不只丁迪，还有一小群交易员也在训练 AI 工具，以从每日微小调整中找出模式。根据丁迪的说法，在过去 4 个季度共 225 个事件中，Xiao Long 对方向的判断超过三分之二时间正确；在超过 40% 的情况下，预测与官方水平相差不超过 20 个点子。新加坡的一名对冲基金交易员使用类似工具，也得出相近结论：自 3 月以来，数据显示人行对人民币较慢升值的容忍度提高。
这些尝试出现在更大的政策不透明背景下。人行不像一些主要央行那样频繁说明，市场只能从中间价推断意图，而在境内市场人民币被限制在中间价上下 2% 的交易区间内。自去年 4 月初唐纳德·Trump 发动全球关税以来，人民币已对美元升值逾 8%；同时 Goldman Sachs Group Inc. 估计人民币仍被低估超过 20%。Stephen Jen 指出，中间价体系是结构性「半透明」的：它会提供讯号，但仍保留裁量空间。
China’s daily yuan fixing is being probed with AI by a small set of traders trying to infer People’s Bank of China intentions from tiny adjustments. Ding Di, in a Shanghai bank investment unit, has used the Gemini-powered Xiao Long model for about a year; on March 4 it forecast 6.9122 for the dollar-yuan fixing versus the official 6.9124, missing by only 2 pips. Di said the model was helped by detecting a change in the day-before setting method, implying the PBOC may have given more weight to overnight yuan moves.
Other traders using similar tools say the pattern has become clearer. Over 225 episodes across the past four quarters, Xiao Long correctly called the direction more than two-thirds of the time, and in over 40% of cases it was within 20 pips of the official level. A Singapore hedge fund trader with a comparable model reached the same broad conclusion: before the war, the fixing showed an appreciation bias, but since March the PBOC appears more tolerant of slower yuan gains.
The backdrop is an opaque policy regime. The PBOC rarely briefes market participants, leaving the fixing as the main public signal, while the onshore yuan trades in a 2% band around it. Since Donald Trump launched sweeping global tariffs in early April last year, the yuan has strengthened more than 8% against the dollar, even as Goldman Sachs says it remains more than 20% undervalued. Stephen Jen describes the system as structurally semi-transparent: it offers clues, but retains discretion, and AI is still vulnerable to hallucinations even as human oversight improves its usefulness.