• Member Since 11th Apr, 2012
  • offline last seen Yesterday

Bad Horse


Beneath the microscope, you contain galaxies.

More Blog Posts758

Mar
13th
2020

Hugo nominations due Friday the 13th · 1:56am Mar 13th, 2020

A reminder that Hugo nominations are due tomorrow, Friday March 13th, and that, as far as I can tell, fan fiction is eligible for all fiction Hugos. Fan fiction, fan blogs, and other SFF-ish writing are eligible for the "Best Fan" Hugo. If you're eligible to make nominations, that means you're registered for Worldcon and you already know how to make nominations.

I'm not personally very eligible; the only fiction I published was the last chapter of "Moments", and "Think Inside the Box", that script I published in Worst of Bad Horse. I can't post it on fimfiction due to knighty's stupid rule against scripts. I'd had this deluded hope that some of you would pay a buck via Patreon to read it online, but that didn't happen. I give up. It's here, you cheap bastards.

Did I mention that I'm really pleased with the essays in Worst of Bad Horse? Which, as far as I know, no one except my editor RBDash47 read. (And which I still owe some people copies of. I haven't forgotten!) It is however not so much a work of fantasy as a fragmentary system of philosophy organized around stories you've already read.

I did self-publish Stupid Unicorn (PDF here), which is technically eligible for Best Graphic Story, in much the same way that I am technically eligible to run for President in 2020. Ilya Leonov's going to do a video reading of it for YouTube! Well, a reading, with the pictures.

Report Bad Horse · 329 views · #Hugos #Stupid Unicorn
Comments ( 12 )

I am technically eligible to run for President in 2020

Fun fact: this is three demographic facts.

I did read the essays in Best of Bad Horse, and showed some of them to my brother (who is not into ponies).

Comment posted by JustSomeRandomUsername deleted Mar 31st, 2021

5219336
Definitely by July 1, because there are several people at Trotcon I owe copies to.

Hopefully earlier than that. I ran out of copies, and want to revise it before printing more copies, and haven't yet.

5219335 Neat! Did he think they were crazy?

I give up. It's here, you cheap bastards.

Well, I believe in the Starving Artist, see.
Money ruins vision.
It also buys me cake. Soooo...
:yay:

I'd had this deluded hope that some of you would pay a buck via Patreon to read it online, but that didn't happen.

That's nuts and does not work for digital goods. With digital goods, you monetize enterprises through a value proposition, and you monetize consumers through traps, habits, and addictions. You were treating readers like enterprises rather than like consumers. That's just silly.

Did I mention that I'm really pleased with the essays in Worst of Bad Horse? Which, as far as I know, no one except my editor RBDash47 read.

I read the one with pictures. On chains vs. hierarchies, smelling things, chaos, and Aristotle.

By the way, there are some infinite things, namely Categories, that contain (faithful) representations of all their own sub-things. A functor gives the equivalent of a "subset" of a category, and functors form a category. So the category of categories contains all of its own subcategories as a category of functors.

This is sort of like defining a set that contains its own powerset as an element, not as a subset. I think that would be legal within ZF if you threw out the Schema of Specification axiom, which we all know is just some bogus hack created by a few mathematicians that were butthurt over Russel's 1337 sploit. It's been nearly 120 years since that particular pony sprouted wings. It's time to move on.

By the way, what do you think about cleaning data for fun?

5221427

With digital goods, you monetize enterprises through a value proposition, and you monetize consumers through traps, habits, and addictions. You were treating readers like enterprises rather than like consumers. That's just silly.

That suggests that one of the important functions of corporations is to alienate humans from their activities enough that they can reason about them and do them competently rather than just flail out at every problem with their immediate emotional response.

In other words, humans can think, but not for themselves.

But how about threats? Do threats work?

This is sort of like defining a set that contains its own powerset as an element, not as a subset.

Either you meant to write something else, or I don't yet believe you. I doubt that any infinite set can contain its own power set as either a subset or as an element. The power set of the integers has the cardinality of the reals.

I think that would be legal within ZF if you threw out the Schema of Specification axiom, which we all know is just some bogus hack created by a few mathematicians that were butthurt over Russel's 1337 sploit. It's been nearly 120 years since that particular pony sprouted wings. It's time to move on.

"Move on" as in "allow Russell's paradox as valid rather than paradoxical"? Not me. I'm more of a constructivist.

By the way, what do you think about cleaning data for fun?

About what I think of cleaning toilets for fun.

5223277

That suggests that one of the important functions of corporations is to alienate humans from their activities enough that they can reason about them and do them competently rather than just flail out at every problem with their immediate emotional response.

I think that's actually correct, at least for companies where value prop matters (which, acknowledged, is definitely not all companies). Employees are assigned roles, and they're expected to do things in a way that justifiably satisfies those roles. Along the purchasing chain, a few of those roles involve checking to make sure downstream employees are acting sensibly. This only has to work somewhere along the purchasing chain for it to give reasonably good results.

In other words, humans can think, but not for themselves.

Also broadly correct in the markets I'm familiar with. Advertisements, news, and supply chain convenience matter more than how competitive a product is. All of those things are decided by people that are not the typical consumer.

Ransomware is another market I'm somewhat familiar with. Threats work extremely well there.

Either you meant to write something else, or I don't yet believe you.

Nope, I meant to write exactly that. A set has cardinality 1 when it contains a single element, even if that single element happens to be the set of all reals. The cardinality of its powerset is still 2, which is much smaller than the cardinality of the set of all reals.

About what I think of cleaning toilets for fun.

https://voca.ro/2XLoup304cz *
* This was created using https://fifteen.ai/, which is not the same TTS you'd use for most of the characters.

5223884

https://voca.ro/2XLoup304cz *

That was really good! Except for the artifacting, the shaky sound. I am intrigued.

A set has cardinality 1 when it contains a single element, even if that single element happens to be the set of all reals.

You wrote of a set containing its own powerset as an element. I was thinking that was impossible without sloppy formalisms. I'm tired and old and have a headache, so I'm not going to try to figure it out now. Maybe it is posslble.

5223934
Yeah, I don't think it's possible within ZF. In any theory where it is possible, you'd have to add in other constraints to prevent paradoxes.

I don't know how 15's TTS works, but I can tell you how Cookie's works. Cookie's works with any voice from the show, and it works with both English and phonetic transcriptions. The caveat is that the results are a bit noisier than 15's. There are a few barriers to improvements:

  • Familiarity with sound or generative neural networks for speech data. The ramp-up time on both of these is pretty large, so relatively few people are working on this.
  • Errors in our phonetic transcriptions. I suspect we'd see a large improvement if we had a good way to clean the phonetic transcriptions. Our phonetic dictionary only has one pronunciation for most words, which I suspect is incomplete in most cases. Our forced aligner also gets things wrong sometimes, and we have no way to detect that.
  • Noise in the clips. We have a lot of data with noise in the background, and there are various kinds of noise in the generated clips. Removing that noise would both improve the quality of our dataset and would give us a way to improve generated clip quality independent of neural network improvements.
  • Lack of statistical sophistication. We're sampling clips from the network naively using a sequence of uniform normal distributions (over a latent space). If we had a way to tune the sampling distribution based on feedback, we could probably give people way more control over the generated clips.

Here's a clip that someone painstaking made using Cookie's TTS: https://voca.ro/dHuceoqkKP2

5224227 You write "we" as if you were involved. Are you?

Login or register to comment