Hakan Kjellerstrand

It proposes a research program using #AI to tackle the longstanding problem of many free parameters of Standard Model.

Our (Sergei Chekanov and me) physics paper "Discovering the Underlying Analytic Structure Within Standard Model Constants Using Artificial Intelligence" has now been accepted and published in the Particles journal: www.mdpi.com/2571-712X/8/... .

The Picat PPL page includes documentation of the features of Picat PPL as well as over 550 probabilistic programming models. (In total, it's now over 115 000 lines, including comments and output of the model.)

5/5

The dedicated page for Picat PPL is hakank.org/picat/ppl .
and at GitHub: github.com/hakank/hakan... .

Also, the published files can be downloaded from the Zip file: www.hakank.org/picat/ppl/al... . This file will be updated when changes are done.

4/n

It has two features:
- support for doing probabilistic programming modelling using a fairly high level syntax (such as Gamble and WebPPL)
- support of over 40 different probability distributions (or about 80 depending how one count) with exact calculations, upto Picat's precision.

3/n

It is designed for exploring uncertainty, solving probability puzzles, and experimenting with small-scale probabilistic models - rather than performing heavy Bayesian data analysis.

2/n

Picat PPL (hakank.org/picat/ppl) - Probabilistic Programming Light is is a lightweight probabilistic programming framework implemented entirely in Picat (picat-lang.org).

1/n

Discovering the underlying analytic structure within Standard Model constants using artificial intelligence This paper presents a search for underlying analytic structures among the fundamental parameters of the Standard Model (SM) using symbolic regression and genetic programming. We identify the simplest ...

This is based (and extended) on the findings and data from our first physics paper (preprint): S. V. Chekanov, H. Kjellerstrand: "Discovering the underlying analytic structure within Standard Model constants using artificial intelligence"(arxiv.org/abs/2507.00225 )

5/5

but instead follow a hierarchical structure within a high-dimensional functional space. The found analytic solution depends on only two input parameters, representing the simplest mathematical model that could provide a foundation for developing a future theoretical framework to address the SM.
4/n

We outline the core principles of our method and identify the simplest analytic relationships among SM parameters. Our results suggest that the SM parameters associated with quark and boson masses are not randomly distributed,
3/n

Abstract: This paper presents an approach to reducing the number of fundamental parameters in the Standard Model (SM) using genetic programming, a machine learning technique based on evolutionary algorithms.
2/n

Evidence of Relationships Among Fundamental Constants of the Standard Model This paper presents an approach to reducing the number of fundamental parameters in the Standard Model (SM) using genetic programming, a machine learning technique based on evolutionary algorithms. We...

Our second physics paper (preprint) is now available at arXiv: S. V. Chekanov (@svchekanov), H. Kjellerstrand (@hakankj): "Evidence of Relationships Among Fundamental Constants of the Standard Model" (arxiv.org/abs/2509.07713)

Abstract: See next comment.
1/n

GitHub - hakank/fzn_picat_sat Contribute to hakank/fzn_picat_sat development by creating an account on GitHub.

Here's the GitHub repo of the source code of the #Picat SAT FlatZinc solver fzn_picat_sat used in the MiniZinc Challenge 2025: github.com/hakank/fzn_p... .

For the result of the #MiniZinc Challenge 2025: www.minizinc.org/challenge/20... . (As mentioned earlier, the Picat SAT solver got two Silver.)

Picat Through the Lens of Advent of Code Picat is a logic-based, multi-paradigm programming language that integrates features from logic, functional, constraint, and imperative programming paradigms. This paper presents solutions to several ...

For some examples of how to use the bv module:
* Section "4 Day 17: Chronospatial Computer" of our ICLP'25 paper Picat Through the Lens of Advent of Code (arxiv.org/abs/2507.11731)

* My examples at hakank.org/picat/#bv
GitHub commit of these models: github.com/hakank/hakan...

Picat: Examples of the new bv module in v3.9 · hakank/hakank@ca4a0af

#Picat version 3.9 has been released: picat-lang.org
Updates: picat-lang.org/updates.txt

It contains improvements of the SAT solver and - which is really cool - a bit vector constraint module which makes it possible to use arbitrary sized domains (> 2**56) in constraint models.

