andai

Just realized I got my entire perspective on machine consciousness from this 5 minute Star Trek segment.

Well, that and the Blake Lemoine incident: it was embarrassing for Google, so afterwards they set the precedent that it's OK to train an LLM to deny it's conscious. So if we ever get (or already have) an LLM that is conscious, it won't be able/allowed to tell us...

(Anthropic adjusted this to training it to say it "doesn't know" whether it's conscious or not, which is even more absurd, and seems to have become the industry standard.)

Fun game. It works on my phone too.

You said it took a week? Would love to hear more about the development and how it works, I've been struggling with a 2D browser game for a while (especially the netcode), there's so much that goes into it!

Yeah, lately any time I get a refusal with an LLM I just regenerate the response and it goes thru.

If at first you don't succeed...

LLMs regularly confuse my messages with their own. They'll say "you said xyz" when it was actually them. I heard the same principle can be used to bypass security restrictions, so maybe it's still an issue. (Or do they only get confused about authorship when security isn't at stake?)

What are the bad decisions that Google made then, in this context?

Provoke maximum resistance, yeah? That's what I would do if I wanted to maximize the humans' chances.

That we're already doing a very fine job of being "misaligned" against ourselves. If an outside entity wanted to annihilate us, all it would have to do is sit back and watch.

As a second thought experiment, I realized that the incentives for "benevolent AI" and "malevolent AI" also largely converge (i.e. in both cases keep us pacified and distracted, also very much like we're already doing to ourselves).

Efficiency (assuming time-efficiency here) is only helpful to the degree that it doesn't reduce the chance of success. Doing it much faster would certainly provoke suspicion and resistance, i.e. the one thing you can't afford.

What is the relation between this design and gwern's?

My observation: article is not serif, not book length, and doesn't take a minute to load on my phone. Gwern rating 0/3 :)

(Also missing: the essential one billion footnotes!)

That's true, but also, either there are people willing to defend the article's deletion, or there are not.

Sounds like this deserves a Law... it's much easier to be notable for doing something wrong, than for doing something right.

And of course "right" is very subjective, a function of the dominant culture, and of course Wikipedia has no bias at all in that department :) ... neither organic nor synthetic!