Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Frumple]

True, but eyyyyyy. I'll be dead by then if ever so what the fuck do I care?

Heeeey, it's a fair longshot, but we have been making some pretty substantial strides in regards to kicking that aging thing in the ass. Not entirely impossible that'll be a conditional statement before you kick it.

Also, re: the bug thing, it totally depends a lot on the bug. Bees, ferex. They've probably got our collective balls in a beehive at least as much as we've got their collective thorax ready for burn order. There's a fair chunk of stuff we're still pretty dependent on if we don't want basically everything (human) to die.

[Criptfeind]

Well, that's certainly true. But I'd still say conquered is an okay thing to say about bees, we might be reliant on them, but we still control and pretty much enslave them pretty well. Of course, they don't mind, and from their own perspective (if they had a perspective) perhaps they would think they conquered us. Which might be fair. But ayy you know.

[chaotic skies]

The problem with the word "Never" is that it doesn't actually mean "Never" when most people say it; it normally is implied to mean something like "Not now, nor anytime in the forseeable future"; which is not what "Never" actually means.

[Egan_BW]

If you believe in enough multiverses, anything can be true.

[Sergarr]

I think that the AI being unable to adapt "beyond the presumptions of the programmer" is a good thing. It makes bugfixing the problems in AI so much easier, not to mention it reduces the chances of bad unexpected stuff happening.

And you no longer have a significantly useful AI, you merely have a tuppeny-hapenny ElizaBot that has no intelligence and can only deal with input anticipated by the designer.  (My own first Eliza program was typed into the BBC microcomputer in the early '80s... Knowing how it all works takes most of the magic out of it.  Tell me more about your mother.)

It seems that you're somewhat out of touch with the modern state-of-the-art neural networks. Nothing they do is actually "beyond the presumptions of the programmer" (after all, they don't actually change their own code or anything weird and unpredictable like that), but they already do some seriously powerful stuff. And this is just the beginning of the intelligence revolution.

If you believe in enough multiverses, anything can be true.

Nay. 1 will never be equal to 0, as long as it is defined by our mathematical axioms.

[Frumple]

Unless the axioms change, anyway~

Actually, axioms are quite possibly the absolute worst thing to peg a never to, since they're entirely arbitrary in nature and only as immutable as the people using them (which is to say not at all) :V

[Sergarr]

Unless the axioms change, anyway~

Actually, axioms are quite possibly the absolute worst thing to peg a never to, since they're entirely arbitrary in nature and only as immutable as the people using them (which is to say not at all) :V

Axioms are completely and utterly immutable. There's nothing that can actually change axioms, they're utterly and completely embedded in the mathematical super-structure that unites all knowable and unknowable universes.

[Shadowlord]

Nay. 1 will never be equal to 0, as long as it is defined by our mathematical axioms.

For some reason I misread this as "I will never be equal to 0"

[Egan_BW]

i can think of a few multiverses off the top of my head where those exact same axioms claim that 1 = 2

[Sergarr]

i can think of a few multiverses off the top of my head where those exact same axioms claim that 1 = 2

Those ones are logically incoherent and thus can't possibly exist.

Well, actually, they do exist, but they belong in the "0 = 1" mathematical ghetto, which only exists on a basis of not actually existing:

Spoiler: look at how far it is from the decent folks at supercompact (click to show/hide)

[Egan_BW]

There are exactly 648 universes where 648 = 0 makes sense in such a way that you personally from this universe would understand.

[Frumple]

Axioms are completely and utterly immutable. There's nothing that can actually change axioms, they're utterly and completely embedded in the mathematical super-structure that unites all knowable and unknowable universes.

Axioms are literally stuff we invented to start thinking from. Like, that's pretty much the exact definition put in simpler words. They're entirely human constructed and are as utterly mutable as the people making them feel like making them.

You're thinking more along the lines of a platonic ideal or fundamental force, serg, not an axiom. Axioms are logic contrivances we use because they make things (much, much, much) easier, no more, no less.

Maybe we're dealing with some kind of translation error, here? Either that or you defined a different meaning from the general usage somewhere up a bit where I wasn't paying attention, which if so, well... okay. You're using the word wrong, but fair enough

[Starver]

It seems that you're somewhat out of touch with the modern state-of-the-art neural networks. Nothing they do is actually "beyond the presumptions of the programmer" (after all, they don't actually change their own code or anything weird and unpredictable like that), but they already do some seriously powerful stuff. And this is just the beginning of the intelligence revolution.

Very interesting, but I don't see any AI in that.

I also missed the bit about why the author wanted to analyse/compare fonts, so perhaps I missed a previous blog/whatever about the actual aims and expectations. The most interesting part to me was the morphing between cases¹, although you must admit that some of those intermediate forms are effectively unusable (e.g. g to G on that static grid can be clearly seen to have forms indistinguishable from a rough single-story miniscule 'a' form) so that it shares the quality with standard image-morphing techniques². I see no 'understanding', it's just a fancy (and impressive!) statistical analysis then poked and prodded by the human in charge.

Unless, again, I miss something.

¹ Here credited to manually selecting fonts with a smallcaps paradigm, to morph between, but would have been as easy to ask of the analysis to produce vectors from all a-z forms to A-Z forms, but maybe also accented (ä into Ä, etc) and ligatures (æ into Æ), assuming that wasn't beyond the invisaged capabilities of the font-comparer because of the 'fiddly bits'.

² In which the more unguided the initial 'gridding' is, prior to the cross-fading of intermediate distorts, the more it sends you through an intermediate that is not just an uncanny valley but actually 'shows your working'.  Whereas, with understandng and intelligence behind the triangulations, not only can (say) the legs of the source cat be engineered  to become the wheels of the target vehicle but, with effort, the warps can 'wrap' the feet around, and 'hide' the offside leg-wheels behind the descending body that will fully obscure them in a 'believable' way. But that would need one helluva image-recognition algorithm to determine and enact without an actual operator/artist in charge.

[Sergarr]

"Understanding", in my opinion, is simply a result of having a complex and sensitive algorithm, that can successfully solve a given problem, and can quickly adapt to a slightly different version of a problem. Neural networks are currently working very well in that direction, in that they can 1) solve a lot of problems, by the virtue of being universal classifiers, and 2) a slightly different problem usually has a similar solution.

Here's a more relevant, and more visually powerful, example of neural networks being trained to understand different kinds of food, and to identify pictures with the same kind of food on it:

Spoiler: relevant images under cut (click to show/hide)

I think that this algorithm can actually (somewhat) understand things about food. If you think otherwise, it would be a great pleasure for me if you told me why.

[Starver]

I think that this algorithm can actually (somewhat) understand things about food. If you think otherwise, it would be a great pleasure for me if you told me why.

For the moment just going by the image, it looks like it has classified images featuring chips (US: fries) or rice or ?stews? or that come on multiple dishes. But I'm not convinced that it 'understands' food. It's image-recognition that classifies plates with signifcant long-things together, significant granular piles together, etc...  If presented with an un-stacked pile of logs or a sand-dune, perhaps, it might 'understand' these to belong to the chip-based and rice-based picture sets. Until trained otherwise.

It's not unimpressive, but the anticipation was doubtless that something (granularity? repeating shapes?) would differentiate those images, and the resources to find these matches (and so many more types of match, such as whether it is viewing a car or a dog or a tree) were given to it and then honed down to recognising constrained subsets and classifying them.

If I asked for images of vegan food, or pizzas with stuffed-crust and not too much pepperoni, would such a request be any more sensibly provided for than when askng for a beige pickup-truck with extra peppers on top?