curtisf

I'm pretty confused by most of the article.

The focus is placed on "AI Literacy", but it seems to use this to just mean 'volume of AI use'. The discussion of the Netflix case study is extra perplexing, since the summary here admits they didn't find any actual productivity improvement, just that only a few hours of "training" could induce on the order of $50/person/day on tokens.

That seems... the opposite of literacy?

The H1B visa is explicitly designed for high skill (high paying) jobs which companies have (supposedly) demonstrated they cannot find enough citizen workers.

There are much simpler mechanisms to making that would make the enforcement mechanism more effective without destroying the economy, like prioritizing them by salary instead of randomly.

You could also just have a more proactive government which punishes businesses for abusing the visa category.

"Immigrants taking good jobs" isn't an immigrant problem, it's a big-business problem

This has everything to do with art direction, and not technology.

The real world has much more light bouncing fidelity than even modern games. There are still dark things we can't see.

Physically based rendering should be exactly the opposite of what the article is complaining about: it gives you the "correct" way to communicate how light is moving through a space. So the player and the game designer should be able to communicate much more easily, and the artists should be able to focus on actually communicating what they need to, instead of tweaking non physical phong and ambient lighting parameters

Most of the big consumer VPNs include "privacy" with an implication of anonymity in their marketing, so it shouldn't really be surprising

It does seem like, especially for the supposed political goals of the recent governments, there's things to watch out for.

But despite denying it ("This isn’t hyperbole"), it seems like most of this article is, in fact, hyperbole.

The main (incredibly grating) assumption is that government is intently destructive and non-productive:

But only one of them leads to capital accumulation and becomes an engine of generational wealth growth.

...It is unavoidable that the government appropriates and ultimately compromises and destroys a lot of productive capital. Well and good, provided the economy can regenerate it at least as fast as it is harvested.

Government is only unproductive to the extent the people want it to be. Governments fund education, nutrition, and support immigration, i.e., produce essentially _all_ of the available _human capital_. Research and development produce enormous values both for society and the economy (the internet was designed, tested, built, and expanded by a government!). Governments foot the bill for enormous utility and trandportion subsidies to enable entire economies to exist at all.

The only difference between a government agency and a corporation is the profit _motive_. The government can do exactly the same things as a corporation would, and in fact, that's why it's relatively common for governments to charter purpose made corporations to do certain things.

The inefficiency and ineffectiveness of government is mostly a contrivance of people who _want_ government to be ineffective and inefficient -- mostly as a rhetorical bludgeon to further a "small government" agenda (i.e., moving as much economic power out of the hands of agencies chartered to do something _other than_ maximize shareholder value)

Sure. A reasonable model for incoming requests within a short window of time is as a "Poisson process", which means the expected number of incoming requests within any interval is proportional to the length of that interval.

The parameter of that distribution is the expected (aka average) rate. If the intervals are time intervals, then the proper units of the parameter are events/second

It's basically using the "-" embedded in the definition of the eml operator.

Table 4 shows the "size" of the operators when fully expanded to "eml" applications, which is quite large for +, -, ×, and /.

Here's one approach which agrees with the minimum sizes they present:

        eml(x, y             ) = exp(x) − ln(y) # 1 + x + y
        eml(x, 1             ) = exp(x)         # 2 + x
        eml(1, y             ) = e - ln(y)      # 2 + y
        eml(1, exp(e - ln(y))) = ln(y)          # 6 + y; construction from eq (5)
                         ln(1) = 0              # 7

              eml(ln x, exp y) = x - y          # 9 + x + y

                   x - (0 - y) = x + y          # 25 + {x} + {y}

"I would rather read the prompt"

https://claytonwramsey.com/blog/prompt/

discussion: https://news.ycombinator.com/item?id=43888803

All of the output beyond the prompt contains, definitionally, essentially no useful information. Unless it's being used to translate from one human language to another, you're wasting your reader's time and energy in exchange for you own. If you have useful ideas, share them, and if you believe in the age of LLMs, be less afraid of them being unpolished and simply ask you readers to rely on their preferred tools to piece through it.

"Consensus" in this post refers to the "consensus problem", which is a fundamental and well-known problem in distributed systems.

It's not about political consensus.

However, the paper that introduced it and proved it possible, Lamport's "The Part Time Parliament", used an involved (and often cited as confusing) "Parliament" metaphor for computers in a distributed system

"Consensus" in distributed systems need not be limited to majorities; it really just requires no "split brain" is possible. For example, "consensus" is achieved by making one server the leader, and giving other servers no say. A majority is just the 'quorum' which remains available with that largest number of unavailable peers possible.

It could also be useful in low doses to supplement, for example, a seasonal vaccine in a year where they are especially unsure about prevalent strains, or where their predictions were already proved wrong early in the flu season

For optional types, 0 is decoded as the default value of the underlying type (e.g. string? decodes 0 as "", not null).

In the "dense JSON" format, isn't representing removed/absent struct fields with `0` and not `null` backwards incompatible?

If you remove or are unaware of a `int32?` field, old consumers will suddenly think the value is present as a "default" value rather than absent

Isn't it?

You can have Dependabot enabled, but turn off automatic PRs. You can then manually generate a PR for an auto-fixable issue if you want, or just do the fixes yourself and watch the issue number shrink.

