Honored to be the subject of this amazing analysis, Nick! Superb explanations (you praise me too much 🙏🏻🙏🏻). I've tried and failed to put into words what you just described so well here. Important topic and great advice for young writers who want to know how to tap into their uniqueness as humans in the age of AI.
Thanks for the Sardinha synopsis and interesting insights into AI writing. As an ELF business communications teacher and applied linguistics researcher interested in L2 writing and genAI, I resonated with a number of your observations and you further clarified the research territory, too. I liked your analysis of Romero's writing, though that is more of a narrative genre not so applicable to most business and academic communications. I have recently analyzed default writings from GPT, Claude, and Gemini in more of a research writing context and specifically looking at readability from lexical and syntax complexity perspectives since so many L2 English speaking researchers are using ChatGPT to write up their research. but it was good that you helped me take off my blinders and consider other aspects of writing which I will focus on later. I have noticed and therefore have been recently interested in "AI-generated ... exhibit significant differences, with .... a higher density of information-carrying features" -- which makes it difficult to read. ChatGPT was the worst for this. :P
Nice post, Nick. I especially appreciate the emphasis on the way that what feels like imperfection and struggle is where writing grapples with complexities. Writing is making meaning together. These tools can help that process, but they can also interfere. Much of how that actually works in practice is counterintuitive.
This was a great read for many reasons, including that I have been very interested in recent reports about Claude 3 and its writing capabilities. I am not qualified to comment any further because I'm not really a writer or someone involved in academics :(
Thanks, Patrick. Claude is a best AI-writing tool out there right now. I have found it a highly flexible and responsive tool. My main critique is the use of Claude as a single-prompt writing generator--which is a critique that is expansive and includes all AI tools indirectly. Thanks for continuing to engage with my work. I really appreciate it.
Understood on the critique applying across the board - that was well-stated within the post. That's also been my default gut feeling about these tools from Day 1 - incredibly useful in augmenting our efforts, not doing an entire thing for us. Said differently, I have the same mindset when using them for research purposes.
Love this analysis of the weaknesses of generic, synthetic generated writing. Really interesting that it‘s the quirks, nuances and narrative elements married to the experiences one makes that make human writing human.
Yes, the high density of information carrier features hits the nail on the head for me. I am interested in prompting techniques that force LLMs to open up that density and unpack connections. So far, the best I can get are multiplication of paragraphs, all equally dense. Any suggestions?
This was really insightful, Nick. And reassuring too. Like Alberto, I also fear a future where writing is commoditized to the point no one cares anymore, how much of what makes us humans would we lose?
And I agree with your article's premise: I am yet to experience AI-written text that resonates in the same as anything written by other people. Precisely because AI text feels so predictable and formulaic, but also because it's hard (if not impossible) to substitue well-crafted yet ultimately empty write-ups by LLMs for personal experience, emotions, and quirks that will always make their way into human writing, if ever so imperceptibly.
Honored to be the subject of this amazing analysis, Nick! Superb explanations (you praise me too much 🙏🏻🙏🏻). I've tried and failed to put into words what you just described so well here. Important topic and great advice for young writers who want to know how to tap into their uniqueness as humans in the age of AI.
Thanks for the Sardinha synopsis and interesting insights into AI writing. As an ELF business communications teacher and applied linguistics researcher interested in L2 writing and genAI, I resonated with a number of your observations and you further clarified the research territory, too. I liked your analysis of Romero's writing, though that is more of a narrative genre not so applicable to most business and academic communications. I have recently analyzed default writings from GPT, Claude, and Gemini in more of a research writing context and specifically looking at readability from lexical and syntax complexity perspectives since so many L2 English speaking researchers are using ChatGPT to write up their research. but it was good that you helped me take off my blinders and consider other aspects of writing which I will focus on later. I have noticed and therefore have been recently interested in "AI-generated ... exhibit significant differences, with .... a higher density of information-carrying features" -- which makes it difficult to read. ChatGPT was the worst for this. :P
I enjoy your posts. Keep up the great work.
Nice post, Nick. I especially appreciate the emphasis on the way that what feels like imperfection and struggle is where writing grapples with complexities. Writing is making meaning together. These tools can help that process, but they can also interfere. Much of how that actually works in practice is counterintuitive.
This was a great read for many reasons, including that I have been very interested in recent reports about Claude 3 and its writing capabilities. I am not qualified to comment any further because I'm not really a writer or someone involved in academics :(
Thanks, Patrick. Claude is a best AI-writing tool out there right now. I have found it a highly flexible and responsive tool. My main critique is the use of Claude as a single-prompt writing generator--which is a critique that is expansive and includes all AI tools indirectly. Thanks for continuing to engage with my work. I really appreciate it.
Understood on the critique applying across the board - that was well-stated within the post. That's also been my default gut feeling about these tools from Day 1 - incredibly useful in augmenting our efforts, not doing an entire thing for us. Said differently, I have the same mindset when using them for research purposes.
Love this analysis of the weaknesses of generic, synthetic generated writing. Really interesting that it‘s the quirks, nuances and narrative elements married to the experiences one makes that make human writing human.
Yes, the high density of information carrier features hits the nail on the head for me. I am interested in prompting techniques that force LLMs to open up that density and unpack connections. So far, the best I can get are multiplication of paragraphs, all equally dense. Any suggestions?
This was really insightful, Nick. And reassuring too. Like Alberto, I also fear a future where writing is commoditized to the point no one cares anymore, how much of what makes us humans would we lose?
Great breakdown of Alberto's recent post.
And I agree with your article's premise: I am yet to experience AI-written text that resonates in the same as anything written by other people. Precisely because AI text feels so predictable and formulaic, but also because it's hard (if not impossible) to substitue well-crafted yet ultimately empty write-ups by LLMs for personal experience, emotions, and quirks that will always make their way into human writing, if ever so imperceptibly.