pseudonom-

Dhall is the FP config language you're thinking of, I think.

Funny, I did almost the exact same thing: https://github.com/colehaus/hammock-public. Though I project to 3D and then put them in an interactive 3D plot. The other fun little thing the interactive plotting enables is stepping through a variety of clustering granularities.

OpenAI was formed as a nonprofit with a specific charter ("OpenAI’s mission is to ensure that artificial general intelligence (AGI)—by which we mean highly autonomous systems that outperform humans at most economically valuable work—benefits all of humanity. We will attempt to directly build safe and beneficial AGI, but will also consider our mission fulfilled if our work aids others to achieve this outcome.") and the capped-profit entity under which daily operations occur formed years later with the claim that it was instrumentally useful for pursuing that charter. The capped-profit entity remains a subsidiary of the nonprofit. The board in the dispute is the board of the overseeing nonprofit.

So there are many particulars that mean pattern matching to a standard board dispute will lose something. I think it's likely many of the primary actors have, at various times, had non-strictly-pecuniary motives. That one side won doesn't mean that the other side was always a farce.

Are you aware of the history and governance structure of OpenAI?

Unless I'm misunderstanding something, this is precisely why I don't use Home Manager. I've literally never had my NixOS setup break over the course of many years.

https://arxiv.org/abs/2306.11644 is along these lines.

Probably quotes like:

"It must be familiar, roughly C-like. Programmers working at Google are early in their careers and are most familiar with procedural languages, particularly from the C family. The need to get programmers productive quickly in a new language means that the language cannot be too radical."

And not including sum types despite having a sum-type-shaped hole in the language (`if err != nil`).

And some of the discussion about "why no generics" seemed kind of divorced from existing PL knowledge on the topic.

Tim Dettmers recently (https://www.manifold1.com/episodes/ai-on-your-phone-tim-dett...):

"But what we found with these neural networks is, if you use 32 bits, they're just fine. And then you use 16 bits, and they're just fine. And then with eight bits, you need to use a couple of tricks and then it's just fine.

And now we find if you can go to four bits, and for some networks, that's much easier. For some networks, it's much more difficult, but then you need a couple more tricks. And so it seems they're much more robust."

There are other mechanisms for dealing with vanishing and exploding gradients. I (maybe wrongly?) think of batch normalization as being most distinctively about fighting internal covariate shift: https://machinelearning.wtf/terms/internal-covariate-shift/

(Do bear or typeguard allow you to do this using decorators?)

You can push some of this directly into Python type annotations thanks to https://peps.python.org/pep-0646/.

e.g.

  @overload
  def mean(a: ndarray[float, Dim1, *Shape], axis: Literal[0]) -> ndarray[float, *Shape]: ...
  @overload
  def mean(a: ndarray[float, Dim1, Dim2, *Shape], axis: Literal[1]) -> ndarray[float, Dim1, *Shape]: ...

There's some discussion in section 3.2 of https://arxiv.org/pdf/1910.10683.pdf

Ie even lower than my guess of 90% above.

These are not comparable numbers. You're comparing "fraction of people" vs "fraction of outcomes". Presumably an eye-roller assigns ~0 probability to "extremely bad" outcomes (or has a shockingly cavalier attitude toward medium-small probabilities of catastrophe).

Ahh, I was focusing on the "local state" vs "global state" bit. Haskell has ways to address the expression problem: e.g. http://www.cs.ru.nl/~W.Swierstra/Publications/DataTypesALaCa... and http://okmij.org/ftp/tagless-final/course/lecture.pdf

To the extent that I understand your complaint, there is a functional architecture that alleviates this concern: arrowized functional reactive programming. https://blog.jle.im/entry/intro-to-machines-arrows-part-1-st... talks about it.

This can be fairly smoothly achieved already: https://dev.to/gcanti/functional-design-smart-constructors-1...

Chronic hunger in (at least) the US appears to be essentially non-existent for children: https://blog.givewell.org/2009/11/26/hunger-here-vs-hunger-t...

