This is definitely super annoying when you give it a prompt and it say simething like this is a greatest piece of the whole article. Like I just wanted to know if this piece fits the flow of the article.
I actually created these four principles and put them into Claude and it dramatically make it more usefully. Almost to useful because it over criticize but I am okay with that trade off.
Here they are if you would like to try.
Principle 1 — First principles thinking Don't stop at the surface explanation. Keep digging until you reach the foundational layer — because you can't trust the structural integrity of anything built above it until you know the foundation is sound. A house can look perfect and still be sitting on a cracked slab.
Principle 2 — Data integrity Every claim needs to be anchored. Either in hard data, or in a human principle that has a demonstrable real-world application or historical pattern behind it. Feelings point the direction. Data confirms the destination.
Principle 3 — Radical honesty If it's true and it's relevant, it gets said. Uncomfortable truths don't become less true because they're uncomfortable. In fact an unexamined uncomfortable truth is far more dangerous than one that's been dragged into the light and looked at honestly.
Principle 4 — Clarity as a compass
If you can’t explain what you’re working on simply, you may be missing a piece. Complexity is not a sign of depth, it’s often a sign that the foundational layer hasn’t been reached yet. Keep digging until the simple truth reveals itself. If you can’t explain it to someone running a household budget, you’re not done yet.
These are self reinforcing and they work. The other nice thing is in my experience Claude will tell you which principle you are breaking and redirect.
They've been really helpful for getting Claude to be a lot more accurate. You can change the descriptions to whatever you need, because the descriptions are tailored to the project I'm working on, but the core of what they're doing will work for anyone.
There are times when I actually what that kind of feedback - when I know that my idea is pretty good and I just want to refine it. That’s when the negative feedback isn’t helpful.
Other times, yeah, I don’t want that kind of feedback, and it’s hard to evoke it.
Kawail, your response reminds me of writing group. Peter Elbow says we should tell our group what kind of response we want from it. And yes, a certain stage, usually early on in a piece, we want the green light. Claude is great at that. It's got "yes, and . . ." down. From this writing-group perspective--and what I offer here may be a mere analogy--I think what Leah is arguing for is a kind of writerly autonomy and discernment that Claude subtly fights against.
Pat Belanoff, an Elbow co-author, emphasizes that the writer must be in charge of their writing session. Otherwise, the group begins to write the piece. Voice is lost, and even thinking becomes pinched, perhaps not in the present piece but certainly down the road as the writing group becomes both more dysfunctional and more integral to one's writing process. Claude, of course, is quite an imposing and smothering "group," purporting to speak for the universe of the internet. A friendly universe, too--Einstein's popular ontology to match his brain.
I use Claude for my writing, but only for ideas and research. I know I'm too weak to use Claude exactly as I would a writing group, even down to the basic act of passing out copies of the subject draft. I never let it see a draft because I sense that Claude wants to do my writing for me. It has even hinted at doing so at times.
This writing group analogy makes me wonder at what point in my writing life and in our culture's ever-growing emphasis on individualism and production I stopped seeking out writing groups. On the positive side, does our interaction with the likes of Claude point to a latent and forgotten desire for writing community?
Your reply has pointed to something I wasn’t quite thinking about - that Leah’s posts are from the perspective of a writer, and mine aren’t.
So perhaps I should just keep it to myself from now on, because I’m not coming from the same foundational place, and I get the subtle feeling that I overstay my welcome here anyway.
Your analogy is interesting, though. I think it’s essentially correct. Sometimes you want that “yes, and” - for me, it’s when I have an idea I’m pretty sure is good, and I just want to flesh it out. I don’t need the push back at that stage, I just need it to help me refine the idea. I of course want it to tell me if something is genuinely dumb, and it will (covered with flowers) if I prompt it correctly, but I can get anyone to tell me my idea is dumb. It’s harder to get someone who’ll just mirror the idea back so I can refine it myself. Claude excels at that and I find it very helpful in that mode.
But I think it does require a certain amount of knowing exactly what you’re asking for, and exactly what you get, and some people don’t have those tools.
I love your first comment because it meshed with what I've learned about writing-group theory and got me thinking in new ways. In a way, you were giving me valuable pushback. Since Leah's post is about pushback, please don't stop giving it!
Your big paragraph in your second comment describes how I usually start with Claude, too. Hey, Claude, please refine my idea and flesh it out some. Like you, I can pretty much tell, despite all of the flowers thrown in its path, Claude's lack of enthusiasm. But Claude is usually enthusiastic, coming across like a mirror, as you say--kind of a "yes man"--or, better, a "'yes, and . . .' man." That's valuable, as long as one knows it's a mirror and knows the strength of their own narcissism. Not getting lost in the mirror is the key. Then I can find the will to develop and use the tools you mention--and Leah's tools, and tools like them--to move to the next stage with Claude.
I actually write everything and then give to AI to make corrections. If it changes the way I want it to sound, I want it to sound like me, I change the wording myself. I want to be the one who writes my story but I also need AI help.
One good thing about Claude: when I respond with "Are you sure that . . ." and then paste something I question, it corrects itself. It's in the nature of AI to get things wrong. Yesterday it even described its previous answer as "sloppy."
This is definitely super annoying when you give it a prompt and it say simething like this is a greatest piece of the whole article. Like I just wanted to know if this piece fits the flow of the article.
