Thanks for that! I marvel at what can be done with AI (e.g. Claude) and struggle with the problem of attribution. My first use was mainly as a glorified search engine. Then it became a research assistant (or a roomful of them), and finally morphed into a gifted dissertation advisor. And I haven't yet understood it's full value, so thanks for setting expectations.
This is great. I appreciate posts like this. Looking forward to trying some of the things you showed here.
I use Claude Code as an editor in a similar way but have a draft assessment skill that reads through the draft and creates a markdown report for different passes to identify issues with the argument, fact check, or writing issues. This has been very helpful to get quick feedback and revise and then have CC read through again against the original pass and see what has been done.
Like you I also do semi-automated editing for typos and thing like that. I’ll have the style report from the assessment skill and read through issue by issue and I approve the change or skip.
Using the one-sentence-per-line technique helps a lot. Not only is line-diff git actually pretty usable for text then but pushing sentences around in the text becomes much easier. Even without AI.
Hi Mike! I love these posts. I use Claude for similar proof reading and citation management tasks. When I do so, I ask it to edit word documents with track changes and then review and approve them. This seems to accomplish the same thing as your git workflow. But is there a risk Claude will edit some word doc text and not apply it with track changes?
Thanks for that! I marvel at what can be done with AI (e.g. Claude) and struggle with the problem of attribution. My first use was mainly as a glorified search engine. Then it became a research assistant (or a roomful of them), and finally morphed into a gifted dissertation advisor. And I haven't yet understood it's full value, so thanks for setting expectations.
Love these
This is great. I appreciate posts like this. Looking forward to trying some of the things you showed here.
I use Claude Code as an editor in a similar way but have a draft assessment skill that reads through the draft and creates a markdown report for different passes to identify issues with the argument, fact check, or writing issues. This has been very helpful to get quick feedback and revise and then have CC read through again against the original pass and see what has been done.
Like you I also do semi-automated editing for typos and thing like that. I’ll have the style report from the assessment skill and read through issue by issue and I approve the change or skip.
Using the one-sentence-per-line technique helps a lot. Not only is line-diff git actually pretty usable for text then but pushing sentences around in the text becomes much easier. Even without AI.
Hi Mike! I love these posts. I use Claude for similar proof reading and citation management tasks. When I do so, I ask it to edit word documents with track changes and then review and approve them. This seems to accomplish the same thing as your git workflow. But is there a risk Claude will edit some word doc text and not apply it with track changes?