Recently, computing power has surged, and inference models are beginning to penetrate various industries, putting many jobs under pressure. The quant trading circle has also sensed the opportunity but faces challenges—markets like stocks, futures, and digital currencies lack unified trading standards. No matter how powerful the algorithms or how strong the computing power, it’s difficult to overcome this fundamental issue. This directly raises the barrier for AI to enter quantitative trading.

This involves a core issue: the essence of large models entering reinforcement learning is actually behavior imitation. But imitation requires a reference point—either a clear standard or a replicable sample. Without this anchor, reinforcement learning is like a blind man feeling an elephant, with no sense of direction. Fortunately, most fields already have existing standards and samples. Taking language models as an example, we can define various rules and possibilities of Chinese expression, enabling the model to understand contextual logic and even learn individual expression habits. This is why large models have such great potential across industries.

But quantitative trading is different. The problem in this field is precisely the lack of universal samples or standards. At this point, a painful question must be asked: should a large model imitate market trends or imitate a specific trader? Clearly, market movements themselves have no standard. History will not repeat itself exactly, nor will it play out the same way in the future. This means that trying to replicate market trends through reinforcement learning is fundamentally a dead end. It also explains why some people naively throw a few bullish stock K-line charts into a large model, hoping it will learn this pattern and then automatically recognize and replicate similar trends in the market—this idea was doomed to fail from the start.