[Volrath89]

I asked ChatGPT to rewrite the slatesarcodex article as if it were PG:

"I Can Tolerate Anything Except the Outgroup:

Tolerance is important, but it should not extend to the "outgroup". An outgroup is defined as those who are perceived as fundamentally different or opposed. Tolerance towards the outgroup can lead to harm to one's own group and undermine social cohesion. It's better to tolerate ideas and individuals within your own group and be intolerant towards the outgroup. This creates a sense of loyalty and strengthens bonds within the group."

[josephg]

Its hard to focus on the style of that summary you gave when that "summary" doesn't capture what the slatestarcodex article actually says. That summary is vaguely upsetting in how confidently wrong it is.

The actual "I can tolerate anything except the outgroup" article is a ponderous critique of our human tribal tendancy to find an outgroup and hate them. The article finally comes full circle when the author notices the whole essay itself could be seen (ironically) as an attack on another outgroup.

The article is intentionally not making a clear point. The experience I had reading was to come away ponderous, thoughtful and self reflective. I'm not convinced that a simple summary or simplification would be able to give a reader the same experience. (But at least summarizing the article correctly wouldn't hurt.)

Simple language has its place. But its just a style, like flat design or modernism. Its certainly no panacea.

[inimino]

"Vaguely upsetting in how confidently wrong it is" is a good summary of ChatGPT in general, so the summary captures what ChatGPT is better than it captures what the article says. Very few humans are both capable of putting sentences together that coherently, and so incapable of a first-order understanding of the author's point. We don't have ready words for (or widespread recognition of) this kind of LLM-idiocy, because internalizing a probability distribution over words, sentences, and paragraphs without any correlated world model is not something humans do. "Bullshitting" comes close, but there's a difference of degree that becomes a difference in kind.

[Viliam1234]

> That summary is vaguely upsetting in how confidently wrong it is.

Being confidently wrong is an important step towards passing the Turing test. The upsetting feeling is the "uncanny valley". The stupid machine is now expressing arrogance normally reserved to self-confident humans.