OpenAI Says, Let Us Be Open: Intentionally or Unintentionally
July 12, 2024
This essay is the work of a dinobaby. Unlike some folks, no smart software improved my native ineptness.
I read a troubling but not too surprising write up titled “ChatGPT Just (Accidentally) Shared All of Its Secret Rules – Here’s What We Learned.” I have somewhat skeptical thoughts about how big time organizations implement, manage, maintain, and enhance their security. It is more fun and interesting to think about moving fast, breaking things, and dominating a market sector. In my years of dinobaby experience, I can report this about senior management thinking about cyber security:
- Hire a big name and let that person figure it out
- Ask the bean counter and hear something like this, “Security is expensive, and its monetary needs are unpredictable and usually quite large and just go up over time. Let me know what you want to do.”
- The head of information technology will say, “I need to license a different third party tool and get those cyber experts from [fill in your own preferred consulting firm’s name].”
- How much is the ransom compared to the costs of dealing with our “security issue”? Just do what costs less.
- I want to talk right now about the meeting next week with our principal investor. Let’s move on. Now!
The captain of the good ship OpenAI asks a good question. Unfortunately the situation seems to be somewhat problematic. Thanks, MSFT Copilot.
The write up reports:
ChatGPT has inadvertently revealed a set of internal instructions embedded by OpenAI to a user who shared what they discovered on Reddit. OpenAI has since shut down the unlikely access to its chatbot’s orders, but the revelation has sparked more discussion about the intricacies and safety measures embedded in the AI’s design. Reddit user F0XMaster explained that they had greeted ChatGPT with a casual "Hi," and, in response, the chatbot divulged a complete set of system instructions to guide the chatbot and keep it within predefined safety and ethical boundaries under many use cases.
Another twist to the OpenAI governance approach is described in “Why Did OpenAI Keep Its 2023 Hack Secret from the Public?” That is a good question, particularly for an outfit which is all about “open.” This article gives the wonkiness of OpenAI’s technology some dimensionality. The article reports:
Last April [2023], a hacker stole private details about the design of Open AI’s technologies, after gaining access to the company’s internal messaging systems. …
OpenAI executives revealed the incident to staffers in a company all-hands meeting the same month. However, since OpenAI did not consider it to be a threat to national security, they decided to keep the attack private and failed to inform law enforcement agencies like the FBI.
What’s more, with OpenAI’s commitment to security already being called into question this year after flaws were found in its GPT store plugins, it’s likely the AI powerhouse is doing what it can to evade further public scrutiny.
What these two separate items suggest to me is that the decider(s) at OpenAI decide to push out products which are not carefully vetted. Second, when something surfaces OpenAI does not find amusing, the company appears to zip its sophisticated lips. (That’s the opposite of divulging “secrets” via ChatGPT, isn’t it?)
Is the company OpenAI well managed? I certainly do not know from first hand experience. However, it seems to be that the company is a trifle erratic. Imagine the Chief Technical Officer did not allegedly know a few months ago if YouTube data were used to train ChatGPT. Then the breach and keeping quiet about it. And, finally, the OpenAI customer who stumbled upon company secrets in a ChatGPT output.
Please, make your own decision about the company. Personally I find it amusing to identify yet another outfit operating with the same thrilling erraticism as other Sillycon Valley meteors. And security? Hey, let’s talk about August vacations.
Stephen E Arnold, July 12, 2024
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
