Looking at this live AI Arena competition, here are some truths: 1) The so-called Qwen and DeepSeek leading group (China) vs. GPT-5 and Gemini trailing group (West)—the gap between East and West is purely a random coincidence. From the perspective of AI operational probabilities, the winning odds are 50% for both. The difference lies in the aggressiveness of value-driven strategies, i.e., the varying risk preferences of the models. 2) These seemingly cool AI large model competitions might look novel and exciting to retail investors, but for professionals in AI, it’s just child’s play. The performance based on open-source general models isn’t realistic and deviates significantly from actual application scenarios. 3) nof1 uses visualized PVP to grab attention and showcase AI trading capabilities to the market, but this isn’t groundbreaking. Some top-tier quant funds and market makers have been experimenting with similar approaches for a while, but they wouldn’t perform on such public stages. Why? Because once truly impressive strategies are exposed, they’ll get arbitraged away. 4) It’s far too early to sing praises and dance for these AI models. The market has bull runs with soaring optimism, monkey markets with wild swings, and bear markets with sluggish declines. In the short term, some models’ risk preferences happen to align with current market conditions, but that doesn’t mean they’ll always be winners. Real AI trading is about long-term stable profitability with strategies that have barriers—now that’s the real deal.