In my observation, people who use LLMs to assist in their writing 1) consistently underestimate how apparent it is to readers and 2) consistently underestimate how off-putting it is to readers.
I think outsourcing this kind of work to LLMs is a bad idea. You are robbing yourself of the opportunity to build skills. My recommendation: take longer to write. Accept an end product that feels imperfect or unpolished. And publish, publish, publish.
Iāll let you be the judge of how off-putting it is in this case. Here is the prompt I wrote:
Find any obvious ways of making the following much clearer and address any grammatical or orthographical errors:
I can well imagine that you are generally right, although Iād add some caveats:
1. āUsing LLM to assist oneās writingā covers a very broad set of applications. I can prompt an LLM to do little more than check for punctuation, or to simply criticize what I write rather than rewrite it for me. Thereās a wide range of ways in which I could use an LLM that would be neither apparent nor off-putting to the vast majority of people. By the way, I wouldnāt claim that my post above is an example of a subtle use of AI. I agree it is rather obvious there.
2. Even if your claim is generally true, I think it is also fair to say that people commonly overestimate their ability to spot AI-generated content. Because they cover such a wide range of interventions, from very subtle to incredibly obvious, I doubt that the people who are most critical of the tech are quite as good at spotting it as they suspect.
3. Trade-offs are real. The time I donāt spend honing my organic writing skills is time I may spend honing some other skill, or just enjoying life in any of a myriad of other ways. :).
Here is the response that the LLM gave me:
I can imagine that you are broadly right, though I would add a few caveats.
āUsing an LLM to assist oneās writingā covers a wide range of practices. I might ask an LLM to check punctuation or to critique what I wrote instead of rewriting it. Many forms of assistance are subtle enough that most people would neither notice nor be bothered by them. To be clear, my post above is not an example of subtle use. I agree it is quite obvious there.
Even if your claim holds in general, people often overestimate their ability to detect AI generated text. Since the level of intervention can vary from minimal to very heavy, I doubt that the harshest critics are as good at spotting it as they think.
Trade offs matter. Time I do not spend polishing my unaided writing is time I can put into other skills or simply enjoying life in other ways.
Well, you biased me by telling me in advance which one is which, but I much prefer the purely human-written version.
I think editing or rewriting existing text will harm the quality, but the thing that harms the quality the most by far is when the LLM generates original text. It has certain tics or habits when it comes to writing that I really donāt like.
I think you are probably operating under the misapprehension that LLMs write better than you or that LLMs improve your writing. Your purely human-written example is perfectly competent and thereās no reason you should run it through an LLM filter.
Ha, true, this would have been more fun if I hadnāt told you š. Thank you for your kind words on the competence of my writing.
About my supposed missaprahention: It depends. Sometimes the LLM takes all of the edge and spice out of what I say, and then Iāll gladly ditch its suggestions. However, often itāll be more concise and a bit more graceful in its choice of words. Itās most definitely better at grammar and orthography than I am. In the example above, the whole message is shorter without losing any meaningful content. I think removing the āwellā from the first sentence made it a bit less clumsy. In the second sentence the word āpracticesā is slightly closer to what I wanted to express than āapplicationsā. Etc. Be that as it may, it gives me more options, which is nice. I am still free to reject its suggestions or modify them, so Iām happy itās there for me to be used if I find it beneficial, as I often do.
I used to like Grammarly for checking spelling, grammar, punctuation, and copy editing things, but it seems like itās gone downhill since switching to an LLM-based software. Google Docs is decent for catching basic things like typos, accidentally missing a word, accidentally repeating a word, subject/āverb agreement, etc.
I actually donāt agree with the LLMās changes in the two examples you mentioned and I think it made the writing worse in both cases. The LLMās diction is staid and corporate, it lacks energy.
In my observation, people who use LLMs to assist in their writing 1) consistently underestimate how apparent it is to readers and 2) consistently underestimate how off-putting it is to readers.
I think outsourcing this kind of work to LLMs is a bad idea. You are robbing yourself of the opportunity to build skills. My recommendation: take longer to write. Accept an end product that feels imperfect or unpolished. And publish, publish, publish.
Iāll let you be the judge of how off-putting it is in this case. Here is the prompt I wrote:
Find any obvious ways of making the following much clearer and address any grammatical or orthographical errors:
I can well imagine that you are generally right, although Iād add some caveats:
1. āUsing LLM to assist oneās writingā covers a very broad set of applications. I can prompt an LLM to do little more than check for punctuation, or to simply criticize what I write rather than rewrite it for me. Thereās a wide range of ways in which I could use an LLM that would be neither apparent nor off-putting to the vast majority of people. By the way, I wouldnāt claim that my post above is an example of a subtle use of AI. I agree it is rather obvious there.
2. Even if your claim is generally true, I think it is also fair to say that people commonly overestimate their ability to spot AI-generated content. Because they cover such a wide range of interventions, from very subtle to incredibly obvious, I doubt that the people who are most critical of the tech are quite as good at spotting it as they suspect.
3. Trade-offs are real. The time I donāt spend honing my organic writing skills is time I may spend honing some other skill, or just enjoying life in any of a myriad of other ways. :).
Here is the response that the LLM gave me:
I can imagine that you are broadly right, though I would add a few caveats.
āUsing an LLM to assist oneās writingā covers a wide range of practices. I might ask an LLM to check punctuation or to critique what I wrote instead of rewriting it. Many forms of assistance are subtle enough that most people would neither notice nor be bothered by them. To be clear, my post above is not an example of subtle use. I agree it is quite obvious there.
Even if your claim holds in general, people often overestimate their ability to detect AI generated text. Since the level of intervention can vary from minimal to very heavy, I doubt that the harshest critics are as good at spotting it as they think.
Trade offs matter. Time I do not spend polishing my unaided writing is time I can put into other skills or simply enjoying life in other ways.
Well, you biased me by telling me in advance which one is which, but I much prefer the purely human-written version.
I think editing or rewriting existing text will harm the quality, but the thing that harms the quality the most by far is when the LLM generates original text. It has certain tics or habits when it comes to writing that I really donāt like.
I think you are probably operating under the misapprehension that LLMs write better than you or that LLMs improve your writing. Your purely human-written example is perfectly competent and thereās no reason you should run it through an LLM filter.
Ha, true, this would have been more fun if I hadnāt told you š. Thank you for your kind words on the competence of my writing.
About my supposed missaprahention: It depends. Sometimes the LLM takes all of the edge and spice out of what I say, and then Iāll gladly ditch its suggestions. However, often itāll be more concise and a bit more graceful in its choice of words. Itās most definitely better at grammar and orthography than I am. In the example above, the whole message is shorter without losing any meaningful content. I think removing the āwellā from the first sentence made it a bit less clumsy. In the second sentence the word āpracticesā is slightly closer to what I wanted to express than āapplicationsā. Etc. Be that as it may, it gives me more options, which is nice. I am still free to reject its suggestions or modify them, so Iām happy itās there for me to be used if I find it beneficial, as I often do.
I used to like Grammarly for checking spelling, grammar, punctuation, and copy editing things, but it seems like itās gone downhill since switching to an LLM-based software. Google Docs is decent for catching basic things like typos, accidentally missing a word, accidentally repeating a word, subject/āverb agreement, etc.
I actually donāt agree with the LLMās changes in the two examples you mentioned and I think it made the writing worse in both cases. The LLMās diction is staid and corporate, it lacks energy.