Conscription is horribly inapt metaphor for mandatory inoculation.

Banning the playing of third-party Russian roulette, where you hold a mostly unloaded gun to the head of your neighbors, coworkers, and service staff, actually more accurately represents the risks involved to both yourself and the public, and importantly to the personal tax and effort required.

A lien on the property? Although almost all jurisdictions already have property taxes, so it hasn't been an insurmountable problem so far

This could be stated much more succinctly using Jobs to be Done (which is referenced in the first few paragraphs):

Your customers don't want to do stuff with AI.

They want to do stuff faster, better, cheaper, and more easily. (JtbD claims you need to be at least 15% better or 15% cheaper than the competition -- so if we're talking "AI", the classical ML or manual human alternative)

If the LLM you're trying to package can't actually solve the problem, obviously no one will buy it because _using AI_ OBVIOUSLY isn't anyone's _job-to-be-done_

It could, but under the current system, candidates who are affiliated with major parties (i.e., essentially everyone who ends up winning an election) already need to win the support of their party, and the process for this is generally opaque and largely controlled by often less-moderate insiders

Also, having viable third party choices puts more pressure on larger parties to field more widely palatable candidates, or risk losing their majorities

I do not understand what this could mean.

There are clear formalizations of concepts like Consistency in distributed systems, and there are algorithms that correctly achieve Consensus.

What does it mean to formalize the "Single Source of Truth" principle, which is a guiding principle and not a predictive law?

They might not _notice_ but that doesn't mean it's not affecting their ability to use their computer smoothly.

With computers such a huge part of almost everyone's lives now, it's a travesty for one of the largest companies in the world to inflict something so subpar on so many old-style

My favorite concrete example of this is the textually beautiful "xd" crossword puzzle format.

Interesting video story: https://youtu.be/9aHfK8EUIzg (2016)

Data site: https://xd.saul.pw/data

Using words written by other people without disclosure has always been frowned upon. It's called plagiarism.

Plagiarism is bad for a lot of reasons, all of which also apply to the undisclosed use of generative AI.

Livestock emits between 10% to 20% of global greenhouse gases (in carbon equivalent/100y-GWP) [1]

In contrast, all data centers (not just AI) currently use less than 1.5% of all electricity, making up less than 0.3% of global emissions [2]. Although recent increases in data center electricity usage is lamentable, even in the short term future, much of this can and more importantly _will_ be low-carbon energy, and the ratio should continue to improve with time.

A 1% reduction in livestock emissions is therefore about the same as a 50% reduction in data center emissions.

[1] https://thebreakthrough.org/issues/food-agriculture-environm...

[2] https://www.carbon-direct.com/insights/understanding-the-car...

By replacing (some) farmed meat with farmed fungi protein.

Although it's theoretically possible for a disease to infect both fungus and animals, because the biology is so different, the risk is greatly, greatly reduced.

In addition, it may be possible to reduce the use of treatments such as antibiotics which, in their currently mass application to farmed animals, could directly lead to the development of antibiotic resistant in diseases which affect humans and animals.

Formal verification is explicitly NOT testing.

It is a method where a computer verifies a proof that the program adheres to its specification for _all_ inputs (subject to whatever limitations the particular method has).

Types are the simplest kind of formal verification, and with sufficiently advanced dependent type-systems, can be used to prove that programs obey arbitrarily complex specifications. However, this can be extremely laborious and requires significantly different skills than normal programming, so it is very rarely done in industry

I think this discussion dismisses the "physics" of writing code, which rewards laziness.

Effects make _the right thing to do_ (proper sandboxing, testability, assertions, ...) the _easiest_ thing to do.

Build scripts aren't sandboxed because sandboxing bash functions is nigh impossible -- not because people don't want to.

The discussion on assertions is especially confusing, because that is exactly what effect systems excel at. The effect of an assertion would be Assert, and you can choose to handle it however you want, at a higher level. If you want to crash, handle Assert in main by Exit(1)ing. If you want to reject the request but keep the server alive, handle by SetResponse(500)!; CloseRequest()!. If you want to ignore it and trundle on, return to the point of the assertions continuation.

The United States is a country that doesn't require any form of identification to have a phone number. You can buy a "burner phone" at any Walmart, with prepaid minutes and no contract, in cash.

... How about fixing basic things like the cursor position in code blocks, and being able to select text on mobile, instead of unnecessary "AI Agents"?

Or is the frontend now supposedly obsolete since all the work will be done by "AI"?

This is true to the extent that you actually _use_ all of the features of a dependency.

You only need to rewrite what you use, which for many (probably most) libraries will be 1% or less of it

"The Walker Montgomery Protecting Children Online Act" ("HB 1126") was passed essentially unanimously by the Mississippi state legislature.

"number of bytes" is dependent on the text encoding.

UTF-8 code units _are_ bytes, which is one of the things that makes UTF-8 very nice and why it has won

What is the actual model that takes 15 seconds to compute?

If I understand the setting, you are estimating the demand curve for a given price... And there are only 40 such curves to compute.

Surely each curve is fit with only a few parameters, probably fewer than five. (I think for small price ranges the demand curve is usually approximated as something trivial like y=mx+b or y=a/(x+b)+c)

Why does evaluating the model at a particular price need to take milliseconds, yet alone 15 seconds?