Tencent's $850K KDD Cup: Industrial Track Champion Coded Entirely with DeepSeek
赛博禅心 · wechat · 2026-08-13
Tencent offered over 6 million RMB in prizes at the KDD Cup to find a next-generation recommendation system architecture that unifies user behavior sequences and profile features.
Academic Track: Won by students lozyyeah, taking home 2 million RMB. Their CRAFT method uses a flow-matching-inspired transport field to control information transfer between layers, preventing intention overwriting and achieving an AUC of 0.838.
Industrial Track: The champion team 'Sunshot' won with QueryFormer (AUC 0.832). Surprisingly, their entire algorithm was written and debugged using the DeepSeek web interface. They prompted the model with code and errors, ran the output on servers, and defeated human experts globally.
More from Fun
- Sabine Hossenfelder fires back: academia's impact metric turned physics into a paper mill — skdh · 2026-10-16
- Meme: Clicking 'Yes to all' on Claude Code without reading anything — jnack · 2026-10-16
- Minecraft AI mod turns region selection + prompts into autonomous building agent — Acclynn · 2026-10-16
- Throwaway prop code in movies is getting eerily accurate, and AI is why — mike64_t · 2026-10-16
- Distillation copies logits, not hidden CoT — open-sourcing 1.3 leaks IP, not reasoning data — teortaxesTex · 2026-10-16
- Scott Aaronson's 9-year-old son: "I don't think problems should be independent of the axioms" — basedjensen · 2026-10-16