Just this morning I used Claude to build an "omit needless words" app, which lets me paste in text and then marks up that text with strikes through the words that I can leave out.
I used it to edit a document I'm working on, where I accepted about a third of its suggestions. I feel pretty good about it.
If you have a Claude account you should be able to access the artifact I built here: https://claude.ai/artifact/RK7XfnaoxVV1Uji84ETaLr - it uses "Claudeception" - the feature of Claude artifacts that lets them use Claude prompts - so it will be billed against your Claude account (at a cost of probably a fraction of a cent per run.)
- You can use AI to look through docs and do research.
- You can use AI to structure your argument.
- You can NOT use AI to actually write the prose.
- You can use AI to proofread, especially for how effectively you've communicated.
I don't know, man, that sounds a lot like using AI to write. Your writing probably says "load-bearing" less than Claude's outputs, but I'm not sure it's much less AI.
Reading through docs and doing research: of course. They're like much better search engines in that regard. It would be silly not to use them. Obviously, you need to review their search results against their sources.
Structuring arguments: bad idea. Here you're having the model do your thinking for you. The problems that occur with having it write text recur at a macro level; too-obvious patterns. Readers just want to know what the prompt was.
Writing the prose: obviously not.
Proofreading: I think this works pretty great, as long as you have a structure to how you're doing it. "How do I make this better or sharper": bad, you're having the model structure your arguments for you. But for a predefined rubric of stylistic things --- writing you know how to do, that's just tedious to check manually? Hard to see a downside.
> Writing requires thinking, don't let an AI do the thinking for you.
Okay, but reading also requires substantial thinking. Moreover, we can say the same about, say, coding. And I am sure there are people who will say, yes, don't use AI to code. But I do, and the results are amazing, and I still think very deeply: in fact, I spend more time with AI coding (and thinking) than before as I do not get hamstrung on, how do I convert this Word doc to PDF in code again? I spend more time in higher order thinking. Same with when I read text written by an AI. I don't believe that writing is any more sacrosanct that coding. And reading and adjusting AI output requires substantial thinking.
Once upon a time the word processor was invented, and it made it so much easier to arrange your words on paper. iIf bits were wrong, you could just delete them, or copy and paste them across pages. It was actually pretty amazing what you could do; especially if you didn't mind dot-matrix type on fan-fold paper. Of course if you wanted to make a better impression, you'd type it on a typewriter, or better yet, hand-calligraph your writing with a fountain pen.
So LLMs allow you to work at the speed of reading, if not faster. So you can try a paragraph here, completely rewrite the tone. Start a chapter in 1PP and just have it rewritten to 3PP. And sofort and so on. The number of new transformations you can do is endless.
What do a lot of people do? Write a one line prompt and show everyone the oneshot result.
Ok fine, sometimes you don't need to write amazing prose. But you can also be like the early writers who adopted the word processor. If you put the same amount of elbow-grease into using an LLM as you did writing before, you can get some pretty cool results.
>Once upon a time the word processor was invented, and it made it so much easier to arrange your words on paper.
It’s easier to rearrange and rethink and tinker. Hard to imagine now but it was more efficient in the days when you would have to pencil out a memo and have someone type it up, you would make corrections and the final copy would be done in a couple passes. This has been studied, efficiency has actually gone down, in favor of immediacy and direct control.
What it frees you from is organizing your thoughts ahead of time, which takes practice and discipline. Using an LLM is just taking that further, you can avoid organizing your thoughts and ideas a bit more.
If someone uses LLM to help wording their idea in a way that's easier to understand, I am fine with it as long as they don't copy-paste it verbatim into their blogpost.
I am getting a little ranty here, but the usual LLM-isms like "not this, but that" or "that's the real footnote here" simply add zero value to the text, and I can't imagine anyone who has proofread the LLM-generated text even once will opt to keep those sloppy paragraphs. If I see them, I can't help but to think that the author had oneshotted the text. And it strongly suggests to me that the author has put zero effort into their work.
I disagree, the problem is the tool. Generating text instead of thinking about what text to use explicitly is not writing, and it's not worthy of someone else reading it.
Fundamental to exchanging ideas is discovering which ideas should be exchanged and how to express them. The process of writing itself is that exercise. If you use a tool to outsource the thinking then you're not thinking.
I definitely think that "not thinking about what you're writing" is a big danger with AI-generated text, but it's not a given. If you generate a block of text and then go through and edit it to better reflect your ideas and voice - that's the thinking process in action. It's also entirely possible to generate slop by manually typing every word - see any corporate communication ever.
