Generative AI: This is how you can use ChatGPT safely

Your new AI assistant will write poetry for you, generate ideas for your business, check your emails for mistakes and maybe even pass your MBA exam, albeit with only a B- average.

But would you trust a well-meaning AI assistant that can sometimes offer plausible yet inaccurate answers? And does anyone else see the queries that you submit, and if so, who? In other words, how safe is ChatGPT?

ChatGPT and other large language models (LLMs) can provide opportunities at home and at work for efficiencies and improvements – but alongside these opportunities come risks around not just accuracy and bias, but the handling of personal information and security too.

Here’s how to use ChatGPT safely for all your needs.

Training ChatGPT works by giving it previously unseen data sets to establish how well it makes predictions as it reviews the data. Then it is tested to see how well it performs with that new data.

As such, its ability to supply quality information is only as good as the information that was given to it, and the current version of ChatGPT is only trained on data up to 2021. The data that feeds LLMs is likely to contain conflicting evidence, opinions and unreliable conclusions, and so bias issues are as inevitable as accuracy issues.

For personal and business use, the key to navigating accuracy and bias issues with ChatGPT is vigilance. Fact-checking statistics, using multiple sources for information, and engaging critically with AI-generated content with the understanding that generative AI may not understand complex or nuanced topics can mitigate accuracy issues.

Because LLMs can only work with the data that is given to them – and as long as that data creates inaccuracies and bias – humans will need to be involved in the process to identify and remove them, says the Harvard Business Review.

Even though website visits are down for the third consecutive month, ChatGPT still had 180.5 million unique users in August 2023, and students returning to school in September may lead to a bump in traffic, as the education sector cautiously embraces AI.

Many free apps and websites make their money by collecting and selling data, and ChatGPT operates in a similar way. Ensuring safety around your personal data, your children’s data, or the organization you work for is an important part of online security best practice.

ChatGPT does not store user-specific data, nor does it store details of interactions beyond the immediate conversation it is having. But it does save all of the conversations you have with it, store them, and utilize this data to improve its language model. So the human trainers working behind the scenes at OpenAI will be able to see conversations and queries.

To ensure a safer environment for your personal information and security, avoid sharing private information with ChatGPT. Consider disabling chat history and model training. But be aware that even if you disable the history, your chats will still be stored on Open AI’s servers for 30 days so that staff can monitor for abuse.

And if you feel like this is not enough, you can request to delete your entire account and its data by emailing dsar@openai.com.

Lastly, take care when installing extensions and apps on your computer or phone. Some apps and browsers that claim to bring ChatGPT functionality to your browser could be harvesting your data.

For organizations using generative AI such as ChatGPT as part of their day-to-day operations, making sure both their staff and their customers are clear on the boundaries of use is a good first step, and many organizations provide clear and transparent AI policies to ensure best practices.

For organizations looking to build their own LLM AI policies, the Corporate Governance Institute provides some guidance for areas to cover.

“Incorporating cyber-resilience governance into [a company’s] business strategy is one of the most impactful principles when it comes to cyber resilience,” notes the World Economic Forum’s latest Global Cybersecurity Outlook report.