akavi

It's not. If my work (as a software developer) can be replaced more cheaply by a machine, it should be.

I'm still quite a bit better than SotA models, but I imagine that won't be true in 2034.

Pitch accent in Japanese is deterministic based on the mora that is "accented". While it's true the effect of this accent "spreads" across the entire word, you only need to mark a single mora to know the effects word-wide.

> Reading hiragana is slow (and I've been reading hiragana for a long time)- it's slow, and mentally much harder than reading with kanji.

What's the ratio of hiragana-only text that you read compared to Kanji? And does the hiragana text uses spaces between words? My strong suspicion is "low" and "no", respectively. Familiarity breeds comfort with any writing system, and word breaks are a fabulous ergonomic tool for easing reading.

Tones can be written, and all human spoken communication involves conversational cues, Chinese is not special in this respect.

> You would have to invent a system where pitch accent could be indicated for each word

Really not hard to do. A symbol on the syllable bearing the pitch accent would solve the issue

> And then there are the words with little or no pitch accent difference, only context

What's happened is that effectively a written "shorthand" has emerged that has evolved somewhat separately from how people speak. Losing kanji would mean losing this shorthand, in favor of writing more closely akin to the way people actually speak, but this is how the vast majority of written languages work. Preserving this shorthand seems like thin gruel to justify the complexity of kanji.

Also isn't something that a (geo)sharded postgres DB with the appropriate indexes couldn't handle with aplomb. Number of orders to a given restaurant can't be more than a dozen a minute or so.

Definitely TC

Reminder that a mere 15 years ago, Walmart was the unfair monopolist whose market position rendered competition infeasible.

Also unclear to me why "ecommerce" is a market unto itself that we should be concerned with level of concentration in, as opposed to simply a slice of the broader "retail" market (which is much less concentrated)

I feel like every time COBOL is mentioned we get these stories about crazy high comp for COBOL developers, but anecdotally my aunt worked on COBOL projects in the mid 2010s and was paid a much more modest 45 $/hr. Good money for small town middle America where she lives, but nowhere close to what a decent JS dev can get.

Chances increase dramatically if we cooperate in testifying against the perceived "mastermind", as Ms. Ellison did.

Can you make the Record class private to a module, and only export a static function that constructs them?

(I know very little about Java)

You can get strongly typed "shaped" data without objects[0], even in Java: Records[1].

~Unfortunately, I believe they're mutable (and cannot be made immutable).~ Edit: I was wrong, they're immutable.

[0]: I'm using "object" to mean "data bound to methods", since the concept of aggregate data in general long pre-date OOP (eg, C's structs)

[1]: https://docs.oracle.com/en/java/javase/17/language/records.h...

The other thing is, is if there was demand for load-carrying, rough terrain compatible robots, why aren't legged robots like Boston Dynamics' making a dent in the market?

They're strictly superior than the hoverbot on battery life and carrying capacity, are there other downsides I'm missing? Or is there just plain not a market for this thing?

> We also had initial results from the FLOW trial, suggesting that Ozempic prevents diabetics with kidney disease from needing dialysis.

Between this and the 2nd order decrease in renal disease from the reduction in diabetes and high blood pressure, Ozempic would have to end up having some truly awful long term effects to net out negative in utilitarian terms.

The US spends 130 G$/yr (0.5% of the _entire_ GDP) on dialysis, and that undercounts just how awful being on dialysis is in subjective terms.

> Definitely not visiting that spot ever again.

Sounds like there's already a cap on it.

> I should add that lots of states and some cities have their own minimum wage laws but they are all structured the same.

Rule of thumb, there's _nothing_ in US law where you can say "things vary by state, but they're all the same in way X".

In this case, California has no concept of tip credits. All employees, regardless of whether they receive tips, must be paid at least minimum wage (Currently 16 $/hr)

That's going to be hard to do so long as we continue regarding someone who makes a billion from a new cancer drug as as a price-gouging murderer while regarding someone who makes a billion from starting an adtech company with mere indifference.

Is embodied cognition not at least in part about interactivity? I perform action (emit tokens) and receive feedback (non-self-generated tokens)

Ah, good point!

