone to hang their art collection.
While the critique itself is somewhat unique, the overall approach feels a bit generic. Readers have likely seen this type of critical analysis of architectural design before. To make this more compelling, I would encourage you to dig deeper and find a more unique angle or perspective that will really grab the reader’s attention.
For example, instead of just pointing out the flaws, you could explore the underlying reasons behind the design choices - was it a budget constraint, a stylistic preference, or something else? You could also tie it into broader trends or issues in modern architecture and design. The key is to offer readers something new and thought-provoking, rather than just a standard critique.
In terms of the flow, the sections do connect logically, moving from an initial positive impression to more critical analysis. However, the transitions could be smoother. Try to weave the different points together more seamlessly, rather than jumping abruptly between topics.
Additionally, I would suggest tightening up the language in places. Some of the phrasing feels a bit rambling or conversational, when a more concise and polished style would be more effective. For example:
The house has a thoughtful design well-suited for older residents, including an appealing bathroom.
Overall, this has the makings of an interesting piece, but it needs some refinement to really make it stand out. Focus on developing a more unique and compelling perspective, and work on tightening up the writing. I’m happy to provide more specific suggestions as you revise.
Utter hogwash (but you see where all the comments come-from. How many flaccid undergraduate essays and critiques has this “AI” ingested?).
Speaking not of Lex in particular but in general, spelling correction is great (particularly since my old fingers are not so accurate), grammar correction is deficient, and stylistic advice from these things is utterly ghastly.
There.
Dave
But presumably not all your email is (you mention using it for Mimestream). And once its installed, can’t Grammarly monitor everything you type even if you disable its interface in a particular app or website? I agree that writing things which will be published doesn’t seem to be a privacy issue, but for everything else I’d be uncomfortable with a company monitoring all that I write.
Will you consider writing an article about Lex? I’d not heard of it, and clicking through the link I’m still not sure what sets it apart. It claims to provide ‘A premium writing experience.’ but I’d be curious what this means in practice and why you have found it useful. Do you store things in Lex as well as write them (some sort of Google Docs replacement?) or is it more of a first stage and the text gets deposited in your preferred document app/location/format?
1 Like
No idea. My understanding is that those tools basically don’t work anyway.
I’m not quite sure what you were trying to get it to do, but using Lex requires that you prime it appropriately with who you are, what you’re trying to achieve, who your target audience is, and so on. When I use Lex, I’m sometimes asking it specific questions about certain bits of selected text (fact checking, phrasing alternatives that write around things I’m unhappy with in my original, thoughts about how something might be taken by specific audiences, and so on.) Other times, I ask it to evaluate what I’ve written, again with fairly detailed prompt, so I can see if it suggests any areas I might have elided or missed. I often disagree, but that’s fine—it’s not changing anything.
Fair point! As I understand it, Grammarly would have to process everything that is written while it’s active in an app for the purposes of providing the service. However, you can turn off training on your data, and Grammarly automatically doesn’t train on data from people in the EU, UK, and some other European countries.
And no, it doesn’t record data if it’s turned off, or in sensitive places like password fields.
Yes, I plan to. At the moment, I’d describe it as a word processor with a built-in assistant that can do things for you based on what you’ve written and how you prompt it.
I use it purely to draft TidBITS articles, and I don’t go back to documents once they’ve been published in WordPress. Lex really is just a word processor for me. I use Google Docs like that too, but I also have Google Docs that are permanent stores of information and grow over time.
2 Likes
I just asked it for an opinion on the piece. The response you see is what it gave me with no further precision prompting.
From what you say, given your, generally ( ), discursive writing Lex may well be an excellent writing aid. But what horrified me in its general response was the near-corporate speak of its observations. It should have shut-up after saying, perhaps, “Interesting! Do you have a specific problem?”
Dave
1 Like
I have not used Grammarly or Apple’s Writing Tools as I do not write a lot, and when I do, I am fairly confident my proofreading will pick up any issues.
However, I have noticed that Siri sometimes takes liberties with the words I dictate to it. So if I say “going to do something”, I get “gonna do something”. This is gross and not appropriate for those items that I do write.
Have you noticed either tool to substitute “modern usage” forms for what you actually typed?
1 Like
You’ve put your finger on my general concern.
These tools are defined by corporate marketing experts, constructed by corporate programmers, trained on materials intended to persuade corporate executives to support projects or grant budgets, and refined through in-house alpha testing.
(Gross oversimplification, yes, but dammit, this is an argument I’m making! 
)
At this moment in time, the “many of us” who have learned how to read and write in more traditional modes know how to adopt or reject a machine-generated analysis. This is already changing as adolescents and young adults are exposed to a pedagogical environment that embraces AI as a teaching tool.
In another thread we’re discussing Microsoft Copilot, though the concern there is more about marketing and stealth subscription tactics. My main objection to the Copilot experience is that a Copilot icon sits in the lefthand margin of each paragraph I write, beckoning me to tap it and magically see a better paragraph appear for my approval.
Don’t need it. Definitely don’t want it. And wondering what Lex or Copilot might do with these three sentences?
That’s interesting and something I never thought about. I write in multiple genres: technology in general, lasers and optics, science journalism, tutorials on technology, and science fiction. I shift between styles because I have different audiences. I use different jargon writing for lasers/optics and writing for tech in general. I wouldn’t want a grammar checker to homogenize my writing because I tailor white I write for different audiences.
1 Like
Ugh. Got my answer:
It is not necessary. It is not wanted. Considering the potential uses of Lex or Copilot with these three sentences.
Apple Intelligence is perhaps more anodyne:
It is unnecessary and undesirable. I am curious to know Lex or Copilot’s intentions regarding these three sentences.
Don’t need it. Definitely don’t want it.
1 Like
No, not at all. I dislike Siri’s informal contractions too, even if they accurately reflect what I said. If anything, Grammarly and Lex err too much toward the formal side, so their text can be a bit stilted.
At most, these kinds of results are the default case. LLMs are extremely good at writing in any style you can describe. The more I investigate this world, the more I see that what you get out is highly dependent not so much on what you put in, but on how you prompt.
It wouldn’t—to be clear, Grammarly lets you prevent it from training on what you write. Even if you allowed training, my understanding is that it wouldn’t be learning from what you write for the purposes of modifying what it suggests for your text. However, when you reject a suggestion from Grammarly, that apparently is a tiny vote against that in the future, although I haven’t noticed any difference with the words that I wish it wouldn’t mark in my text.
Lex is different. Its “context tags” would let you specify different collections of information for different fields, so its suggestions would be filtered through the appropriate context. But you’d have to set that up.
I have to say that so far, it is fairly easy to pick out AI-written text. There is something about it that does not have the character of human-written text. It is just bland I suppose, whereas humans tend to include unusual words or concepts or even sentence structures to make a point.
I think you give human-written text way more credit than the vast majority of it deserves. 
2 Likes
Or perhaps humans make mistakes in grammar or spelling which AI rarely does?
Nothing is perfect. But humans and software make different kinds of mistakes.
1 Like
Thanks for the review and your experience, I appreciate the discussion of AI tools in the comments. I learned about Grammarly when I went back to college after retiring, and it really helped me catch typos, fix contractions, and choose better words and phrasing in my papers. I’m working on a translation of a Japanese book and will be using it to proofread my draft when it’s ready.
Additionally, similar to @kirkmc I didn’t like having the Grammarly app resident on my computer. It defaulted to a wide-open model (check everything), rather than a controlled model (check only what I tell it to). Rather than deselecting a specific app, I’d rather be able to choose which apps I want Grammarly to check. To each his own.