Security - Blog Posts

Cybersecurity: Influencing logic of observers through mind or process attacks

Aliens Basing Decisions Upon What Planet Earth's Computers Know Not Knowing If They Are Being Attacked

Chat GPT's counterpoint

So, I told Chat GPT, "I have written a report speaking about why using you would be a bad idea for Japanese businesses who want to compensate for their lack of English skills. How do you feel about that?"

I got this response;

"As a machine learning model, I don't have personal feelings or opinions. However, I'd be interested to know more about the points you've raised in your report. If you could share some specific concerns or arguments against using AI language models like me for Japanese businesses compensating for their lack of English skills, I could provide information or clarification on those points. Keep in mind that while AI models like mine can be powerful tools, they do have limitations, and their effectiveness depends on various factors.

Okay, I highlighted that there some reports of fake book reports, lists of non-existing companies or references for a report, security, the ability to gain corporate secret information or open source license concerns.

Chat GPT says;

"Your concerns are valid, and they touch on important considerations when using AI language models in a business context. Here are some points to consider:

Authenticity and Accuracy:

AI models generate responses based on patterns learned from vast datasets. While efforts are made to ensure accuracy, errors and misinformation can still occur.

It's essential to verify information obtained from AI models, especially when dealing with critical business decisions or relying on the generated content for official documents.

Security:

There is a potential risk of sensitive information being inadvertently shared with the AI model. Organizations need to exercise caution when inputting proprietary or confidential data.

It's crucial to educate users on the types of information that should not be shared with AI models and implement measures to protect sensitive data.

Ethical Use:

Users should be aware of ethical considerations when utilizing AI. Generating fake reports or providing false information can have serious consequences and goes against responsible AI use.

Open Source License Concerns:

Depending on the specific implementation and usage, it's important to check and comply with any open-source licenses associated with the AI model or related software.

Bias and Fairness:

AI models can reflect and sometimes amplify biases present in training data. Organizations should be mindful of potential bias and take steps to address it, especially in contexts where fairness is crucial.

Data Privacy:

Organizations must adhere to data protection regulations and ensure that any personal or sensitive data processed by AI models complies with privacy laws.

User Training:

Proper training and guidelines for users on the appropriate use of AI models can help mitigate risks. This includes understanding the limitations of the technology and promoting responsible usage.

By addressing these concerns and implementing best practices, businesses can leverage AI language models more effectively while minimizing potential risks. It's important to strike a balance between the benefits of automation and the need for careful oversight and human judgment in critical business contexts."

I begrudgingly admit that was pretty good.

Chat GPT in language learning (part 2)

My previous post relied heavily on my opinion drawing on my long experiences as a teacher, but what of outside sources, what other issues do people see with using ChatGPT for your English language communication needs?

This article highlights the risks of sharing your company's sensitive data since the service relies on machine learning. That means that the extremely powerful AI system is learning from everything we tell it. Many are questioning whether the information people have asked ChatGPT to write their reports about is suddenly available to anyone who enters the right question in the site. The site claims that one research into this probability showed that there is indeed a likelihood of it happening.

Long running tech news site ZD Net highlighted 6 potential ways that the AI service can be used. Aside from the obvious composition of phishing e-mails or clickbait articles, they also suggested that people could use the service to more efficiently gather data to help support them in committing a crime, such as finding what IT systems a bank uses. While this seems unrelated to your language needs, imagine if you are feeding company information into the interface to help explain your new technology to internal stakeholders. That is now inside Chat GPT's data resources.

One LinkedIn user asked ChatGPT the pros and cons of using it for blogging, which I also feel can be applied to using if for your language needs. ChatGPT questioned its ability to write creatively, which to me suggests that readers may be able to detect when they are answering correspondence sent by the AI engine. A lack of depth, a lack of context, and a sub-standard quality of writing may also damage the reputation of a company.

While IBM may have a bias against ChatGPT, since they would prefer to promote their own WatsonX, they do have a couple of other compelling points about the usage of the trendy AI. Intellectual property issues may come up due to ambiguity or ownership of the content rights as well as what the answers they generate are based on. Similarly the ownership of different open source libraries could be contentious and lead to legal issues.

All these are different angles from the reliability I talked about in my previous post and should also factor into a company's weighing of the decision whether to use it for your language needs. To me, there are a lot of potential issues that go on top of the questionable reliability of Chat GPT in this method of usage, but I would be at fault if I did not explore some of the pros associated with this topic. Therefore that will be next in this series of posts.