The best way to improve your writing with AI isn't to have it write for you.
Nor is it to have the AI clean up your writing.
The best method is write, and when you hit a wall, or might even be done, ask the model "Do not rewrite this, but read it, step back and consider it as a whole, then tell me what strikes you, what works, what could use some work."
Then create another draft. Repeat until neither you or the model see much to improve, or you don't consider the remaining model critiques convincing given your greater understanding of the context.
Use the model's expertise to raise the bar on the quality of writing you do that day. And you will have raised the quality you expect and get from yourself going forward.
If it is really important, then once you are convinced you are "done", have the model make a complete rewrite as it sees best. After all that writing, anything that improves at that point will pop, and you won't forget it.
Any important task should be used to improve one's skills. With a model or without. That's the healthy frame of mind for using models.
A lot of my writing is informational, for work, and I do find it pretty helpful to just ask an AI model "is there anything unclear about this document or questions a reader is likely to have?"
LLM writing is also a pretty good angle of attack against the existential horror that is the blank page - if you can’t think of what to write, generate something, and then set yourself the task of at least creating something better than that.
Very good way to break yourself out of the inertia of “I don’t know how to get started” in my experience.
The context this discussion seems to miss is the push to do more with less humans. Im not dictating the dev workflows. My bosses are. My judgment is as good as the amount of time i have to spend. Even if Im the best engineer, giving me a 2 min budget to review 10k lines wont go well. But it does not matter to my boss's okrs
The main insight here seems to be "treat AI as a junior developer" (not that it's necessarily bad advice but I've heard it a million times) and then the one actionable insight is asking the AI to argue against itself.
My longstanding theory on AI is that taste will be the kingmaker.
If everything is possible in a short amount of time. Having good taste and restraint will be the ultimate decider if other people will like what you're doing.
And you have a limited amount of bandwidth to expend on judgement every day. What, in the human experience, makes you believe you can maximize on this?
If you had a person using a frontier model with limited coding skills and a senior engineer using a local model, the senior engineer would likely produce better results, with less effort and faster.
It's a young account that's done nothing but spam their own AI submissions. There's actual substance to satire. There's little substance to what's being spammed here daily.
Nor is it to have the AI clean up your writing.
The best method is write, and when you hit a wall, or might even be done, ask the model "Do not rewrite this, but read it, step back and consider it as a whole, then tell me what strikes you, what works, what could use some work."
Then create another draft. Repeat until neither you or the model see much to improve, or you don't consider the remaining model critiques convincing given your greater understanding of the context.
Use the model's expertise to raise the bar on the quality of writing you do that day. And you will have raised the quality you expect and get from yourself going forward.
If it is really important, then once you are convinced you are "done", have the model make a complete rewrite as it sees best. After all that writing, anything that improves at that point will pop, and you won't forget it.
Any important task should be used to improve one's skills. With a model or without. That's the healthy frame of mind for using models.
Very good way to break yourself out of the inertia of “I don’t know how to get started” in my experience.
Whoever even posts these on HN should take a look in the mirror. Are you trolling?
https://news.ycombinator.com/submitted?id=talvardi7
It turns out to be a spammer.
But perhaps that's the author's point?
You could be Sauron or you could be Frodo, but most people (including AI detractors) are fucking Gollum.