Science Proves: virtual meetings are not the same as real ones (they claim it's worse, but... 🤣)
https://www.fastcompany.com/90982118/yale-neuroscientists-brain-research-zoom-meetings

Open AI теперь ищет нового CEO — где можно откликнуться на вакансию, никто не знает?

— Я раньше-то — то на лыжах, то на санках!..
— А теперь что?
— А теперь — то на транках, то на антидепрессантах...

https://statmodeling.stat.columbia.edu/2023/11/17/49679/

A kind of a short master class from Andrew Gelman on improving a paper's title and abstract.

https://www.nature.com/articles/d41586-023-03464-x

Publishers and institutions start using ML systems to detect bullshit papers from "paper-mills". Very conservative estimate flags ~2% of published papers.

Andrew Gelman lists some "highly improbable" (to put it mildly) results from social psychology, and some debunks and analysis:
https://statmodeling.stat.columbia.edu/2023/11/25/some/

Kinda useful when you don't want to fool yourself with noisy data and sloppy analysis. 😊

— А вот он чё такой косматый ходит?
— Он — Хагрид.
— Чего-чего делает?

Да ведь word embeddings — это просто АЦП! 😂

Anime, 4chan, Greg Egan and superpermutations:
https://www.quantamagazine.org/sci-fi-writer-greg-egan-and-anonymous-math-whiz-advance-permutation-problem-20181105/

ДМК Пресс снова готовит что-то любопытное:
https://dmkpress.com/catalog/computer/data/978-5-93700-250-1/
"Алгоритмы и структуры для массивных наборов данных"

Да, "массивные наборы данных" в качестве перевода Big Data немного смущает. Заголовки в духе "Реально-временная аналитика" вместо человеческой "Аналитики в реальном времени" тоже вызывают подозрения.

Тем не менее, в книге разбираются распространённые и важные структуры данных
1. на основе хеширования: старые-добрые хеш-таблицы, фильтры Блума, HyperLogLog;
2. (приближённые/вероятностные) потоковые алгоритмы;
3. разные варианты B-деревьев и LSM-деревья;
и некоторые другие.

В деталях не разбирался, но что-то стоящее должно быть можно из неё извлечь.

https://statmodeling.stat.columbia.edu/2023/12/06/exploring-pre-registration-for-predictive-modeling/

Jessica Hullman provides a (long) summary of a paper on preregistration for predictive modeling aka machine learning by Jake Hofman, Angelos Chatzimparmpas, Amit Sharma, Duncan Watts, and Jessica Hullman.

Simply put, they suggest (informal) preregistration as (another) tool for honest researchers to avoid fooling themselves and overfit to the test data hacking too much hyperparameters while comparing to undertuned baselines (among other mistakes).

At any rate they provide two short sets of questions that anyone should answer to themselves before starting real experiments and measurements.

https://buttondown.email/hillelwayne/archive/when-would-you-ever-want-bubblesort/

The Bubblesort algorithm is mainly famous for being slow, but it possesses other curious theoretical properties too. In particular it can be viewed as an "anytime" or incremental algorithm.

Apparently it can be useful in Game Dev for some particular tasks.

— Как вам удалось так похудеть? В чём ваш секрет?
— Перестал жрать говно в Твиттере...

Видишь в итальянской пасте укроп с петрушкой, и невольно думаешь "откуда тут африканские названия?!"

I recently read a prescient short fiction story "See no evil" by John R. Pierce, where he anticipated in great detail the blueprint for most modern AI startups, already 55 years ago!

Unfortunately the only reference to it I could find is this one: https://www.isfdb.org/cgi-bin/title.cgi?68334

http://aria.org.uk/wp-content/uploads/2023/10/ARIA-Mathematics-and-modelling-are-the-keys-we-need-to-safely-unlock-transformative-AI-v01.pdf
"Mathematics and modelling are the keys we need to safely unlock transformative AI"

David “davidad” Dalrymple outlines the case for mathematical "world models" and formal verification as a fruitful or even necessary approach to AI alignment and safety.

The 3-page PDF features very nice layout and a long list of relevant publications (including deep technical ones) for further study.

But the best part: ARIA (aka Advanced Research and Invention Agency) is going to provide funding for research in this area, and starts to collect (preliminary) research projects proposals.

If you have a question and think to yourself "I'll ask a LLM about it!" better ask two of them and compare the answers providing your feedback to the community:

https://chat.lmsys.org/

😊

— Всё, чтобы вы отъебались улыбались!

Моему коту в этой жизни нужно только одно — укрытие от жизни.

https://www.stochasticlifestyle.com/chatgpt-performs-better-on-julia-than-python-and-r-for-large-language-model-llm-code-generation-why/

Evidently ChatGPT 3.5 "understands" #Julia significantly better than other languages including Python and JavaScript, not even mentioning Go, C and C++: https://arxiv.org/abs/2308.04477

Chris Rackauckas gives some points as to why it's not that surprising from the perspective of teaching novices. And also speculates that large volume of so-so tutorials and examples for very popular languages might hurt LLMs' learning.

He also advertises his diffeqpy library connecting optimized solvers implemented in Julia (including code generation for GPUs) to #Python #machinelearning libraries, and points to some curious papers with impressive benchmarks. 😊

Хотел сократить "психологический уют" до "псих. уют", но получилось как-то неблагозвучно... 😏