LLMs are fundamentally a dead end though. If we ever create AGI, it will be a qualitatively different thing from an LLM.

“For too long, the average American has been squeezed by the big banks and financial elites,” Trump wrote. “It’s time we take a stand — together.”

Do people actually believe for a fraction of a second that Donald Trump of all people is on the side of the average American? And that the way to “take a stand” is with a crypto pump and dump scheme?

If you click the “talk” link at the top of the page, there’s a discussion about the original header image for the article, “Boops boops in a bucket,” depicting just that. Because the image committed the unforgivable sin of being slightly whimsical on a Wikipedia article about a fish, someone went on a lengthy crusade to get it replaced.

The talk page is an all-time classic of petty Wikipedia drama.

It’s named as attack ad bait. “So-and-so voted against the Kids Online Safety Act” sounds bad to the uninformed voter, and there are a lot of uninformed voters.

You’d be right in this case too. It’s extremely sketchy, it’s pretty much absolute censorship power with only an informal promise that it won’t be used for anything nefarious (but a refusal to actually codify anything preventing that). “Harmful content” is left very conveniently vague.

I feel like there’s a primal part of my brain that would tell me not to drink something that looks like there’s mold growing on it that would be difficult to get past.

It’s a fundamental problem with the tech in general. It inherently has no concept of “I don’t know” and will just be confident, specific, and wrong.

For example maybe you need a working time machine at your destination, such that the earliest point possible to travel to is the moment the first time machine was switched on.