v4nn4

Hi, author here. Full disclosure: I am not a linguist and not in NLP, just interested in learning German for my own sake. I wrote a version by hand pre-2022 with a very small scope as a fun project. I recently revisited it using Fable 5 and was able to reach very high accuracy, that is a perfect agreement on forms where multiple gold sources agree. There is still work to be done on rare verbs and forms tied to dialects, where the gold sources disagree.

I am documenting a bit in the blog post the verification loop, and I am planning to look into replicating this for other languages as well. If any linguists, NLP professionals or language app builders are interested in using the library through Python or an API, let me know! Cheers.

Indeed nice catch! Not as simple as lowercasing the input, but as you said it was fixable and is now fixed. Thank you!

The filters selection seems to return a union not an intersection which is a bit confusing, at least to me.

Yes indeed. You can tap on the orange square to remove it, put the red Z on the top left corner, the yellow L next to it and complete the puzzle!

Thanks for the feedback, will look into it. I have too distinct interfaces for mobile and desktop so it could be that you are stuck with the desktop one that uses click (vs touch).

Thanks! The difficulty is based on the number of steps taken by my solver to solve the puzzle using backtracking and force moves calculations. Humans make better guesses, so it is not just luck ;).

I’ve been watching this space for a while and built my own puzzle with Cursor. Vibe coding speeds things up, but getting the idea, difficulty balance, and UI right is still tricky. Probably depends a lot on the type of puzzle (word-based vs. object placement, etc.).

Shameless plug: https://play7fold.com

Stochastic calculus is required to derive closed formulas and approximations used to calibrate SDE models. Similarly to deep learning, the secret sauce lies in the training, less in the inference. The code used by banks is closed source, and the research papers are missing said secret sauce. Calibrating models in a production environment handling correlation, multi-curves, stochastic funding, discrete dividends, etc. is not a solved problem. Interest rate derivatives modeling heavily relies on change of measure, even when using simple models.

Good point. On a serious note, I probably overreacted, sorry about that. I have been working as a derivatives quant for a decade and thought the claim that stochastic calculus was not used/useful was ridiculous.

The comment above is probably from a bot. You do need an extensive understanding of stochastic calculus to maintain quant models code, let alone explain what it does to regulators.

Wtf wow

Pretty neat for a high school side project!

Totally second this. I often mention this scandal when discussing pseudo sciences, along with Elisabeth Teissier case. Quite embarrassing for my country but surely a good laugh. Love that this ended on HN !

Grichka thesis : https://tel.archives-ouvertes.fr/tel-00001502/document

Igor thesis : https://tel.archives-ouvertes.fr/tel-00001503v1/document