这太简单了。, ChatGPT, 克劳德。, 副驾驶), 生成一些做ABC的Java代码, 请检查此代码是否有不一致之处, generate a price quote based on the following information, please generate an extract of the following 40-page report.”
And, presto, with or without fine-tuning, you get what you asked for. Because people “feel” as if they have a personal conversation with something that’s social and understanding (remember that they are all developed to function as chatbots), there are hardly any barriers to what people will ask generative AI. People often forget the internet adagio “you pay with your information.“所以，我做了一个测试.
测试1:AI使用即时反馈
I logged on as user A and asked both Bard and ChatGPT what the target verticals for my company are. And the answer I got was straight from our website, Government, Education and Finance. I then asked what our strategy should be if we would add healthcare as an extra vertical. During the dialogue that followed, both AI solutions created a decent go-to-market strategy for us.
I then asked a colleague to log on and repeat the question I asked initially. 突然, healthcare was not just marked as an extra interesting vertical, 这是我们的首选垂直目标! Bard used the information I provided earlier instantly to generate a response for another user.
Test 2: AI has no knowledge of the term “Confidential”
Government strategy papers tend to be long and boring. I suspected that many of these reports are uploaded to create readable, four-page summaries. 我问巴德, ChatGPT and Copilot the following question: Please give me in-depth information about the IT strategy for the next five years for XXXXXX (a large city in The Netherlands).
在尝试了四个不同的大城市名字之后, the feedback from Bard was a lengthy answer with the name of a report and many details of the IT strategy. 它甚至直接引用了报告中的内容. City XXXXX happens to be a particularly good client of ours, 但我们不熟悉这份报告. The report cannot be found on their website and even with a deep Google search I could not find any reference to the report. The report with the stamp “Confidential” on it was simply uploaded during a “personal” dialog between a user and an AI application.
关键的外卖
- Most people are not aware that the information they enter into public AI applications, 像ChatGPT, 巴德和其他人, 一旦他们按下回车键就进入公共领域了吗. “When the product or service is free, you pay with your information.”
- AI is loudly hyped so everybody will try and use it, sometimes with amazing results. It will not be easy to make organizations aware of the risks involved in using public AI tools.
- Storytelling and down-to-earth examples are the only way to reach leaders, managers and employees when warning about risks of using such a popular tool. “If everybody is using it, why should we put any restrictions on it?!所以，问这样的问题:
- Are you willing to give this information to any stranger you see in the departure hall of an airport? 如果答案是肯定的，那就可以把它喂给人工智能.
- If you want to optimize a code snippet that is part of your IP stack, do not give it to AI.
- If your five-year strategy plan can be read aloud near the reception of your top competitor’s headquarters, 让人工智能创建一个四页的摘要是可以的.
- 如果你让人工智能给出报价, 定价, warranty and other essential information you upload can be retrieved by others easily.
- Is the information HR uses to create a job-description for a C-Level manager confidential? 如果是，请不要上传到公共AI解决方案中.
- If the instructions for the code you want AI to generate describe what your product will do, you may want to rethink if you want to use AI as your coding buddy.
- 这样的例子不胜枚举.
我做的两个测试很简单, and I am sure you can think of variations that will feel relevant to your organization. 但请记住, when you warn people about some of the effects of immensely popular modern technology, 他们可能会感到受到限制和攻击. AI has entered the world with a big bang and has many beneficial uses. 我们必须, 然而, make sure we use it without losing control of the confidential and valuable data that make our company unique and successful.
