Grok and Claude are too complementary when reviewing my work. And seems to work faster than it should. Both found the errors in the stories that were there, but I couldn't quite explain them. Organizing complex ideas, checking for logical errors, and unfulfilled promises appear to be the most effective uses.
I have to say, I did find a genuinely touching moment in Pratchett's Going Postal. The substituted message that starts with "Who will speak for the dead?" Almost brings me to tears.
So far, the most helpful everyday thing I’ve found from AI is to take the draft and redo it for various IQ levels. Just seeing the differences helps think about style in a different way, and sometimes it offers new insights - what’s missing in the simplified version is often as productive as what’s included in the higher-level examples.
If you're fully aware of the work to which it is comparing your work, that's not a problem. I already have an idea of what works well and what doesn't, so it's basically like another set of different eyes on it.
Some of the advice, I concur with and implement. Some of it, I think is simply wrong and don't. No different than a human editor in that regard.
"What you say if you were an editor who had to look at 100 other stories in competition for 1 publication"
"What would you say if you a redditor who hated it. What would you say if you were a redditor who loved it?
"What would you say if you were a hardcore fan who actually liked the story, but disagreed with the redditor who hated it? What demographic does this hardcore fan fit into?"
Every chat session I start with Claude is "Read the entire KB, brief tech speak only, dont flatter me" and then enter.
Every time it responds differently, but they all look more like 1950s papers than 2025 teen cell phones. Occassionally, they will look more lively, and I just close those sessions, because it skipped some of the docs.
Grok and Claude are too complementary when reviewing my work. And seems to work faster than it should. Both found the errors in the stories that were there, but I couldn't quite explain them. Organizing complex ideas, checking for logical errors, and unfulfilled promises appear to be the most effective uses.
Vox, have you tried the various GPT-5 family in this role? They seem to have dialed the sycophant setting quite a bit down from version 4 to 5.
Not in this role, but I've tried it. I was not impressed with it.
I have to say, I did find a genuinely touching moment in Pratchett's Going Postal. The substituted message that starts with "Who will speak for the dead?" Almost brings me to tears.
Be a little bit careful about asking AI to review your work. Challenge it to make sure that it really did. Case in point: https://amandaguinzburg.substack.com/p/diabolus-ex-machina
So far, the most helpful everyday thing I’ve found from AI is to take the draft and redo it for various IQ levels. Just seeing the differences helps think about style in a different way, and sometimes it offers new insights - what’s missing in the simplified version is often as productive as what’s included in the higher-level examples.
If you're fully aware of the work to which it is comparing your work, that's not a problem. I already have an idea of what works well and what doesn't, so it's basically like another set of different eyes on it.
Some of the advice, I concur with and implement. Some of it, I think is simply wrong and don't. No different than a human editor in that regard.
AI is very good at unexpected angles of analysis.
The same model/same prompt can be very consistent. A good measure of yourself.
No it’s not. It’s All preprogrammed like you.
The angles are half the fun:
"What you say if you were an editor who had to look at 100 other stories in competition for 1 publication"
"What would you say if you a redditor who hated it. What would you say if you were a redditor who loved it?
"What would you say if you were a hardcore fan who actually liked the story, but disagreed with the redditor who hated it? What demographic does this hardcore fan fit into?"
Every chat session I start with Claude is "Read the entire KB, brief tech speak only, dont flatter me" and then enter.
Every time it responds differently, but they all look more like 1950s papers than 2025 teen cell phones. Occassionally, they will look more lively, and I just close those sessions, because it skipped some of the docs.