There are serious attempts to estimate the externalities that arise from automobile use in general: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.661...

Adjusting those estimates (only) for inflation, they work out to about $2.14 per gallon in driving externalities.

"Carbon taxes are regressive" is too simple, and I'd argue wrong, for a variety of reasons:

- Some experts just flatly deny it: "low-expenditure households devote a smaller share of their budget to gasoline than do their counterparts in the middle of the expenditure distribution" --- https://dspace.mit.edu/bitstream/handle/1721.1/63747/isgasol...

- Carbon taxes are often less regressive than alternative sources of revenue e.g. sales taxes

- Depends on how the revenue is spent. If, for example, it's used to expand the Earned Income Tax Credit in the US, the overall tax and spend pattern would be progressive.

- The impacts of excess carbon are also "regressive" in that global warming affects poor people more severely. This means that the net effect of a regressive tax which reduces a problem with an even more regressive incidence could be progressive.

Does anyone have any impressions on how this compares to CLRS? (https://en.wikipedia.org/wiki/Introduction_to_Algorithms)

~30% of her students, from either school, would not eat at all over the weekend. No, not that they would eat Cheetos or drink Coke. They plain-jane went hungry.

Can you help me reconcile that with the data from the USDA as described here? https://blog.givewell.org/2009/11/26/hunger-here-vs-hunger-t...

They describe children in 0.3% of households as skipping a meal almost every month and the number of children that did not eat for at least one day in almost every month is too low to measure.

You can use switch statements for a pattern matching analogue. See the "Exhaustiveness Checking" heading at https://www.typescriptlang.org/docs/handbook/advanced-types.....

I was able to find one reference to this spider in the context of mosquito elimination:

Regarding ecological issues surrounding the release of sterile TMs, of particular importance is that a reduction in the number of Anopheles is unlikely to have a negative impact on local food chains. According to (Pennetier et al. 2010) there are no birds, fish or other insects that feed exclusively on Anopheles mosquitoes, the only known exception being Evarcha culicivora, an East African jumping spider that feeds indirectly on vertebrate blood by selecting blood-carrying female mosquitoes as preferred prey.

from https://www.researchgate.net/profile/Alicia_Coupe/publicatio...

From the article:

"That might sound disturbing from an ecological perspective, but mosquitoes don’t make up a significant portion of any known predator’s diet; there’s just not a lot of meat there. “So far there’s no evidence really that seems to show that Anopheles gambiae is a key species in the ecosystem,” Jonathan Kayondo, a senior research officer at the Uganda Virus Research Institute and a member of Target Malaria’s scientific team, told Fong. “There’s nothing that exclusively feeds on it. So I’m finding it hard to see how that would collapse the ecosystem, because that’s the fear most people have.”

I suspect (based on stuff like https://arxiv.org/abs/1510.08419) that you'd have had more luck with property tests instead of unit tests.

In some languages, you can ensure that no `.toString()` equivalent is available. Then, your ID is truly opaque and can only be used in the prescribed ways.

Not sure what GP is thinking of, but here's an example of embedding tools for reasoning about performance in Idris: https://www.youtube.com/watch?v=4i7KrG1Afbk&feature=youtu.be...

When the compiler handles this, it is usually called "automatic deforestation". One technique for this (that Haskell's compiler implements) is "stream fusion". A brief overview of stream fusion is given at https://stackoverflow.com/questions/578063/what-is-haskells-... and a lengthier paper is at http://fun.cs.tufts.edu/stream-fusion.pdf.

e.g. https://github.com/gabesoft/nix-conf/blob/master/.nixpkgs/co...

Though there are a couple more declarative ways to handle user package installs.

An EDSL and a domain-specific API are not the same thing. For example, one typically expects the description of a program's logic and its execution to be separate in a DSL (i.e. building up expressions from the syntax made available in the EDSL and then later applying an interpreter (possibly multiple interpreters) to these expressions) while this is not generally true for domain-specific APIs.