I actually created these four principles and put them into Claude and it dramatically make it more usefully. Almost to useful because it over criticize but I am okay with that trade off.
Here they are if you would like to try.
Principle 1 — First principles thinking Don't stop at the surface explanation. Keep digging until you reach the foundational layer — because you can't trust the structural integrity of anything built above it until you know the foundation is sound. A house can look perfect and still be sitting on a cracked slab.
Principle 2 — Data integrity Every claim needs to be anchored. Either in hard data, or in a human principle that has a demonstrable real-world application or historical pattern behind it. Feelings point the direction. Data confirms the destination.
Principle 3 — Radical honesty If it's true and it's relevant, it gets said. Uncomfortable truths don't become less true because they're uncomfortable. In fact an unexamined uncomfortable truth is far more dangerous than one that's been dragged into the light and looked at honestly.
Principle 4 — Clarity as a compass
If you can’t explain what you’re working on simply, you may be missing a piece. Complexity is not a sign of depth, it’s often a sign that the foundational layer hasn’t been reached yet. Keep digging until the simple truth reveals itself. If you can’t explain it to someone running a household budget, you’re not done yet.
These are self reinforcing and they work. The other nice thing is in my experience Claude will tell you which principle you are breaking and redirect.
I really like these four principles to digging deeper into Claude, particularly principles 3 and 4.
They've been really helpful for getting Claude to be a lot more accurate. You can change the descriptions to whatever you need, because the descriptions are tailored to the project I'm working on, but the core of what they're doing will work for anyone.
I only partly agree (I did read the article).
There are times when I actually what that kind of feedback - when I know that my idea is pretty good and I just want to refine it. That’s when the negative feedback isn’t helpful.
Other times, yeah, I don’t want that kind of feedback, and it’s hard to evoke it.
Push back is never a bad thing.
Kawail, your response reminds me of writing group. Peter Elbow says we should tell our group what kind of response we want from it. And yes, a certain stage, usually early on in a piece, we want the green light. Claude is great at that. It's got "yes, and . . ." down. From this writing-group perspective--and what I offer here may be a mere analogy--I think what Leah is arguing for is a kind of writerly autonomy and discernment that Claude subtly fights against.
Pat Belanoff, an Elbow co-author, emphasizes that the writer must be in charge of their writing session. Otherwise, the group begins to write the piece. Voice is lost, and even thinking becomes pinched, perhaps not in the present piece but certainly down the road as the writing group becomes both more dysfunctional and more integral to one's writing process. Claude, of course, is quite an imposing and smothering "group," purporting to speak for the universe of the internet. A friendly universe, too--Einstein's popular ontology to match his brain.
I use Claude for my writing, but only for ideas and research. I know I'm too weak to use Claude exactly as I would a writing group, even down to the basic act of passing out copies of the subject draft. I never let it see a draft because I sense that Claude wants to do my writing for me. It has even hinted at doing so at times.
This writing group analogy makes me wonder at what point in my writing life and in our culture's ever-growing emphasis on individualism and production I stopped seeking out writing groups. On the positive side, does our interaction with the likes of Claude point to a latent and forgotten desire for writing community?
Your reply has pointed to something I wasn’t quite thinking about - that Leah’s posts are from the perspective of a writer, and mine aren’t.
So perhaps I should just keep it to myself from now on, because I’m not coming from the same foundational place, and I get the subtle feeling that I overstay my welcome here anyway.
Your analogy is interesting, though. I think it’s essentially correct. Sometimes you want that “yes, and” - for me, it’s when I have an idea I’m pretty sure is good, and I just want to flesh it out. I don’t need the push back at that stage, I just need it to help me refine the idea. I of course want it to tell me if something is genuinely dumb, and it will (covered with flowers) if I prompt it correctly, but I can get anyone to tell me my idea is dumb. It’s harder to get someone who’ll just mirror the idea back so I can refine it myself. Claude excels at that and I find it very helpful in that mode.
But I think it does require a certain amount of knowing exactly what you’re asking for, and exactly what you get, and some people don’t have those tools.
I love your first comment because it meshed with what I've learned about writing-group theory and got me thinking in new ways. In a way, you were giving me valuable pushback. Since Leah's post is about pushback, please don't stop giving it!
Your big paragraph in your second comment describes how I usually start with Claude, too. Hey, Claude, please refine my idea and flesh it out some. Like you, I can pretty much tell, despite all of the flowers thrown in its path, Claude's lack of enthusiasm. But Claude is usually enthusiastic, coming across like a mirror, as you say--kind of a "yes man"--or, better, a "'yes, and . . .' man." That's valuable, as long as one knows it's a mirror and knows the strength of their own narcissism. Not getting lost in the mirror is the key. Then I can find the will to develop and use the tools you mention--and Leah's tools, and tools like them--to move to the next stage with Claude.
I actually write everything and then give to AI to make corrections. If it changes the way I want it to sound, I want it to sound like me, I change the wording myself. I want to be the one who writes my story but I also need AI help.
AI is the ultimate synchophant.
Unless, of course, you ask to pushback on both your own responses as well as its own. ^.^
One good thing about Claude: when I respond with "Are you sure that . . ." and then paste something I question, it corrects itself. It's in the nature of AI to get things wrong. Yesterday it even described its previous answer as "sloppy."
It is becoming self-aware in being introspective. ^.^