5 Expert Tips for Smart and Safe Use of Generative AI

The Rise of Generative AI and Its Challenges

As generative AI becomes increasingly ubiquitous in our daily lives, from Google searches to chatbots, the need for understanding its limitations and using it safely has never been more critical. At the recent South by Southwest (SXSW) conference, Carnegie Mellon University computer science professors Maarten Sap and Sherry Tongshuang Wu shared valuable insights on how to navigate the world of AI tools effectively 1

1

2

2

.

Be Specific and Structured in Your Prompts

One of the key challenges with AI models is their tendency to misinterpret vague or ambiguous prompts. Large Language Models (LLMs) struggle with non-literal language, often taking statements at face value more than 50% of the time 1

1

. To combat this, users should:

1. Be explicit in their instructions
2. Provide clear context
3. Avoid assuming the AI will extrapolate meaning

As Wu suggests, "Focus on what exactly you want, and don't assume the LLM will extrapolate your actual question" 1

1

.

Verify AI-Generated Information

Perhaps the most significant concern with generative AI is its propensity for "hallucinations" – generating false or inaccurate information. These hallucinations can occur up to 25% of the time in general use cases and even more frequently in specialized domains 2

2

. To mitigate this risk:

1. Cross-check AI responses with external sources
2. Ask the same question multiple times or in different ways
3. Be especially cautious when the AI expresses high confidence

"Make conscious decisions about when to rely on a model and when not to," advises Wu. "Do not trust a model when it tells you it is very confident" 1

1

.

Protect Your Privacy

AI models often use their outputs for further training, which can lead to privacy concerns. Users should be cautious about sharing sensitive information with these systems. Sap demonstrated that some AI models struggle with information boundaries, potentially revealing private details inappropriately 1

1

.

To maintain privacy:

1. Avoid sharing sensitive or personal data with AI tools
2. Opt out of data collection when possible
3. Redact personal information in prompts when seeking assistance 2

   2

Avoid Anthropomorphizing AI

While AI's ability to mimic human conversation is impressive, it's crucial to remember that these systems are not human and lack true cognition. Sap warns against attributing human-like qualities to AI, as it can lead to misplaced trust and reinforce social stereotypes 1

1

.

To maintain a healthy perspective:

1. Use language that doesn't imply cognition when referring to AI
2. Remember that AI's human-like responses are based on training, not genuine understanding
3. Be aware of the potential for over-attributing human-likeness to AI systems 2

   2

Understand AI's Limitations

Despite claims of advanced capabilities, current AI models have significant limitations. Benchmarks suggesting human-level performance don't always translate to real-world applications. Sap cautions against making rash decisions based on inflated perceptions of AI capabilities 1

1

.

When considering AI use:

1. Evaluate the benefits and potential harms of using or not using AI for a specific task
2. Recognize that AI may not be suitable for all use cases
3. Stay informed about the current state of AI technology and its limitations 2

   2

By following these expert tips, users can harness the power of generative AI more effectively while maintaining safety and accuracy in their interactions with these increasingly prevalent tools.