Sorry, I'm told that:

"Over half a million people in Gaza are facing catastrophic conditions, with extreme shortages of food and basic necessities. Another 1.07 million people are on the brink of famine."

Well, the brink of famine isn't going to be much better.

For parents of small children, you've probably seen Miss Rachel. Here's a video related to this that I saw just before famine status was declared. I know it's biased, but still worth a watch: www.youtube.com/watch?v=Qwpa...

(If you're confused by this because Sabine doesn't seem anti-science: she's a sort of two-headed person, as suggested by the commentary starting at 58:44. She loves scientific results, but for some reason also makes attack videos that paint scientists in general in a bad light.)

(correction: Glenn Beck devotee. Unsure about Fox)

(Sure LLMs can say they are 80% sure or whatever, but it writes that number with just the same process it uses to write hallucinations)

This 80000 Hours podcast didn't discuss those subjects either, but it did cover lots of important and interesting ground. www.youtube.com/watch?v=ny4X...

Trump's USA vs China: so many important facts and ideas, I wish everyone would see this.

They speak about how Democrats don't understand China either, which is a big problem. We should learn from China and we don't. Democrats' strategy was bad; Trump's is much worse. www.youtube.com/watch?v=UqBa...

But I still suspect that knowing facts well enough to care about them is the main antidote to this "influencer" bullshit. And for me, resistance to facts is the greatest mystery to be solved.

I think people aren't getting some very basic memos. For example: ourworldindata.org/much-better-...

(I don't *really* think people are more commonly full of shit, but among those that talk and thus supply training data, yes. And even the quiet types are often highly overconfident. Learned that from my father, a Fox News devotee.)

> none of which express uncertainty very much

Oh and it's trained on basically the whole internet too, which is also generally biased against expressing uncertainly. At best, people don't bother talking when they don't know. At worst (and more commonly) people are just so full of shit omfg wtf.

Oh and yeah, obviously an LLM develops language-agnostic concepts as the diagram shows, but less so in models with less "model scale" (I expect this depends more on compute than on model size, beyond some minimum. And training data is important.)

Anthropic's post: www.anthropic.com/news/tracing...