Welcome to the US government. How can we secretly piss you off today?

[Hauldren Collider]

I'm wondering why actuaries are all that needed, and if they'll be less in demand as risk models get better.

As they say, there are only two certain things in life: Death and taxes. Actuaries are as secure as anyone.

[Asher]

Have you even seen a line of my code? You've not seen any of the projects I've worked on, any of the solutions I've designed, or any of the critical thinking problems I've solved outside of software.

This is fair; I'm only extrapolating from technical discussions you've been involved in, where you couldn't seem to understand theoretical issues over learned dogma...

No, see -- in that thread you and KH were obsessed with theory while I've dealt with the problem in reality.

I've designed embedded software, I don't think either of you have. You guys were clearly out of your element when discussing the kinds RNGs that run on embedded systems.

Go write a Java program constructing ~1 million numbers between 1 and 99, in streams of 6, using java.util.Rand and run it inside J2ME. Have it round up the outliers in frequency of occurrences of individual numbers.

Compare the results with seeding just once on the program start, and seeding to the value of the instruction counter at the start of each stream of 6.

It'll blow your mind, apparently.

All the theory in the world is useless when faced with the realities of the world. I suspect you and KH are used to robust, secure RNGs that run on compute clusters and have no ****ing idea how they differ from those that run on low-power ARM chips...

[Kuciwalker]

Also, re: actuaries, there is a legislative floor on demand because of the mandate for actuarial certification of defined-benefit plans.

[Kuciwalker]

No, see -- in that thread you and KH were obsessed with theory while I've dealt with the problem in reality.

I've designed embedded software, I don't think either of you have. You guys were clearly out of your element when discussing the kinds RNGs that run on embedded systems.

An LCG is both trivial to implement in an embedded system and also does not require re-seeding in this context.

[Asher]

An LCG is both trivial to implement in an embedded system and also does not require re-seeding in this context.

I hear you talking theory again. I don't see you testing your theory. Stop wasting time.

And java.util.Random is a LCG with a 48-bit seed, so you're in luck.

[Asher]

See, you don't understand economics either. Productivity improvements in a given sector are just as likely to increase quantity consumed as decrease it...

You apparently do not know how to read. Which is a shame for such a well-rounded individual.

[onodera]

I think gays of the world would pay top dollar for an Asher/Kuci/HC sex tape. The tension...

[Asher]

While I'm waiting for Kuci's stunning results (which I suspect will never come once he sees them...), it should be noted that for all of KH's vaunted expertise in RNGs, none of his RNGs use the method used in embedded systems (LCG). Which makes his experience somewhat worthless. In fact, the low order bits are quite terrible in LCG... Most good game developers make liberal use of bitmasking to work around that.

[KrazyHorse]

You're a ****ing idiot, Asher.



LCGs with a decent set of choices for parameters are more than capable of doing the job required in the other thread, and reseeding WILL ONLY DEGRADE THE STATISTICAL PROPERTIES.

[Asher]

Why don't you use LCG for your Monte Carlo simulations?

[KrazyHorse]

Once more: never send a software engineer to do a quant's job.



There's a reason you and your ilk are so much lower on the totem pole than us...

[Asher]

Once more: never send a software engineer to do a quant's job.



There's a reason you and your ilk are so much lower on the totem pole than us...

Quants are seldom on any totem pole, unless you count the ones strung up on them in 2008...

[KrazyHorse]

Why don't you use LCG for your Monte Carlo simulations?

Because there are better available choices, but for an application that merely wishes to avoid improbable clustering over a few thousand/million draws, it's perfectly adequate.

[Asher]

LCGs with a decent set of choices for parameters are more than capable of doing the job required in the other thread, and reseeding WILL ONLY DEGRADE THE STATISTICAL PROPERTIES.

Define decent set of choices for parameters. Cross-reference your requirements with java.util.Random.

[KrazyHorse]

Quants are seldom on any totem pole, unless you count the ones strung up on them in 2008...

We have more quants now than we did then.