AITechNode9 months ago
Tencent Tests AI Strategic Reasoning via Gaming
1 min read
Tencent researchers developed the "Think in Games" (TiG) framework, training AI models on _Honor of Kings_ match data. Using supervised and reinforcement learning with GRPO, Qwen3-14B achieved 90.9% strategic accuracy after 2,000 steps, surpassing Deepseek-R1 at 86.7%. Smaller models outperformed larger ones under certain conditions.
Level
Hype check
Tap to vote and see what everyone thinks.
