Whether it be curiosity, or the business pressure of integrating Artificial Intelligence (AI) into their unique business model, many people seem to be turning to ChatGPT for use in both the workplace and for personal tasks.
ChatGPT was developed by Open AI, and soon after its launch, over a hundred million people began using it within a span of two months. Microsoft soon after became an important investor in the startup, spending 10 billion United States dollars to further the development of GPT4, and integrating it into the popular Bing search engine. You can now find Bing with GPT4 on both the App Store and Play Store, for easy access and use.
ChatGPT is designed for ease of use, particularly for automated tasks like writing emails with relative ease. For example, you can use the details of a job listing, through ChagGPT, to generate the perfect cover letter for it, or a HR assistant can generate a task listing from scratch if they only have a sketchy idea of the task they are advertising.
The risk, however, of divulging confidential information is high when it comes to these tasks. After all, on the one hand, if the job listing is of a common sort, sharing it with two hundred million ChatGPT users will hold no risk. But, if it includes information that a competitor could use to identify your own business, it could be potentially damaging. This became evident when Samsung experienced a breach of confidential information from staff using ChatGPT; it is thus important to consider and weigh the risks of using it, before using the system in a professional context.
Interestingly, the GPT4 model, the most recent and the most accurate of the GPT family, has a voice-to-text feature that is incredibly precise. The danger in using this feature however, is the potential for transposed information to be part of the training set that helps the AI to improve its answers, as has been experienced with the Samsung breach. This is concerning, as it allows developers to paste code and use it as an input to ask ChatGPT to improve it, a feature which became immensely appealing, as it could significantly speed up development processes and coding.
The challenge with this has caused Italy, and other European countries to take up a stance against ChatGPT, claiming it infringed upon the EU General Data Protection Regulation. Italy will likely retract this opinion by the end of April, as it is replaced with an age verification check – users must be over 18.
Generative AI requires a massive amount of data to produce its predictions; improvement depends upon feedback it receives from its users. Herein lies the problem for businesses, with their staff providing the necessary feedback that could include proprietary material, without even knowing it. The solution for this then is a simple one – use ChatGPT or Bing only for information that your business would otherwise share openly.
The perplexity of your content, and its burstiness (variations in perplexity) can be calculated to detect if generative AI had been used to generate it using GPTZero, an AI by Open AI. As AI services like ChatGPT improve, with the improvement of GPT4, these detection services need to evolve in order to remain effective.
Ultimately, it is up to the business to decide whether to use search engines to retrieve information or to generate information. When it comes to the use of ChatGPT, however, it is important to differentiate between confidential content and information that would be shared publicly. If the output of ChatGPT would normally be considered confidential, then it should not be submitted or shared.
