First, the political macro backdrop. The extreme political polarization in the US leading to the so-called culture wars. This is a massive driver of toxicity on social media. This conflict machine is relatively new, people fondly remember centralized social media as being far less "political" just a few years ago.
Second, the mobile revolution. Which leads to a dumbing down of engagement. Before, people would sit behind a PC with a large screen and functional keyboard, enabling deep engagement as seen here on HN. Today, people sit on the toilet, look at a tiny screen with endless content, and any engagement (most never engage at all) is very shallow and lazy: a like, a retweet. It's not a conversation, it's amplification. In the rare case where somebody produces original content (a self-composed tweet), Twitter's format incentivizes hot takes and makes context and nuance impossible or impractical.
Third, amplification. It's a specific choice by Twitter (UX, algorithms) to promote and reward the worst opinions. It's a complete inversion from real life.
Hence, rather than stating that human conversation absolutely does not scale, I'd refine that conclusion. It does not scale in these particular conditions.