I’ve come to the same conclusion about AI and my creativity and understanding. I see terrible examples of colleagues shamelessly saying “I put your figure into AI and it said this”. I was pretty insulted and embarrassed for them.
Great take on AI in writing, particularly academic works. I recently downloaded data from NIH RePORT on K99 and F32 postdoctoral fellowships to look at success rates by institute since 2018. I'll have to dive back in to run an experiment on AI text detection across the years of data I already have!
Thank you, Stephen, for another great post! People often ask me about my use of AI, and after I lay out the pros and cons (there are!), I often also point them to this #Substack. It is a little foggy out there, so your insights are always welcome.
Disclaimer: Grammarly often checks my grammar, and I still make mistakes, and those are all mine.
This human enjoyed this essay very much, and is grateful for your survey of the slopverse. The dilemma for editors and consumers of the scientific literature used to be bad enough with the daily torrent of legit papers and discoveries---the drinking at the fire hydrant metaphor was commonly used. Let's hope some clever individuals (like you) can come up with solutions for the new floods.
With all the low effort writing out now, I follow roughly this process for reading:
1. Check the title. If the subject matter potentially interests me, I'll skim.
2. Is it it obvious AI slop? If so, move on. If the subject matter interests me enough, I'll ask AI about it myself. If not, actually read the article.
3. Pass the article to an AI (usually 2) for a fact check and analysis. Pass the AI what I took away and see if my conclusions are supported.
It varies, but now whatever anyone writes needs to pass an interview before I'll read it. An obvious solution is Pangram, at least for things like arXiv.
I find that if I form my own opinion first, double checking with AI helps my understanding. If I start by having an AI summarize, the result is bad and more often I find myself just skipping the process.
Thanks for your honesty and analysis! I've ramped up my use of Claude over the past month or so, and it keeps growing as I see what it can do. I’ve found that it has freed my mind of so much that was crowding my mental space. It's actually helped my creativity and writing! I've played around with different methods and I'm curious to hear yours. One thing that has worked well is telling it to not rewrite my words, but to be a real-time editor. I like using the Claude Docs for this because I can write and chat at the same time, then reorganize things myself or have Claude do it when it's a formatting thing.
Thank you for articulating this. I know that feeling of overwhelm, and to be honest it's tricky trying to read selectively. Sticking to know authors/institutions is the obvious shortcut but as you point out, this risks making the world smaller. I'm going to have to think of a new way to prioritise what I give my attention to!
I’ve come to the same conclusion about AI and my creativity and understanding. I see terrible examples of colleagues shamelessly saying “I put your figure into AI and it said this”. I was pretty insulted and embarrassed for them.
Great take on AI in writing, particularly academic works. I recently downloaded data from NIH RePORT on K99 and F32 postdoctoral fellowships to look at success rates by institute since 2018. I'll have to dive back in to run an experiment on AI text detection across the years of data I already have!
Thank you, Stephen, for another great post! People often ask me about my use of AI, and after I lay out the pros and cons (there are!), I often also point them to this #Substack. It is a little foggy out there, so your insights are always welcome.
Disclaimer: Grammarly often checks my grammar, and I still make mistakes, and those are all mine.
This human enjoyed this essay very much, and is grateful for your survey of the slopverse. The dilemma for editors and consumers of the scientific literature used to be bad enough with the daily torrent of legit papers and discoveries---the drinking at the fire hydrant metaphor was commonly used. Let's hope some clever individuals (like you) can come up with solutions for the new floods.
With all the low effort writing out now, I follow roughly this process for reading:
1. Check the title. If the subject matter potentially interests me, I'll skim.
2. Is it it obvious AI slop? If so, move on. If the subject matter interests me enough, I'll ask AI about it myself. If not, actually read the article.
3. Pass the article to an AI (usually 2) for a fact check and analysis. Pass the AI what I took away and see if my conclusions are supported.
It varies, but now whatever anyone writes needs to pass an interview before I'll read it. An obvious solution is Pangram, at least for things like arXiv.
I find that if I form my own opinion first, double checking with AI helps my understanding. If I start by having an AI summarize, the result is bad and more often I find myself just skipping the process.
Thanks for your honesty and analysis! I've ramped up my use of Claude over the past month or so, and it keeps growing as I see what it can do. I’ve found that it has freed my mind of so much that was crowding my mental space. It's actually helped my creativity and writing! I've played around with different methods and I'm curious to hear yours. One thing that has worked well is telling it to not rewrite my words, but to be a real-time editor. I like using the Claude Docs for this because I can write and chat at the same time, then reorganize things myself or have Claude do it when it's a formatting thing.
Thank you for articulating this. I know that feeling of overwhelm, and to be honest it's tricky trying to read selectively. Sticking to know authors/institutions is the obvious shortcut but as you point out, this risks making the world smaller. I'm going to have to think of a new way to prioritise what I give my attention to!