Some thoughts on participating in an AI programming challenge organized by a major Chinese corporation

To give a brief background: a while back, I participated in an AI programming competition organized by Trea, but I encountered some unfair treatment that I’d like to share with everyone. I’m also curious to know how judges typically score competitions like this.

My entry transforms my personal notes on the underlying mathematics of LLMs into an interactive interface, which I have deployed on GitHub Pages.(I've already deleted the code repository. :( ).

Here is a brief overview of the process: over 14,000 independent developers participated in the preliminary round. The submission deadline was July 15. The review period ran from July 16 to July 21, with the 350 qualifiers announced on the afternoon of July 21. Of these spots, 300 were determined by judges officially selected by ByteDance, while the remaining 50 were awarded to entrants who posted their work on TikTok (known as Douyin in China) and ranked in the top 50 based on likes and comments.

The overall atmosphere of the competition was reasonably cordial—until the list of those advancing to the next round was announced on the afternoon of July 21st. Many independent developers immediately sought out Trea’s operations team to demand an explanation.

1. The review process was a "black box," and the scoring criteria were even altered (a community member pointed this out).
2. A significant number of entries from contestants went unreviewed or unexamined.
3. Some posts that did not comply with the preliminary round guidelines were able to advance.

Developers are questioning these specific points: is the reality living up to the official pre-launch marketing? Clearly, it is not. On a broader level, this represents a crisis of trust between the community and the product.

I wasn't selected either—I didn't even make the top 2,000. Everyone is questioning the fairness of the competition here. While it is undeniable that the top 300 entries included truly outstanding work worth learning from, the fact remains that these merits do not overshadow the community's desire for fairness or their doubts regarding the contest's integrity.

Here are my main takeaways from this competition.

1. The rules were a joke; they changed constantly, and the organizers themselves disregarded them, even granting special exceptions to entries that didn't meet the requirements
2. The community atmosphere is quite poor; a handful of members enjoy mocking those who ask genuine questions, and the official team turns a blind eye—unless a topic draws a crowd of dozens, at which point they only show up to watch the drama unfold.
3. As for the competition review process, luck plays a significant role. Some participants reported their services being accessed from IP addresses in Shanghai or Beijing, while others found their servers completely untouched. Perhaps those entries simply didn't appeal to the judges' tastes. The same applies to my own work; the link click count visible in the post shows no increase.

Finally, there is no need to deliberately alter your work just to cater to the judges' preferences. No matter how well you execute it, people will inevitably focus on the weakest aspects, nitpick, and dismiss both the work and the creator as worthless. In my view, a project is a success if it helps even a small number of people solve real-world problems—something that cannot simply be judged by someone holding a title like "judge."

This was perhaps the worst programming competition experience I’ve ever had. Trae’s operations team focused entirely on talking about how hard they worked—mentioning, for instance, that they had put in five consecutive days of overtime reviewing tens of thousands of submissions. They refused to disclose the shortcomings of the rejected projects or the actual scores of the ones that advanced; instead, they simply had community members aggressively promote the narrative of the team's hard work.

I saw many rejected projects that actually held significant social value. One such project—an app designed by a couple to help children suffering from depression or mental health issues—failed to advance, which struck me as particularly ironic. I provided the couple with free technical support on the side, and they were very grateful; at that moment, at least, I felt that the true meaning of the competition lay not in rankings, but in helping those who genuinely need it. Yet, Trae—and ByteDance—you have truly betrayed the original intent behind promoting this competition.

The screenshot shows the logo designed for their app, right? I really like it—it’s a sprouting seed, symbolizing new hope.