https://www.fused.io/workbench

Huh, pretty nice and impressive. It makes possible writing, visualizing and debugging pretty sophisticated GIS functions in Python. Look at the catalog.

An illustration of my training goals and largely my training program... 😂

https://x.com/davidtolnay/status/1883906113428676938

The future is (quietly and unevenly) here: an "AI-maintained" library. In Rust in this case. And yep, it exhibits all the traits of an "AI-maintained" you can expect at present day and age. Including segfaults. In Rust. Yeah, that's the true staple of advanced AI.

But I blame YAML. It's such a pile of smelling garbage no sane human would willingly maintain a library to handle it. And those who attempt inevitably go insane. Only AI stupid enough yet to dive into.

https://corticallabs.com/cl1.html

A brain in a vat, but unironically... 🥴

I'd like to attract your attention to one funny observation.

The seminal "Gopher paper" of 2021 (so old!) among great many things investigated (sections G.3.3 and G.3.4) initialization of a new LLM weights by weights on a existing pre-trained LLM.

Basically, this is the direct knowledge transfer from one "mind" to another, like SciFi writers dreamed about last 70 years or so.

There are other methods of pretty direct "knowledge transfer" of course: model "merging", "patching", etc. It's so mundane for ML nobody even trying to make a fuss...

Utrecht University seems to be very serious about Differential and Probabilistic Programming — they have three (3) PhD positions open:

https://www.uu.nl/en/organisation/working-at-utrecht-university/jobs/phd-position-in-programming-semantics-and-category-theory

https://www.uu.nl/en/organisation/working-at-utrecht-university/jobs/phd-position-in-dsls-for-high-performance-computing

https://www.uu.nl/en/organisation/working-at-utrecht-university/jobs/phd-position-in-probabilistic-and-differential-algorithms

Category Theory, HPC DSLs and Algorithms... In Utrecht, The Netherlands... 😊

Application deadline is 21 April.

— место в жизни
— место в жизе
— место в жиже

https://designer.microsoft.com/

A quite little thing from Microsoft. Reminds me of postcards and booklets templates in PowerPoint. At some point I even employed that to make a couple of postcards... Now it uses AI of course. 😊

https://gemini.google/overview/canvas/

The Gemini Canvas landing promises (among other things) to make you interactive (as in "HTML/JS app") quizzes for your studies. Coincidentally I was flipping through an extended lecture notes/lightweight textbook (80+ pages, pretty comprehensive thing), which has many multiple-choice tests for self-assessment. In particular, it has an entry test for pre-course assessment.

So I figured, I'd upload the PDF into Gemini and ask it to make an interactive quiz for the pre-test specifically. And it did!

Though, at first it simply extracted the test from the PDF into a Markdown document, but at least it was the right test, nothing added, nothing subtracted. And after I asked about an interactive app it generated HTML/JS for the quiz which Canvas can show inline.

Funny enough, one of the "correct answers" in the PDF was pointing to the fifth answer for a four-answer question, and Gemini corrected it to the right one. 😁

Yeah, these LLM become not only fun but also useful and usable. 😊

https://www.youtube.com/watch?v=Umy7hsw7pF0

A beautiful video on New Zealand's nature and its history.

Overpass API + YOLOv11 + SAM2 + Colab + HuggingFace

"Statements dreamed up by the utterly deranged"

I'm frightened by the fact I know what that means and how that came to be...

https://www.youtube.com/watch?v=G6R7UOFx1bw

Really interesting, much more than I anticipated. Some deep ideas about both past and future.

https://www.bigfootmap.com/

If you spot a Bigfoot/Yeti/Sasquatch/etc. you can report the encounter here. Otherwise you can analyse the data, but I have no idea what you can make of it. 😁

Судя по тому как программисты реагируют на бенчмарки разных LLM, мы понимаем, что свой код они не бенчмаркают, и вообще никогда ничего не бенчмаркали... 😏

https://www.youtube.com/watch?v=YdOXS_9_P4U

Damn the Ancient Greeks! They figured out so much from literally nothing: no instruments, no data, the most basic math. By the sheer power of reason. And a lot of patience it seems.

https://www.youtube.com/watch?v=LI7koSi5J6w

More hype!!! 😃

So soon already!

Apparently, Google Gemini "thinks" the https://www.geminidata.com/ is somehow related to Google Gemini... 😏

Sweet cats are made of fur...

Turn your unwieldy Python code into beautiful math formulae and render them right in a Jupyter Notebook:
https://github.com/google/latexify_py
😂

TIL: Applicative instance for functions comprises K and S combinators:

instance Applicative ((→) a) where
  pure b = λa → b
  h ⊛ g  = λa → (h a) (g a)

😳