Korva is a scroll of the internet’s best ideas.
What didn't work
For a first-time author, a book felt about 300 times harder than a blog post one thirtieth its length. Dictating a rough draft by voice to text and editing it afterward didn't work, because writing a book means wrestling with the sentences. Feeding an AI all of an author's blog posts to see whether it could mimic his voice didn't work either.
Feedback that does
What did work was feedback. Give it old writing samples, different drafts, and the chapter in progress, then ask where the voice is weaker than in other work, whether an anecdote shows up more than once in the book, and where the flow breaks. Ask it for everything that sucks about the writing, along with the weakest and strongest points. Some of the answers get thrown out and some get used, and it is hard to be objective about your own writing, while an AI digests all of it immediately and points out obvious flaws. A human editor is still phenomenal, and the AI adds more rounds of micro edits between the human ones.
Faster research
Research speeds up too. Perplexity is very useful for pulling information quickly, and Claude Code synced up to a computer can query files that hold information on companies.
Why is AI weaker at imitating voice than critiquing it?
Language models produce text by predicting likely next words, which pulls output toward common phrasing and away from the quirks that make a voice recognizable. Critique asks something different: compare a passage to other passages and flag where it diverges. Models handle that kind of pattern comparison across long texts well. The same ability explains why a model can notice a repeated anecdote that a tired author misses.
Why is it so hard to judge your own writing?
Psychologists call it the curse of knowledge. In a 1990 Stanford experiment by Elizabeth Newton, people tapped out the rhythm of well-known songs while listeners tried to name them. The tappers predicted listeners would guess about half of the songs, and listeners guessed about 2.5 percent. Writers face the same gap, since the author hears the intended meaning that a reader never receives.
More from The Prof G Pod with Scott Galloway
