moelf

not to mention there's always the possibility of what's roughly described in _End of Science_, at least for physics anyway.

interesting to frame problems as strong/weak link, is kinda useful for thinking about certain things.

- - -

citing Nobel laureate's take on "if I'm working today I won't..." is not too convinsing

    >In my day you could get a faculty job with 0 postdoc papers

can't imagine how that would work today, the world changes why science as a social activity that has costs can be immune to the change?

Too bad while bad music and bad novels only take up some shelf space and make searching a little harder, science projects costs $$$ and so the real Q is how to distribute $, current approach is not that good and gatekeeps some good stuff out, but what's the alternative?

what do we say to people who has the argument of "but the web is already littered with spam blog and SEO stuff"

why people don't compress plain text:

    2.5G filtered_08cdfa755e6d4d89b673d5bd1acee5f6.sampled.jsonl
    834M filtered_08cdfa755e6d4d89b673d5bd1acee5f6.sampled.jsonl.lz4

Think of Stable Diffusion but doesn't need the expensive multi-step diffusion yet still achieve similar result.

but the tokenizer is dataset-driven... it tokenize the most common pattern in your dataset to improve efficiency, so it's 100% about dataset?

except CERN actually does a lot of things backwards in terms of technology (partly due to our unique needs for hardware, but absolutely no excuse for software).

If what CERN is doing has even the remotest real-world application, CERN would have looked much less world-leading than it is today. European treaty is what made CERN viable -- because otherwise it's very hard to fund ultra long-term, almost no return projects.

AI is the opposite of that and you probably don't want to have a CERN: plenty of $ flowing into AI project without government intervention && you don't want the slowness of CERN.

>I’ve been thinking a lot about how AI can reduce some of the world’s worst inequities.

huh, closeAI is certainly not helping at this moment in time.

it's not using the same algorithm right?

the Julia REPL allows you to type unicode with \blah, also if you use `?` help mode, you can give it a unicode sequence and it will tell you how to type it, so it's nice quality of life

    help?> Ψ₂ˣ
    "Ψ₂ˣ" can be typed by \Psi<tab>\_2<tab>\^x<tab>

>These are quite readily available from many other retailers and sometimes for even cheaper than $300.

    https://www.google.com/search?q=Western+Digital+SN740+NVMe+2230

where? Aliexpress and eBay?

>happens at runtime

is not technically true, because that implies a massive slow-down. instead it's more accurate to say behavior-wise it's always equivalent to a dynamic dispatch, but because Julia's Just-Ahead-of-Time compilation, often you eliminate the dynamic dispatch during run time.

we're actively trying to move more end-users of high energy physics (HEP) into Julia https://github.com/JuliaHEP/

All platform is a lie, IIRC a long time ago they actually promised a Linux version, but never realized that.

you're right and it's a pain point for YEARS

https://forum.affinity.serif.com/index.php?/topic/42372-file...

>languages defined by a standard, not by a single implementation.

any of the IronPython, Cython, Jython worked well enough to be continued and works well with the rich CPython ecosystem 100%?

yeah general chicken and egg problem but I guess if we don't collectively take a bit of initiative, we can never have better things!

because that's way too expensive? if you follow every Programming@ on every public instance and nobody is doing the centralized data distribution, you have a all-to-all situation?

Julia packaging scene has the unique advantage that it handles binary dependency (building and versioning) just like a regular Julia package. For example, CUDA.jl will "just work", and you don't need to manually install some specific version of CUDA for TensorFlow etc. which also makes reproducibility more robust. Where in Python the conda.yml doesn't work across OS/architecture, AND often doesn't track binary dependency (despite that's what Conda is designed to do...)

>This is all according to a major new report published on Thursday 20 October 2023 by an international collaboration led by the University of Sussex,

future is here!

it's called JIT, comparing to R, it probably would speed up some moderately complex calculation by 100x to 1000x, but if you don't have the type of workload, feel free to use R -- people should use whatever suits them the best!

Op here, I'd also like to say I blame this on Python's lack of broadcasting syntax, (dot in Matlab and Julia)

Which forces Numpy to be designed in a way to automatically broadcast, guessing what user wants is hard in nested container

Julia

when built-in array type is not performant nor serves as an interface (abstract class), it's as if it doesn't exist. I guess the point is Numpy predates Python's array by a long shot.

Compare and contrast with languages with built-in N-dimensional array for scientific computing.

that's what I'd go for, throw more errors when `__contains__` is called with ambiguity.

but, cough, mature and stable scientific computing foundation?

https://www.programmingfonts.org/#julia-mono FTW

"not bad" is a weird bar to cross for a language that:

    1. decided generics is not needed
    2. went on for a decade
    3. implemented a version of it that is not performance-sensitive
    3.5. $$ by google

I wish the effort is united between dt and rt (and maybe digiKam for management), then we'd have a Blender-level competitor against LightRoom maybe