It used to be that a well-written text was decent evidence that the speaker knew what they were talking about. But now, LLMs produce well-written text, making it no longer signal education or attention to detail. Everyone learns to avoid em dashes, and countersignalers turn to all lowercase letters. AI companies notice and...

At least it's not the default, yet. But who knows where RLHF and incentives to imitate humans will lead? Texts full of typos, which the AI introduces too? Flexible use of the language, including my favourite compound-words -style and inscrutable loan words from German? Rotating fashion trends where what's considered good style will shift from month to month so that at least new posts can still be judged a bit.

Realistically, that game is almost over. The style of written communication is no longer useful indicator for anything. Maybe we should be happy, if it makes people actually focus on the actual content. But I find it more likely to do the opposite; we first check who wrote the text, before deciding if it's worth reading at all. Curation will help with this, too.

Recruiting has been increasingly moving to recorded-video applications for a while now, although I think it was mostly covid that made it stick. I presume it's primarily used as a tool for discrimination based on visible factors, especially when it can be implemented in an automated manner. And even more significantly, filtering out anyone who doesn't want to themselves to it. But it's a better signal for effort too, as typically you can only use the same appliation for multiple positions, and speaking well cannot be outsourced to LLMs so easily. Sure you could make the videos with AI too, but that's a lot more effort, and possibly detectable in live interviews.

The question follows: if the signaling value has been lost, why stick to writing the texts yourself? ChatGPT would do better work with it, if you keep correcting the inaccurate statements it tends to insert everywhere. I think that typically the writing process itself is a tool of thinking, and that is the reason to write it down in the first place. That's not true for when you want to communicate an already-crystallized idea though.

When I'm trying to say something, reading my own ideas seems to be more effort than writing things down. I'm not sure how much of that is just having to read it again when I've already written it, though. When programming, at least, it's way harder to read code and make sure it behaves as intended, compared to writing it yourself. I'd be surprised it the same didn't apply to technical. or all, writing. However, if the tooling around this improves we might soon have the LLMs check that all points were addressed, at least, and it should already be feasible with good prompting.

And lastly, maybe you could just enjoy the writing itself. I sure don't, most of the time. But I enjoy the idea that I have written something, and that's sufficient.