I find this article deeply thought-provoking and at times disturbing, for multiple reasons. I am not yet sure what to make out of it, as many parts are very alien to my "feeling of what is right and wrong", while the others are pending more debate, so for now I am just leaving it here for further processing.

https://www.timdavis.com/blog/probabilistic-engineering-and-the-24-7-employee

#AI #LLM #vibe #danger #probabilistic

"Probabilistic Machine Learning", Murphy (2012, 2022, 2023)

MIT licensed PDF drafts published by the author are at https://github.com/probml/pml-book.

As the #PML name suggests, these books present #ML from the #probabilistic perspective—that is, ML presented the right way.

The 2022 and 2023 books (An Introduction and Advanced Topics) are the updated, expanded version of the original 2012 book. The new books provide a comprehensive coverage of ML as it was, circa 2021.

These books are not only comprehensive in the coverage of ML, they are also self-contained in that they provide all requisite mathematical background, without slapping the reader across the face with fine turns of phrases. They are written in an authentic, genuine, heartfelt style, a bit of a rarity amongst AI/ML publications, today. They are, in my view, the most effective self-study guides for upper-level undergraduate students, beginning graduate students, and experienced IT practitioners who aims to study the concepts in depth—they who are dissatisfied with just making shallow API calls.

The older book uses MATLAB, and the newer ones use Python. Naturally! But using probabilistic DSLs—Figaro, Church, Anglican, WebPPL, etc.—with these books maybe even more effective.

#Probabilistic #Computer

https://arxiv.org/html/2601.13229v1

#probabilistic #programming languages

https://github.com/automata/awesome-probabilistic-programming

#Introduction I've been around the Fediverse for a few years, but at some point I ended up abandoning this account 🙈

So... let’s try again.

I teach computer science at VRAIN-UPV (Universitat Politècnica de València, Spain). My research interests include things like explainable and symbolic #AI, #probabilistic #logic programming/term rewriting, #causality, #concurrency, #programming languages, #reversible computing, program #verification, and #debugging.

Outside of work, I'm into #photography and I'm a big #sci-fi fan (books, movies, and TV shows). I also enjoy traveling, cooking, and getting outside for a walk or a run.

Languages: Spanish (native), Catalan, English, and some Italian.

fly51fly (@fly51fly)

하버드 연구진(M. Zhao 등)은 LLM(대형 언어 모델)이 통계적 분포에서 무작위 수를 생성하는 데 큰 어려움을 보인다는 실험 결과를 보고합니다. LLM이 난수 생성과 확률적 샘플링에서 체계적 편향을 보이며 시뮬레이션·확률적 추론 응용에서 신뢰성 문제를 야기할 수 있음을 지적합니다 (arXiv:2601.05414).

https://x.com/fly51fly/status/2013733255078650246

#llm #randomness #evaluation #probabilistic

The CEO of this company is total Silicon Valley right wing -pilled, but his tech is fucking solid.

This is the right way to do #AI.. Fuck GPU’s, design the hardware for probabilistic computing from scratch.

Result: 10000x less power required.

This opens the way to me running AI some day: On my own device, without a network connection.

https://youtube.com/watch?v=Y28JQzS6TlE

#extropic #tsu #thermodynamic #probabilistic #computing