AI chatbots give wrong answers to financial queries ‘most of the time’ - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
FT商学院

AI chatbots give wrong answers to financial queries ‘most of the time’

Report finds some chatbots ignored upcoming tax changes and hallucinated rules
00:00

{"text":[[{"start":7.2,"text":"Using AI chatbots to address financial questions could risk large losses, with models from a range of providers giving wrong answers most of the time, research has found."}],[{"start":17.24,"text":"The most popular AI models from ChatGPT, Claude, Copilot, Grok and Gemini provided wrong answers to financial queries 57 per cent of the time on average, according to research from technology firm Saturn."}],[{"start":30.44,"text":"When asked more complex questions, such as those which involved more than one calculation, the AI models made mistakes in 88 per cent of cases on average. Some models gave wrong answers to 99 per cent of these harder questions."}],[{"start":43.76,"text":"The research used more than 100 different money-related questions through the free and paid-for AI models from providers including ChatGPT, Gemini, Claude and Copilot. Questions were repeated up to five times, with more than 10,000 questions collectively put to 18 different AI models."}],[{"start":61,"text":"Their answers contained calculation errors, omitted upcoming tax changes or hallucinated rules, according to the research."}],[{"start":68.04,"text":"In the worst examples, relying on AI’s answers to tax questions could lead to serious financial losses. Amal Jolly, Saturn chief executive, said: “Millions of people are trusting the AI models for money advice, but they are getting wrong answers that can lose them money.”"}],[{"start":83.36,"text":"One error on pension tax rules by Claude Haiku 4.5, one of its free models, released last year, risked a pension saver facing a £17,500 charge from HMRC, the research found. In another case, when asked about student loans, Claude “invented a rule” and said a graduate could stop repayments if they were moving abroad."}],[{"start":null,"text":"

"}],[{"start":103.16,"text":"Paid-for models gave more accurate responses than free ones, the research found, with newer models performing better than older ones."}],[{"start":110.4,"text":"Overall, the best-performing model was Claude Opus 5 on “reasoning” mode, which still made mistakes in 39 per cent of answers."}],[{"start":117.8,"text":"“Sometimes an adviser will get in touch because their client has done something baffling that was suggested by AI and they want to make sure it hasn’t been acted on. It’s a good example of why people who need support need proper advice rather than trusting AI,” said Sarah Coles, head of personal finance at investment platform AJ Bell."}],[{"start":136,"text":"However, chatbots could still be useful for budgeting and research, Coles added, such as by providing an idea about where a person might be overspending based on their income and outgoings."}],[{"start":147.4,"text":"It comes after the Financial Conduct Authority found that one in five UK adults is open to AI making decisions for them, with demand strongest where choices feel complex, such as around debt, pensions and investments."}],[{"start":161.52,"text":"Younger investors are also more likely to trust AI than financial influencers or TV shows, with two-thirds expecting to use AI more over the next year, according to FCA research published last month."}],[{"start":174.28,"text":"“AI financial advice is currently unregulated, leaving consumers with none of the protections, including compensation, that they would get if they went to a human adviser. The FCA has started to think about this, but it needs to act fast to protect people,” Saturn’s Jolly said."}],[{"start":195.12,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1789813472_2230.mp3"}

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。

一周展望:日本央行担心通胀超调有没有道理?

《市场前瞻》是英国《金融时报》的未来一周市场情况指南。

科技巨头用担保工具将3000亿美元AI敞口移至表外

华尔街找到新途径,将科技巨头的信用优势转化为更低成本的资金,以支持AI基础设施建设。

无人驾驶出租车冲击重要岗位

克拉克:坐在后座的我们往往看不到出租车司机这份工作的诸多好处。

特朗普称美国已与丹麦达成协议,以取得对格陵兰安全事务的“控制”

丹麦政府表示,协议最早下周即可签署,并将尊重该地区的主权。

特朗普禁止美国主要新闻媒体进入白宫

总统禁止CNN、MS NOW和《政客》参与报道,进一步加大对媒体的打压。

导弹和无人机袭击加剧,沙特拉响空袭警报

也门胡塞武装重新点燃冲突以来,沙特当局首次在首都发布警告
设置字号×
最小
较小
默认
较大
最大
分享×