A Rant About Bilibili's Recommendation System

My Bilibili recommendations are full of things I don’t want to watch, and I’m very unhappy about it right now:

  1. Enough with those interview question videos. How can you claim to understand concurrent programming when you talk about application-layer spinlocks without even considering interrupts? And writing code to check for cycles or intersections in a linked list you don’t understand is a sign of a bad engineer. If I had a teammate like that, I’d either make them brush up on their data structures or just kick them off the team.

  2. I have no interest in watching videos on algorithm problems, especially those that advocate for “not studying something because it’s rarely or never tested”—that’s practically dancing on my last nerve. I want a proper education grounded in Discrete Mathematics and Its Applications and Introduction to Algorithms, supplemented with training in parallel algorithms, not just an accumulation of problem types and solution patterns.

  3. The videos on modern C++ and system details are even more likely to be viewed through my highly critical lens. I couldn’t care less about what the syntax looks like; I want to see an analysis of the evolutionary trends and the value behind them, not just watch some flashy feature or accumulate scattered bits of knowledge.

  4. When I finally see a video that looks interesting and click on it, I discover it’s a re-upload from YouTube with an AI-generated, machine-translated voiceover. I leave immediately. What’s the point of having machine-translated YouTube videos on Bilibili?

  5. Why is there such a flood of videos on research methodology and academic topics (especially AI)? I don’t have a supervisor to please. You can’t just treat me like a graduate student and push a ton of meaningless deep learning content and paper analyses at me just because I clicked on a few related videos.

  6. Some of the math videos are indeed interesting, but the recommendations are mainly for graduate school or college entrance exams, which makes me feel like the algorithm is “insulting” me. I have zero interest in solving those complex integrals. Even if you master them, you’ll never outperform symbolic computation. But when it comes to the real deal—the history of mathematics, proofs, abstraction—they can’t deliver.

  7. Then there’s a bunch of videos with titles like “Top-Tier University Student…” or “High School Student Does X…” I have no interest in these either. The value of a person’s thoughts and knowledge doesn’t come from their title, and the validity of their words should be tested by reality and logic. Besides, those who use their age to show off their talent are a bit boring. They haven’t experienced how truly jarring a real mismatch between age and cognition can be. Instead of reinventing the wheel in isolation, they’d be better off acknowledging their limitations and standing on the shoulders of giants.

  8. Just when I finally scroll to a gaming video or a female vlogger I want to watch, I get bombarded with a heap of cybersecurity content. I have very limited interest in cryptography and network security; professional matters should be left to the professionals. My taste, or rather my instinct, is to evaluate things from a performance and systems perspective, not a security one. I might be able to do some high-performance computing and systems programming, but I can’t be a security expert.

I admit, I probably haven’t used Bilibili enough to properly “train” the algorithm for my profile, but its personalization is truly awful. It might be able to guess what kind of person I am and my general field, but it fails to recommend content with sufficient depth and information density. It just matches superficial tags—the signal-to-noise ratio is far too low, resulting in a terrible experience.