1/3

Results from the #MiniZinc Challenge 2025: minizinc.org/challenge/20...
As earlier years, Google OR-tools got all Gold medals. We (#Picat team) got two Silver (again). Choco: 1 Silver, 1, Bronze; Yuck: Silver; SICStus Prolog, Pumpkin, iZplus, and Atlantis: 1 Bronze.

Congrats to all!

Krazydad Free Puzzles Free printable and interactive sudoku, mazes, and logic puzzles.

Puzzle creator Jim Bumgardner (aka krazydad, krazydad.com) is "vide coding" a puzzle solver/generator using Constraint Programming (w Google OR-tools): "Lets make minesweeper sudoku part 2" (www.youtube.com/watch?v=aNdT... ).

His other YT videos: www.youtube.com/@jbum

Our paper “Discovering the underlying analytic structure within Standard Model constants using artificial intelligence” by Sergey Chekanov (anl.gov/profile/serg...) and myself (hakank.org) is at arXiv: arxiv.org/abs/2507.00225, in the category hep-ph.

It is a Python program but since it's a beta (about v0.000006), I'll not publish it.

If you're interested, please DM (or mail me).

Earlier today I read about Warp Terminal (warp.dev) and it looked neat. However, it uses a remote AI for the query.

Since I rather want to use my local LLM (via LM Studio), ChatGPT (4o) and I did Terminal Assistant instead. Examples: hakank.org/terminal_ass...

Cashier's Change Puzzle Search Analysis of a Coin Change Puzzle Involving a Cashier 1. Introduction The inquiry concerns a specific logic puzzle presented as a dialogue between a customer and a cashier named Robert. In this scenari...

Just for fun, I also asked Google Gemini to do a Deep Research on this problem: docs.google.com/document/d/1...)
and NotebookLM Deep Dive podcast: notebooklm.google.com/notebook/280...

Note that the paper also discusses variants to this puzzle.

A Puzzle That Will Test Your Thinking Pocket Full of Change, But No Change to Give?

Here's my (new) Litmus test of LLMs: The No change problem (from medium.com/puzzle-spher... )

#Picat program: www.hakank.org/picat/no_cha... showing two constraint approaches (and a brute force variant).

LLMs might get one of the two solutions, but often only after a lot of corrections.

LLMs and Einstellung Effect Research Large Language Models and the Cognitive Bias of the Einstellung Effect I. Introduction: Defining Large Language Models and the Einstellung Effect in the Context of AI Large language models (LLMs) repr...

And I did another Research with Google AI Studio but restricted it to be only on Einstellung Effect. Here's the generated document: docs.google.com/document/d/1...
And then using Google's NotebookLM to create a neat 2 person discussion about the document: notebooklm.google.com/notebook/ef8...

ChatGPT - ILP Rule Induction Shared via ChatGPT

In March 2023, I tested GPT4 with a couple of traditional ILP (Inductive Logic Programming) problems, and some other problems: hakank.org/gpt/

Yesterday, I checked how GPT 4o would do: chatgpt.com/share/67d479...

In summary: quite much better.

The other day I happen to read about Einstellung effect (en.wikipedia.org/wiki/Einstel...) and thought that it seems to be one of the problems that are in LLMs.

So I asked GPT 4.5 + Deep Research for a report on LLM Cognitive Biases Research. Here's the report: chatgpt.com/share/67d47b...

#Picat version 3.8 is available: picat-lang.org/download.html

picat-lang.org/download.html
Updates: picat-lang.org/updates.txt
E.g. new if/then/else syntax variant, improved error handling, debug flag (-d) at command line, new between/4, and some other bug fixes.

Unfortunately. Pop-11 is a fun language and the Poplog distribution contains a huge number of examples of GOFAI (Good Old Fashion AI) that nobody seems to care about nowadays...

Here's my Poplog page:
www.hakank.org/poplog/

Nope, I learned about it some weeks ago so it's on my TODO list. It does seems fun.
One must love a language that has a built-in function for shortest path (www.uiua.org/docs/path). When I browsed the site some weeks ago it had astar but that's now deprecated (in favor of path).
Are you in APL/J/K?

Thus there are now two "CP" in my life. The real (and for me the first): Constraint Programming, and now Competitive Programming.
I hope that will not confuse neither you nor me.

So today I have toggled the autocomplete off. What a mental relief!