Comment by refulgentis

It's pretty easy to pattern-match LLM writing even when there's been a lot of work put into it, and it wasn't one-shotted by the LLM.

I bring this up because this is a very smart person, with an interesting story I've been waiting to hear for years, and an important point, but I couldn't read it.

Not because LLMs were involved: whatever, that's fine.

First, I'm reading then get an uneasy feeling when I see the "That wasn't X—it was Y.", which is a tell of GPT 4o at chatgpt.com or 4.1 on API. [^1 for sentences that got my attention]

Then, as I'm reading, I keep getting a weird "attention reset" buzz and I find it hard to follow. I note that there are no less than 15 sections, each 3-5 paragraphs. This is / was unnatural in writing. 0 flow.

Tips I'm taking away for myself:

- Actively read for "snappy" sentences from the LLM, and then actively eschew them -- you can't be familiar with every LLM's tells, but here, I'd try to notice the repeat structure in a completely different, and the cadence of the phrase ("snappy", in my verbiage)

- Marketing-type writing is best helped by an LLM if you can get it to give you individual feedback items that you have to address, or at least, a set of suggestions. Code works well with LLMs because the metastructure doesn't communicate meaning to a reader, there isn't "flow": in prose, the way the text was assembled can be betrayed by the structure.

[^1] A) "This wasn't just a cultural mismatch—it was a fundamental scale and complexity mismatch" B) "This wasn't about buying a struggling phone company—it was our strategic entry into the future of computing platforms"