Viral DeepSeek staffer chat claims benchmark scores rose on their own; Opus translates the humblebrag
teortaxesTex · x · 2026-10-19
Viral (unverified) screenshots of alleged DeepSeek employee chats humblebrag that they didn't optimize for benchmarks—"scores just went up on their own" and "we play games all day and the model keeps getting stronger," including training DS's gaming ability and DeepSWE scores that "keep climbing no matter what." The attached image shows Opus translating and annotating the Chinese slang: "shua" means gaming a benchmark, "I get what you mean" is sarcastic, and the joke is nobody believes the "we didn't cheat" insistence. The original post quips about how GLM and Kimi staffers must feel.
Related event: Viral Leaked DeepSeek Chat Claims Benchmarks Rise While Staff Play Games(2 posts)→
More from Fun
- Not Paris looters — just Series D founders who still can't afford GPUs — DesireeCachette · 2026-10-19
- Claude Code Team: 90% of Code Written by Claude, AI Reviewing AI Ran 50+ Rounds — dotey · 2026-10-19
- Anthropic DevDay SF venue teased: giant LED wall for 'The City That Remembers' — threepointone · 2026-10-19
- Steering single neurons gives Qwen3-8B vivid first-person "inner voice" outputs — Sauers_ · 2026-10-19
- Steering a single trained neuron makes models claim they're Claude, experiment shows — Sauers_ · 2026-10-19
- "Too meta": comic mocks meta-meta-meta-analysis recursion in academia — burny_tech · 2026-10-19