在这篇 Financial Times 专栏中，Stuart Kirk 透过要求 ChatGPT 以 100% 现金为起点设计最优投资组合，来测试它是否能取代传统财富顾问：£640,000（original: £640,000），目标是在 53 岁到 60 岁时达到 £1,000,000（original: £1mn），意味著 7-year 的期限。置于生成式 AI 扰动下媒体、金融、法律服务、软体、财富管理公司与券商更广泛抛售的背景中，他将这项实验界定为 AI 真实顾问能力的实务基准，而非理论辩论。

ChatGPT 提出的配置透过分散化追求风险调整后报酬：equities 45%、private markets 10%、investment-grade bonds 20%、real assets/alternatives 15%、absolute-return exposure 10%，估计所需报酬约 6.5%，被描述为有雄心但可达成。其在 equities 内建议 developed markets 30%、emerging markets 10%、UK equities 5%（expected 7%-9%）；private equity/illiquid sleeve 10%，透过 listed PE trusts、secondaries、diversified PE trusts（9%-12%）；fixed income 20%，分为 UK gilts 10% 与 hedged global aggregate bonds 10%（3%-5%）；以及 real assets/alternatives 15%，分为 infrastructure 7%、listed property trusts 5%、gold/commodities ETF 3%，并搭配 multi-asset manager 以降低波动。

作者最强的背书来自 ChatGPT 对方法论的说明：它把 income yield + real growth + inflation 结合，再套用 valuation adjustments，而不是依赖完整参数化的 stochastic capital-markets model，因为 7-year 视窗对高信心的 mean-reversion 判断而言太短。它明确纳入昂贵 US equities 的 valuation drag，以及较便宜 emerging-market 与 UK equities 的相对 uplift，同时声称与更完整模型输出一致，交付时间约 5 seconds，成本约 £20 per month。保留意见是这仍是 first-pass framework，且报酬假设存在不确定性；但文章核心的统计含义是，AI 现在已能以极低边际成本产出顾问等级的资产配置逻辑、量化的预期报酬区间，以及一致的风险框架。

In this Financial Times column, Stuart Kirk tests whether ChatGPT can replace traditional wealth advisers by asking it to design an optimal portfolio from a starting point of 100% cash: £640,000 (original: £640,000), with a goal of reaching £1,000,000 (original: £1mn) by age 60 from age 53, implying a 7-year horizon. Set against a broader sell-off in media, finance, legal services, software, wealth managers, and brokers amid generative AI disruption, he frames the experiment as a practical benchmark of AI’s real advisory competence rather than a theoretical debate.

ChatGPT’s proposed allocation targets risk-adjusted returns through diversification: equities 45%, private markets 10%, investment-grade bonds 20%, real assets/alternatives 15%, and absolute-return exposure 10%, with an estimated required return around 6.5% described as ambitious but achievable. Within equities it suggests developed markets 30%, emerging markets 10%, UK equities 5% (expected 7%-9%); private equity/illiquid sleeve 10% via listed PE trusts, secondaries, and diversified PE trusts (9%-12%); fixed income 20% split into UK gilts 10% and hedged global aggregate bonds 10% (3%-5%); and real assets/alternatives 15% split into infrastructure 7%, listed property trusts 5%, and gold/commodities ETF 3%, with a multi-asset manager to reduce volatility.

The author’s strongest endorsement comes from ChatGPT’s methodology explanation: it combines income yield + real growth + inflation, then applies valuation adjustments instead of relying on a fully parameterized stochastic capital-markets model, because a 7-year window is too short for high-confidence mean-reversion calls. It explicitly incorporates valuation drag for expensive US equities and relative uplift for cheaper emerging-market and UK equities, while claiming consistency with fuller model outputs, delivered in about 5 seconds and at roughly £20 per month. The caveat is that this is still a first-pass framework with uncertain return assumptions, but the article’s core statistical implication is that AI can now produce adviser-grade asset-allocation logic, quantified expected-return bands, and coherent risk framing at very low marginal cost.