Admit that it's already too complex for us to understand how it arrived at a particular response or behavior. Just like a person is, qualitatively.

We seem to get by not really understanding how or why other humans behave or respond, in particular cases. We get by by understanding generalities about behaviour, and by creating societal norms, laws, and social contracts.

Maybe we need the same for AIs. But who should the laws and social contracts apply to? Saying it should apply to AI developers is like saying parents should be blamed for the misdeeds of their offspring. That consequentialist morality and justice used to be the case, but not in the most advanced societies today.

A bit of a quandary.

Coordinated Information Apparatus Battlefield Earth

Not "In fact" but rather "Additionally", because those two observations about randomness are independent of each other.

There is a clear difference between
1. very occasionally "a pattern appears" randomly in the random data sequence (and yet it is apparently still perfectly random) ...
and
2. a perfectly random sequence can nonetheless be an encoding (representation) of something else that is non-random.

I must admit it is hard to think clearly about the semantics of "randomness".

For example, does an infinite-length random sequence necessarily contain non-random subsequences, in fact, ALL non-random subsequences?
This is the Boltzmann brain issue generalized to non-physical random sequences.

I guess, when we "check" a sequence for randomness, we have to ASSUME we are not so terribly unlucky as to be checking during the infinitesimally rare time period of the sequence generation in which it's generating regularity purely by accident. How many times its it sufficient to sample different parts of the sequence to be sure it is not "usually somewhat regular". What if we are extra unlucky and are in the middle of a very long accidental regularity in the sequence?

So can we ever be sure something is random just by looking at the random data sequence?
Or must we instead analyze its generation process and say nothing could predict that process exactly, so we may as well say its output data is random.

a "true" random number generator will still generate a stream of data which has an analyzable pattern in it, purely by chance. (Infinite Monkeys problem).

If your is-it-random-or-not evaluator happened on one of those randomly coherent strings, would it say "hey this isn't random!" ?

If fact, perfectly random data is the most efficient (most compressed) representation of any information. (Smallest Kolmogorov Complexity representation).

It seems there are a lot of paradoxes with the concept of randomness.

I don't foresee that there will be any Uber and Lyft drivers by 2030, except a small and diminishing subset to serve the niche of people who won't get in a self-driving taxi.

And how TF is Vbasic more prevalent than Perl?

Agreed. But this should be a Fedora Spin so those who want it can get it easily and thise who do not, need not even see it.

are required to re-invest all of their profits in their operation, and not distribute them among owners and shareholders.

FWIW, this Nobel Laureate (Hinton) disagrees with you about consciousness. Maybe you should be less certain about your credences.

Anyway, there was some discussion about the Goblin Problem and its relation to consciousness it in the latest Last Week in AI. Always worth a listen.

Remember Solyndra?

Jesus, this is a brain-dead take. Anthropic is the AFAICT only for-profit company that takes AI safety and alignment seriously.

Dario (and others') work on Constitutional AI is AFAICT the only realistic solution to this very real engineering problem. And they publish what could be their secret-sauce constitution, verbatim, under a Creative Commons license.

The people behind ChatGTP are essentially designing a neural net training system and general purpose inference and expression system based on the neural net. Then they are feeding in pretty much all of the human expressions of knowledge or communication in the public or cheaply commercially purchasable (e.g. used books) domain, as learning material for the neural net.

Essentially, they are creating a very large library with a very fast, efficient, disinterested librarian function to help a person find what they're looking for. I'm sure a lot of crimes have been planned in the past with the assistance of library visits. Do we start rounding up librarians? Or, in a more apt analogy, do we start rounding up the university professors of library science, who trained all those librarians?

are way, way better (and more cost effective) than anything you're driving. 60mph in 4 seconds and drives itself. You should try one before commenting. By the way, enjoying those gas prices?

you're welcome.

This. And note the headlines for what afaict is the same product. For Meta they are enabling sexual predators. For Apple they are ... making style and design choices?