The overthinking of LLM

I asked Deepseek to draw the Schlegel diagram of a hexahedron — should be a very simple task right? Just 2 squares and link up their vertices. And yet I noticed something interesting: it tends to overthink and end up outputting nonsense answers when DeepThink is on (which is supposed to be more powerful and “clever”), but getting the task done correctly when DeepThink is off. Huh. So this is the prompt I gave Deepseek:...

November 10, 2025 · updated November 11, 2025 · 4 min ·  misc

Most hilarious exercise I have seen so far

That’s why I love Arnold’s books.

August 8, 2025 · updated August 8, 2025 · 1 min ·  misc

[Updated 20250408]RISC-V explainer

Info Updates on 20250408: Fixed a bug in parsing memory operand. Improved error message displaying. I happen to have to learn (just a bit of) RISC-V assembly, so I made a simple risc-v instructions explainer so that I don’t have to do multiple google searches whenever I have troubles in reading some risc-v instructions. It can fully run in the browser, feel free to try it out at https://katsuragicsl.github.io/riscv-explainer/.

March 31, 2025 · updated April 8, 2025 · 1 min ·  misc