I'm on version 6 of a writing an agent voice, style, tone, and orientation guide I have been developing over the course of the past year. Every time the agent writes something using the guide, I edit it and then update the guide where needed. Slowly, the agent is beginning to sound eerily close to my natural writing voice. No, I didn't use it for this comment :)
AI for proofreading/spelling correction and reordering works quite well. My words, but perhaps structured in a way which makes more sense, with sentiment replaced by synonyms.
Basically an automatic "second-pass" which I would've caught had I re-read it, but automatically.
I use it to write in German, because my German sucks.
I first write the thing myself, then feed it through Deepl Write to see how the AI would write it. It's been very helpful.
I write everything else myself. However I don't know how I feel about spending my time on Earth typing personal responses to AI-generated messages. Do those people deserve my time and energy?
I said this once in my recent "Trustworthy AI" class that... "I think writing with AI is one of the most detrimental things you can do".
The reaction was pretty much silence and most people probably thought I was some angry, arrogant, anti-ai purist... which isn't the case. It can be acceptable to use a language model in code generation but requires new forms of discipline and tooling.
But for for just regular old writing... Should never, ever be generated. I'll likely never change that stance.
Also, don't use AI to translate into a language you can't read back. It glosses over assumptions you didn't know you made, and the output might not mean what you intended.
Just this morning I used Claude to build an "omit needless words" app, which lets me paste in text and then marks up that text with strikes through the words that I can leave out.
I used it to edit a document I'm working on, where I accepted about a third of its suggestions. I feel pretty good about it.
If you have a Claude account you should be able to access the artifact I built here: https://claude.ai/artifact/RK7XfnaoxVV1Uji84ETaLr - it uses "Claudeception" - the feature of Claude artifacts that lets them use Claude prompts - so it will be billed against your Claude account (at a cost of probably a fraction of a cent per run.)
So let me get this straight:
- You can use AI to look through docs and do research.
- You can use AI to structure your argument.
- You can NOT use AI to actually write the prose.
- You can use AI to proofread, especially for how effectively you've communicated.
I don't know, man, that sounds a lot like using AI to write. Your writing probably says "load-bearing" less than Claude's outputs, but I'm not sure it's much less AI.
Reading through docs and doing research: of course. They're like much better search engines in that regard. It would be silly not to use them. Obviously, you need to review their search results against their sources.
Structuring arguments: bad idea. Here you're having the model do your thinking for you. The problems that occur with having it write text recur at a macro level; too-obvious patterns. Readers just want to know what the prompt was.
Writing the prose: obviously not.
Proofreading: I think this works pretty great, as long as you have a structure to how you're doing it. "How do I make this better or sharper": bad, you're having the model structure your arguments for you. But for a predefined rubric of stylistic things --- writing you know how to do, that's just tedious to check manually? Hard to see a downside.
I didn't see in the article where he recommended AI to structure an argument? He did suggest it can be helpful for structuring data.
> Writing requires thinking, don't let an AI do the thinking for you.
Okay, but reading also requires substantial thinking. Moreover, we can say the same about, say, coding. And I am sure there are people who will say, yes, don't use AI to code. But I do, and the results are amazing, and I still think very deeply: in fact, I spend more time with AI coding (and thinking) than before as I do not get hamstrung on, how do I convert this Word doc to PDF in code again? I spend more time in higher order thinking. Same with when I read text written by an AI. I don't believe that writing is any more sacrosanct that coding. And reading and adjusting AI output requires substantial thinking.
> Okay, but reading also requires substantial thinking.
I don't think anyone would argue reading a book takes as much thinking as writing one…
We have copious evidence for behaviors like note taking and writing creates deeper thinking and retention that passively watching / reading
https://www.scientificamerican.com/article/why-writing-by-ha...
Once upon a time the word processor was invented, and it made it so much easier to arrange your words on paper. iIf bits were wrong, you could just delete them, or copy and paste them across pages. It was actually pretty amazing what you could do; especially if you didn't mind dot-matrix type on fan-fold paper. Of course if you wanted to make a better impression, you'd type it on a typewriter, or better yet, hand-calligraph your writing with a fountain pen.
So LLMs allow you to work at the speed of reading, if not faster. So you can try a paragraph here, completely rewrite the tone. Start a chapter in 1PP and just have it rewritten to 3PP. And sofort and so on. The number of new transformations you can do is endless.
