Text math tasks trainer

[Anon]

I found a dataset called "GSM8K" (Grade School Math,  >8k tasks) and made game with 3 modes: speed mode, endless mode and most hardcore - memory mode. In speed mode you need to correctly answer 5 questions as fast as possible.  Memory mode is a mode in which every next word appears only after previous disappears; you starts with base speed of 1000, which increases by 25 each times you answer correctly and decreases by 50 each times you answer wrong. After 5 tasks it ends and new base speed saves.

This game is both speedreading and mental calculation trainer.

I used this dataset just for test, original purpose was to make somewhat similar but with more serious math tasks with many ranges of difficulty including olympiad level, so that as you progress in solving the problems, the difficulty of the problems increases. And with many fields like number theory, combinatorics, probability theory, geometry, algebra, calculus etc. So you can track your rating separately in different fields. If you know good datasets for this, please write.

2 options:

1) Download as exe file https://drive.google.com/file/d/1gzs08SOUP770OU8yAkKSpThz1mS9uRZS/view?usp=sharing

Super fast and simple, you don't need to pre-install anything.

https://www.virustotal.com/gui/file/2e6e5c939c1dc8b43b4e7e7be5cf4b6da977f239e0d24f9d3b46309fcb04141e/detection

(Unfortunately there are several false positives that virustotal gives to all exe files built with pyinstaller)

2) Download rar archive and run main.py file (but you need to install Python and several libraries)

So you can edit anything you want,  even completely change dataset of tasks. Also you can use this option if you don't want to download exe file from some random guy from the Internet, which is pretty reasonable lol :D

You can watch the gameplay in the attached video.

[Anon]

Just achieved 1625 speed level in memory mode? Anyone here to compete?

I think this is one of the truly working ways to train speed reading. In fact, one of the main bottlenecks is a familiar thing: when you read a text, you forget what you just read and go back. This is a deficit of semantic working memory (a domain of working memory, separate from verbal/audial, which works directly with meanings, and not words, like verbal/audial WM)

"For performance on the active-passive comprehension task, we found that semantic WM, but not phonological WM, predicted comprehension of dative sentences when controlling for comprehension of transitive sentences" - https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2022.887586/full

"Semantic working memory: Evidence for a separate system that maintains meaning" - https://psycnet.apa.org/record/2002-95004-011

среда, 9 апреля 2025 г. в 02:12:21 UTC+3, Anon:

[Anon]

Additional training load on the brain if English is not your native language

среда, 9 апреля 2025 г. в 16:32:15 UTC+3, Anon:

[Anon]

I experimented with Omni-MATH dataset, but there are too few tasks (there are 4428 of them in total, but after cutting the tasks that have an answer in the form of complex LaTex notation, for example something like this: \\[\n\\frac{\\cos 1^\\circ}{\\sin^2 1^\\circ}\n\\], there are several times fewer left. And if you make a progression of tasks by difficulty level, then for each difficulty level separately there will be even fewer tasks. This dataset is great, but too small

среда, 9 апреля 2025 г. в 02:12:21 UTC+3, Anon:

[Anon]

An example of my experiments with the Omni-MATH dataset:

среда, 9 апреля 2025 г. в 17:04:59 UTC+3, Anon:

[Anon]

Now it's 1725

среда, 9 апреля 2025 г. в 16:32:15 UTC+3, Anon:

[Anon]

1975

четверг, 10 апреля 2025 г. в 18:13:52 UTC+3, Anon:

[Anon]

Achieved 2500 speed level in memory mode

четверг, 10 апреля 2025 г. в 18:54:15 UTC+3, Anon: