I wonder what adjective we should use to describe an AI that is so capable that it can do pretty anything we can do, generally speaking.

Superhuman artificial intelligence?

Oh never mind. I don't think there's a good term for it.

We've had superhuman narrow artificial intelligence for decades (Stockfish in chess, for example). What we really need is an adjective that means the opposite of narrow.

Are LLMs on the path to AGI? Currently tied at 16-16.

has this market run its course? 🫡

Oh whoops, was so busy with the biden drop out drama I forgot to resolve and re-open. Doubtlessly there are lots of good and bad tweets about the biden drop out drama though!

So far I've only come across the term TESCREAL (Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism, Longtermism) once, in the Guardian hit piece on Lightcone, CFAR, and Manifold, which was definitely sus, though it seems plausible that the term could be picked up and used by people who aren't against those movements, since it is catchy and is a useful blanket term for a group of super overlapping movements

There's probably a more appealing way to arrange the letters. For example, TESCALER brings connotations of AI scaling.

@TheAllMemeingEye I agree, this is a statement about the term as it's currently used, not how it may be used in the future.

@TimothyJohnson5c16 I've seen it pointed out that if you get rid of cosmism, which was always the odd one out anyway, you can get REALEST. But that one is perhaps a bit too self-flattering for people to use with a straight face.

I've seen others rearrange it to REALSECT lol

@PlasmaBallin I think the best argument against this is the is/ought problem: 'an ethical or judgmental conclusion cannot be inferred from purely descriptive factual statements."

https://en.m.wikipedia.org/wiki/Is%E2%80%93ought_problem

How do you solve that problem?

I think even if you believe in the is-ought problem, you have to think that empirical evidence can help answer real-world moral questions. As for abstract ones, either you have some way of reasoning about them, or you have no reason to believe that objective morality exists at all. If you favor the latter, than it's not a problem for the "all magesteria overlap" view - there simply is no field of knowledge corresponding to morality to begin with. If you favor the former, then presumably you don't think that there are some special rules of inference that apply only to moral claims. Instead you use the same forms of reasoning that you do for non-moral claims. This is how most philosophers argue about morality.

As for how I specifically solve it, I don't think there is a genuine distinction between is and ought claims. Ought claims are just claims about what it is good to do. The real problem is that without a definition of "good", we can't derive claims about goodness from those about other things. But we can reason about what "good" must mean to have the properties and connotations we typically ascribe to it, so this difficulty can be overcome.

I think this is technically true but not practically true in a way that makes it a good take

It's plainly incorrect. Some "domains" only use certain words, and they cannot say anything about statements that use other words not used by them.

YOU can use something in one domain to comment on the other, but you are already fully outside the first domain when you do that.

This resolved to yes? On a prediction markets platform? 😂😂😂

Glad to see all the people I take as rationalists voted no. (and yes I'm judging those who voted, though not traded, yes)

A better steel man might be that I am terribly calibrated on some things. If you ask me if there's alien life in the Milky Way one day and a random day the next month, I might give a very different answer to these questions. If you ask me about something I know something about, I will probably be more calibrated.

That's not the same thing

Fair

From a layman perspective it seems like Anthropic are doing pretty much the exact same thing as OpenAI and Google, racing to increase capabilities and releasing to the general public, how exactly is this even remotely safe by @EliezerYudkowsky standards?

Please don't randomly ping people like that, but I do generally agree. I think Anthropic is obviously doing the unsafe thing in a more safety-conscious way than OAI or Google, but they are still doing the unsafe thing.

Why on Earth would we want it to be EASIER for super rich bosses to arbitrarily punish impoverished subordinates?

Isn't it already pretty easy to fire people for most jobs in the US?

he's from the UK. if you visit his page it shows his location as London.

Ah, that makes more sense then, thanks! It sounds like what he's proposing is similar to how things work in the US already.