Disclaimer: I'm a game developer.

This looks like interesting stuff. I recall playing around with Microsoft's similar, but more limited, demo that recreated Quake. I imagine that hands-on this probably behaves basically the same way.

Without knowing how they actually implemented this, I'm curious at how honest they are when they say it "simulates physics and interactions". The math behind rigid body, IK, collisions, cloth deformations, etc., etc. doesn't really seem to be in the wheelhouse of a generative learning machine. I'm guessing what this does is take the currently rendered scene and the user input as "prompts" and then it generates successive frames based on how the scene would most likely evolve over time. Essentially the same as how LLMs generate videos, except with the added real-time user input. So, no actual physics calculations - it just looks like it.

At least, that's my guess.

<quote>I'm trying to read the comments and this damned thing keeps popping up obscuring the view.</quote>

In Firefox with uBlock extension right-click on the pop-up and select "Block Element" and then "Create".

<quote><p>They've owned Rare for decades and revived exactly zero of their classic IPs, so I would take any of his spit-balling on what they might do with a huge grain of salt.</p></quote>

Do these not count?

Banjo-Kazooie Nuts & Bolts (Xbox 360)
Perfect Dark Zero (Xbox 360)
Killer Instinct (Xbox One)
Battletoads (Xbox One)

https://www.youtube.com/watch?v=wBzqOa9y02I

This excellent music video captures the melancholic escapism uniquely provided by gaming.
Years later and it still brings the feels... :,)

Enjoy.

Kent Brockman: Now, here are some results from our phone-in poll: 95% of people believe Homer Simpson is guilty. Of course, this is just a television poll which is not legally binding. Unless proposition 304 passes, and we all pray it will.

Bender: What should we point it at first?
Fry: I dunno. Try it on me!
[Zap]
Fry: Ow! My sperm!
Bender: Wow! Neat! Mind if I try that again?
[Zap]
Fry: Huh, didn't hurt that time.

Yes, I know. ;)

Designed by me.
Built by me.

And you best hope, not pointed at <a href="http://www.youtube.com/watch?v=i68cEsALWt0">you</a>.

I am certain you are mistaken on this:

and somewhat on this (it's correct, but I think you're missing the point of what this means):

From the wikipedia link you specified:

An algorithm with O(f(n)) computational complexity requires on the order of f(n) operations to complete for input size n in the worst/average case (it could much require fewer in the best case, but we don't care about the best case, in general, or in this discussion, in particular).

A problem that is O(f(n)) has no known algorithmic solution that has computational complexity better than O(f(n)).

An "operation" in the Big-O sense refers to the logical/mathematical concept of a fundamental unit of computation (think Turing Machine). Whether an operation is implemented wholly in hardware or software (or half-and-half or whatever) is completely irrelevant to a discussion about it's computational complexity. An operation is simply a step of execution that must be performed by an algorithm and Big-O tells you very approximately how many of them it will take for it to arrive at a solution (on average/worst case).

Anyways, I think my point - that computational complexity kills Kurzweils singularity speculation - stands, but this topic has long since faded from front page discussion, so I'll let it drop. :)

D.

That doesn't sound right.

From your example, a computer that can run an O(2^n) algorithm on up to n=1000 inputs in less than or equal to 1 second, must be able to execute at least 2^1000 operations per second! You couldn't do that with computer composed of 1 processor for each atom in the universe, each running in parallel.

Also, n is the number of input to the algorithm. It has nothing to do with its implementation. I think the value you are thinking of is k, which is a scalar that represents computational overhead of the algorithm, and that is certainly implementation dependent. Big-O says nothing about k.

Lastly, you're right, there are good approximations for many hard problems. However, there are many hard problems for which no good approximation can be found (or at least finding the approximation is also a hard problem). For these problems any uber-AI will be no better than us, lowly, normal intelligences.

D.

The thing that bothers me about Kurzweil's super-intelligence/singularity thing is it completely ignores computational complexity.

Every problem has a strict lower bound on the number of operations (and/or amount of memory space) required to solve it. Even simple sorting requires that every element to be sorted is looked at, at least, once. In other words, in the best possible case, sorting can be accomplished in O(n). No amount of super-speed intelligence can reduce that.

Further, difficult problems stay difficult regardless of how much processing power you throw at them. Take this example from the wikipedia page:

So, even ignoring all physical constraints, intelligence is computationally limited. No matter how much "smarter" we can make an AI (or whatever) there are problems whose complexity is great enough that any increase in intelligence quickly becomes irrelevant.

D.

ps. Apologies for any inaccuracies in this post. It's been quite a while since I took this stuff in university. If anyone more knowledgeable in complexity theory cares to correct me, I'd appreciate it.

I think part of the problem is the perception of teachers as shown by the media and pop culture. For instance, look at North American cartoons. At best teachers are depicted as ancient, boring, and/or uncaring. At worst, they're downright monstrous. Contrast with the view of teachers presented in a lot of anime. They are intelligent, beautiful, and often even super-heroic.

Here's an excerpt from the press release as posted on Kotaku: http://kotaku.com/5028663/hasbro-sues-over-scrabulous-facebook-game

This burns me up in so many ways. >:(

Not only is this a trademark case rather than a copyright one (as far I can tell, no lawyer here), but as a Canadian, it disgusts me that they're using the DMCA to dictate what can or cannot be published in Canada. We're having a tough enough time keeping crap legislation like that out of our country without amoral corporations smearing it across the boarder.