Some vibe-writing maxims:
- While writing, have two windows open: one for the writing, and the other one for ChatGPT. Previously he used to consult ChatGPT a couple of times for internet research, but now the role has transitioned to be a more conversational thought partner, helping you riff-raff on the idea for the essay. (Let’s say you’re writing an essay about tarrifs, and you want to understand what the general consensus is, so that you could provide an unique insight that can counter-position to the general consensus. It’s particularly helpful to provide this). Deep Research is also effective in productizing the ‘secondary research’ part of any essay where you have to crawl all over the internet to then derive the analysis from (Deep Research is basically a Mckinsey level research analyst at your doorstep helping you with this part of the process)
- It also helps with alternative word seeking. Let’s say we want to find an alternate word to “outrage”, and you can give some references to ground them so that it generates a close cousin you’re expecting. This can be done on a word level, sentence level, or even on a paragraph level.
- AI can help in fleshing out the essay with a meta-analysis, your first pass roughest-rough draft (even half-way through), you could probe ChatGPT to analyse if you’re getting to the core of the argument quickly? is it building momentum? is it a good opening, or there could be some alternate openings, closings?
- In essays which involve more argument analysis, I also apply some of these questions inspired by Lex Fridman’s interview style. For eg:
| Question | Purpose |
|---|---|
| Can you steelman the case for …? | Elicit strongest argument for.. |
| Can you strawman the case for..? | Address counterarguments for.. |
| Can we break this down into first principles… | Analyse from basic axioms |
| What happens if we take this to the extremes.. | Test robustness when it comes to the edge cases |
- The more we use AI-assisted writing, there is a big risk of the writing sounding like AI. The line needs to be tread carefully where you use AI for asking good questions, identify flaws, improve/edit content, not to actually put words in your mouth. I also use this quite extensively to remove the filler words such as dwelve, em-dashes and the usual AI-generated copypasta: `Identify and remove any generic or overused phrases that make the writing sound artificial. Replace common AI phrases with more original, specific language.
