Achieving excellence means ensuring that the output meets or exceeds the standards that would be expected of a human creator by law, ethics, tradition, or industry requirements. Together, these principles form the bedrock of trust and value in GenAI-generated works.

This lesson delves into the critical importance of upholding originality and excellence, and the strategies you can employ to achieve them.

To ensure the originality and quality of GenAI creations, it’s essential to implement strategies that can verify the uniqueness and accuracy of the content produced. Here are some strategies you can use to improve content generated by GenAI:

• Prompt GenAI to provide references in its responses and then review those references yourself. Why? Because GenAI often lies. And, yes, it can totally make up references to support its lies, too. But also because that’s a good first step in spotting hints of plagiarism or copyright infringements.


WARNING

That’s why you never want to prompt a GenAI model to factcheck or check for plagiarism in its own responses and just accept its answer. Nor should you ever assume that you’re using a different model to check the first model’s work. You could easily be using two different applications built atop the same GenAI model.

• Use a different GenAI model to cross-reference generated content against existing works, flagging potential duplicates for review. For example, prompt Google Smart Search or Perplexity to factcheck the response you got from Claude or ChatGPT. Some GenAI search tools like Perplexity will automatically cite its sources for you to review, too. Make sure to stop and actually review the sources it gives you.
• Consider embedding digital watermarks or metadata within GenAI-generated content, creating a traceable digital footprint that confirms its origin and authenticity. See the deep dive below for more on digital watermarks and metadata.
• Know the data source. The development of GenAI models should be transparent and ethical, with clear documentation of the data sources and training methods used. By this, we mean that the maker of the GenAI model or application — either a vendor or your internal AI development team — should provide transparent information on how the model was trained and what data was used. This transparency not only builds trust but also allows for the scrutiny necessary to maintain the integrity of models in the rapidly evolving GenAI landscape. Discovering how the model was trained can help you estimate the risks in terms of the likelihood of originality and reliability of outputs.
  
  For example, the GPT models that power ChatGPT were trained on data scraped from the internet. That may mean a higher risk in the potential of plagiarism and copyright infringement than, say, a specialized model that AI scientists at your company trained only on company data.
• Consider asking your employer or GenAI vendors to use Causal AI to reveal how the GenAI model came to its decision (output) as a deeper check of its functioning within ethical guidelines. This is a highly technical strategy that your AI or IT department or GenAI vendor will need to do for you, as it involves developing causal inference algorithms that can work alongside or within GenAI models, creating training datasets that include causal relationship information, designing new model architectures that can combine causal reasoning with natural language generation, and developing metrics to manage the two. However, once in place, the Causal AI can reveal how the GenAI did what it did so you can better evaluate the originality and quality of its outputs.