What do a lot of people do? Write a one line prompt and show everyone the oneshot result.
Ok fine, sometimes you don't need to write amazing prose. But you can also be like the early writers who adopted the word processor. If you put the same amount of elbow-grease into using an LLM as you did writing before, you can get some pretty cool results.
>Once upon a time the word processor was invented, and it made it so much easier to arrange your words on paper.
It’s easier to rearrange and rethink and tinker. Hard to imagine now but it was more efficient in the days when you would have to pencil out a memo and have someone type it up, you would make corrections and the final copy would be done in a couple passes. This has been studied, efficiency has actually gone down, in favor of immediacy and direct control.
What it frees you from is organizing your thoughts ahead of time, which takes practice and discipline. Using an LLM is just taking that further, you can avoid organizing your thoughts and ideas a bit more.
You nailed it on what I've always wanted to say.
If someone uses LLM to help wording their idea in a way that's easier to understand, I am fine with it as long as they don't copy-paste it verbatim into their blogpost.
I am getting a little ranty here, but the usual LLM-isms like "not this, but that" or "that's the real footnote here" simply add zero value to the text, and I can't imagine anyone who has proofread the LLM-generated text even once will opt to keep those sloppy paragraphs. If I see them, I can't help but to think that the author had oneshotted the text. And it strongly suggests to me that the author has put zero effort into their work.
Agree completely. The problem is not the tool, it's all the pointless filler (or even hallucinations) that the tool generates that people leave in.
I disagree, the problem is the tool. Generating text instead of thinking about what text to use explicitly is not writing, and it's not worthy of someone else reading it.
Fundamental to exchanging ideas is discovering which ideas should be exchanged and how to express them. The process of writing itself is that exercise. If you use a tool to outsource the thinking then you're not thinking.
I definitely think that "not thinking about what you're writing" is a big danger with AI-generated text, but it's not a given. If you generate a block of text and then go through and edit it to better reflect your ideas and voice - that's the thinking process in action. It's also entirely possible to generate slop by manually typing every word - see any corporate communication ever.
> Of course if you wanted to make a better impression, you'd type it on a typewriter, or better yet, hand-calligraph your writing with a fountain pen.
Why do people feel the need to invent the past that never existed?
I lived in that past myself, what are you thinking about?
I'm on version 6 of a writing an agent voice, style, tone, and orientation guide I have been developing over the course of the past year. Every time the agent writes something using the guide, I edit it and then update the guide where needed. Slowly, the agent is beginning to sound eerily close to my natural writing voice. No, I didn't use it for this comment :)
My brain immediately went to "I wonder how many people got this article and used AI to summarize it for them."
I agree. Writing manually and thoughtfully and then letting the AI figure out the nitty gritty implementation works very well.
AI for proofreading/spelling correction and reordering works quite well. My words, but perhaps structured in a way which makes more sense, with sentiment replaced by synonyms.
Basically an automatic "second-pass" which I would've caught had I re-read it, but automatically.
Proofreading yes, automatic "corrections" no
The article opens with
> Writing requires thinking, don't let an AI do the thinking for you.
And then the author goes on to admit that they use AI for coding. Does coding not require thinking or what?
I agree that we shouldn't be delegating thinking to AI, but stopping at writing seems disingenuous.
Similar recent discussion: https://news.ycombinator.com/item?id=49767937
I use it to write in German, because my German sucks.
I first write the thing myself, then feed it through Deepl Write to see how the AI would write it. It's been very helpful.
I write everything else myself. However I don't know how I feel about spending my time on Earth typing personal responses to AI-generated messages. Do those people deserve my time and energy?
I said this once in my recent "Trustworthy AI" class that... "I think writing with AI is one of the most detrimental things you can do".
The reaction was pretty much silence and most people probably thought I was some angry, arrogant, anti-ai purist... which isn't the case. It can be acceptable to use a language model in code generation but requires new forms of discipline and tooling.
But for for just regular old writing... Should never, ever be generated. I'll likely never change that stance.
Also, don't use AI to translate into a language you can't read back. It glosses over assumptions you didn't know you made, and the output might not mean what you intended.
this SPECIALLY for arabic the shit I see everyday
- why stop at writing? put this advice for coding too
- sure its good to ask AI feedback about your module or code review or identify vulnerabilities or better ways of doing your existing code
- but maybe stop asking AI to generate entire projects from scratch?