But the model is downstream of the "patch" tokenization, so the cut-down in resolution (compression) of the image has already occurred prior to the point where the model can direct greater "attention".

I think the synthesis is that I'm proposing a per-pixel tokenization with a transformer block whose purpose is to output information at a compression level "equivalent" to that of the patch tokens (is this what an autoencoder is?), but where the attention vector is a function of the full state of the LLM (ie, inclusive of the text surrounding the image)).

Naïvely, I'd think a layer like this that is agnostic to the LLM state needn't be any more computationally costly than the patching computation (both are big honks of linear algebra?), but idk how expensive the "full context attention" feedback is...

(I apologize to anyone who actually understands transformers for my gratuitous (ab|mis)use of terminology)

Speaking as someone with only a tenuous grasp of how VLMs work, this naïvely feels like a place where the "embodiement" folks might have a point: Humans have the ability to "refine" their perception of an image iteratively, focusing in on areas of interest, while VLMs have to process the entire image at the same level of fidelity.

I'm curious if there'd be a way to emulate this (have the visual tokens be low fidelity at first, but allow the VLM to emit tokens that correspond to "focusing" on a region of the image with greater resolution). I'm not sure if/how it's possible to performantly train a model with "interactive" data like that, though

I think I meant more fusional inflections. 们, 了, 着, 的, etc, don't affect the pronunciation of the root morpheme.

Chinese has fewer inflections than English, but does have 儿 as a nominalizer (and one which is quite productive in northern dialects)

> Why can't we have both multiple staircases and fire-retardant construction?

Because that

1) that costs more money (if you object, "How dare we value money more than human lives?", replace "costs money" with "requires a greater amount of the limited quantity of human labor"), which means fewer homes are built, which means it costs more to purchase/rent a home.

2) Means buildings have to deal with the inherent topological constraints of double-loaded corridors (ie, apartments can't "cross-cut" the building, allowing cross-ventilation/greater number of windows.

As you've demonstrated, you can always construct sum types in typescript with the use of explicit discriminants:

    T = {tag: "L", payload: L} | {tag: "R", payload: R}

The real issue is typescript doesn't have pattern-matching, which make operating on these sum types inelegant

I'm Indian-American, as I stated above, and the list of subcategories are my own ethnic taxonomy.

I'm not sure what part of the Lingayat distinction bothers you. My grandparents would characterize it as "Lingayat" vs "Brahmina"; would you have preferred that framing?

As an Indian-American, "South Asian" is a perfectly reasonable ethnic grouping in the United States. Collectively, that represents ~1% of the US population, too small a grouping to reasonably split out when talking about ethnicity in general terms.

All cultures are fractal. For my own ethnic heritage, for example, I could insist on splitting out:

* South Indian Dravidians from North Indian Arynans

* Kannadigas from other South Indians

* Lingayats from Vaishnavites

But that level of detail (absent special contexts where its relevant) would almost always confuse rather than clarify.

It's any of:

Rust + (garbage collector)

Swift + (cross platform support)

Kotlin + (proper pattern matching)

Unfortunately, none quite hit that central sweet spot, IMO

That's not quite a sufficient rule. Eg, 1 Bahraini Dinar is 1_000 Bahraini Fils.

It's a shame that along with the clause against bills of attainder, banning laws that are intended to solely harm a specific individual or group of individuals, the constitution didn't also ban bills of... abtainder (?) banning laws that are intended to solely _benefit_ a specific individual or groups of individuals.

Would be the best antidote to this sort of thing, but is probably untenable to amend at this point.

I found this hard to believe, but confirmation bias is a hell of a drug, so I ran my own quick blinded experiment: 3 cups of hot water, one of which had a small pinch of saffron swirled around in it, one with a 1/4 tsp of turmeric, one plain.

That it was trivial to tell which was which is an understatement. There's a floral sweetness to saffron that is absolutely unmistakable to me.

Rents came down significantly during 2020 in San Francisco (I paid ~2500 $/mo in Nov 2020 for a place that rented for 3300 $/mo the year before)