5 Quick Tips to Enhance AI Usage for Better Results and Safety

In today's world, dodging artificial intelligence (AI) is getting tougher by the day. Take Google searches, for instance—they're now featuring AI-generated responses. With AI becoming such a staple in our lives, ensuring its safe use is more crucial than ever. So, how can you, as an AI user, navigate the world of generative AI (Gen AI) safely?

Also: Here's why you should ignore 99% of AI tools - and which four I use every day

At SXSW, Maarten Sap and Sherry Tongshuang Wu, assistant professors at Carnegie Mellon's School of Computer Science, shed light on the limitations of large language models (LLMs), the tech behind popular Gen AI tools like ChatGPT. They also shared tips on how to use these technologies more effectively.

"They are great, and they are everywhere, but they are actually far from perfect," Sap pointed out.

The tweaks you can make to your daily AI interactions are straightforward. They'll shield you from AI's flaws and help you extract more accurate responses from AI chatbots. Here are five expert-recommended strategies to optimize your AI use.

1. Give AI better instructions

AI's conversational prowess often leads users to give vague, brief prompts, similar to chatting with a friend. The issue here is that with such minimal guidance, AI might misinterpret your text, lacking the human ability to read between the lines.

During their session, Sap and Wu demonstrated this by telling a chatbot they were reading a million books, which the AI took literally rather than understanding the exaggeration. Sap's research shows that modern LLMs struggle with non-literal references over 50% of the time.

Also: Can AI supercharge creativity without stealing from artists?

To sidestep this, be more explicit in your prompts, leaving less room for misinterpretation. Wu suggests treating chatbots like assistants, giving them clear, detailed instructions. It might take a bit more effort to craft your prompts, but the results will better match your needs.

2. Double-check your responses

If you've used an AI chatbot, you're familiar with "hallucinations"—when the AI spits out incorrect information. These can range from factually wrong answers to misrepresenting given information or agreeing with false statements from users.

Sap noted that hallucinations occur between 1% and 25% of the time in everyday scenarios, with rates soaring above 50% in specialized fields like law and medicine. These errors are tricky to spot because they sound plausible, even when they're off the mark.

Also: AI agents aren't just assistants: How they're changing the future of work today

AI models often reinforce their responses with phrases like "I am confident," even when wrong. A cited research paper revealed that AI models were confidently incorrect 47% of the time.

To guard against hallucinations, always double-check the AI's responses. Cross-reference with trusted external sources or rephrase your query to see if the AI's response remains consistent. It's easier to catch errors if you stick to topics within your expertise.

3. Keep the data you care about private

Gen AI tools are trained on vast datasets and continue learning from new data to improve. The problem is that these models might regurgitate training data in their responses, potentially exposing your private information to others. There's also a security risk when using web-based applications, as your data is sent to the cloud for processing.

Also: This new AI benchmark measures how much models lie

To maintain good AI hygiene, avoid sharing sensitive or personal data with LLMs. If you must use personal data, consider redacting it. Many AI tools, including ChatGPT, offer options to opt out of data collection, which is a wise choice even if you're not using sensitive information.

4. Watch how you talk about LLMs

The conversational nature of AI can lead users to overestimate its capabilities, sometimes attributing human traits to these systems. This anthropomorphism can be dangerous, as it might lead people to trust AI with more responsibility and data than they should.

Also: Why OpenAI's new AI agent tools could change how you code

To counteract this, Sap advises against describing AI models in human terms. Instead of saying, "the model thinks you want a balanced response," he suggests, "The model is designed to generate balanced responses based on its training data."

5. Think carefully about when to use LLMs

While LLMs seem versatile, they're not always the best solution for every task. Available benchmarks only cover a fraction of user interactions with LLMs.

Also: Even premium AI tools distort the news and fabricate links - these are the worst

Moreover, LLMs can exhibit biases, such as racism or Western-centric views, making them unsuitable for certain applications.

To use LLMs effectively, be thoughtful about their application. Evaluate whether an LLM is the right tool for your needs and choose the model best suited to your specific task.

• Want more stories about AI? Sign up for Innovation, our weekly newsletter.