Every time I post something, I weigh whether it's worth the effort of structuring the idea and finding the right words. I think it usually is, and not for the audience. The public square has moved online, and writing in it is a small civic act, like voting. We owe it some care: to reason rather than react, to admit what we don't know, to counter-balance the noise instead of adding to it.
There is a newer reason, though, one we are only beginning to see. The archive of all this writing, books and forums and blogs and arguments, compiled into terabytes of text, is the raw material for the tools we are building. Language models are trained on it. Mathematically the thing they optimize is narrow: predict the next word given the previous ones. Nothing about answering a question, writing code, or proofreading a paragraph was implemented directly. Those abilities are emergent, squeezed out of having read an enormous amount of what we wrote. The source data is the power. And the source data is billions of small public discourses.
The hard part is what we call alignment, getting the result to be useful, honest, and harmless. This turns out to be difficult, and the difficulty is telling. We have built an interactive mirror of modern humanity, and we don't like what we see in it.
The answer is not to delete the opinions that bother us. They are natural; a radical opinion is easier to put into words than a measured one, because that is how we react to stimuli. The answer is older than the technology. We practice the exercise of dialogue, and we take care how we express ourselves.
The mirror only knows what we put in front of it.