Karsten Konrad 🇪🇺
@zabong69
Data Scientist and nerd. PhD in automated theorem proving just before AI became a subfield of Linear Algebra. Machine Learning, statistics, data visualisation, biology, recreational maths, computation, old pocket calculators, and video games
I am using adversarial review to check my math results, i.e. GPT Sol checking results aggressively. For the record: recursively differentiable quasigroups of order 14 and 18, and IQG2(35,4) and IQG2(38,4) all exist. These were all open according to the Handbook of Satisfiability (2021).
The Handbook of Satisfiability (2nd edition from 2021) has a whole list of open problems. Knock yourself out! (QG5(18) and FQG(3^4) have been solved already)
Target on next open problem. It is a sight to behold how my AI buddy relentlessly attacks a problem.
I think I will remember this conversation with Claude for a while. Yes, I talk to it in German.
In the 1990s, the automated theorem provers loved quasigroups, and for some reason this algebraic property below was very popular. (QG5-identity).
"Hey Claude, how do you think I look like?" - I wish, Claude, but thanks anyway.
I keep generating and mutating over several generations, and I am pretty sure that accidentally summoning a demon will happen at some point in the near future.
The fractal Mandelbrot set is sometimes called "Apfelmännchen" ("little apple man") in German. My local Gemma 4 model and I clearly had a misunderstanding when I asked it - in German - to write Python code that draws a Mandelbrot set.
Currently things looking to get wild... lot's of chaos and fractals I never knew existed...
It seems, my little chaos math visualizer will remain the only app so far that I build with Fable.
I am very grateful for @sconradi.bsky.social for rekindling my interest in visualizing math. I clauded a small app that combines genetic algorithms and visualization that finds me "interesting" formulas to look at... and learn more about these chaotic systems of equations.
I am experimenting with genetic algorithm to "grow" interesting iterative equations, but since I never really was into that part of maths very often, I barely know what I (or Claude) is doing.
Computational photography is wild. The iPhone can do things that are almost surreal.
My gadget now manages to tronslate natural language into formal logics - and can prove then.
Building a software that 100% fits your own needs as an industrial data scientist is extremely enjoyable. I never tried cocaine in my life, but it can't be much better than this.
I have clauded a Data Mining suite that combines machine learning and optimization into one visual design approach. 50+ nodes for machine learning and EDA (Linear, Gradient Boosting, UMAP, Preprocessing, visualizer, Python node) and 2 nodes for optimization. So far, it's going well. 260 Unit tests.