10 questions to ask vendors claiming to use generative AI

- So what's the catch? Do all of the benefits and use cases of GenAI and talent acquisitions sound too good to be true? Well, there is a catch, at least five of them. Let me walk you through these and provide you with 10 key questions to ask vendors and internal development teams who are leveraging GenAI. First, you should never use GenAI solutions for anything that you would want to claim copyright on. Copyright protection only applies to works created by people. Second, there's no guarantee of privacy or security with public GenAI solutions like ChatGPT, so you should never enter any personal candidate or sensitive or confidential company information. Third, you may have heard that large language models can hallucinate. That may sound funny, but it does mean that Generative AI can sometimes make up information. For creative content like outreach messaging where accuracy isn't important, this isn't really an issue. But if you tried using chat GPT for something like compensation research and analysis, you cannot implicitly trust its output. Fourth, solutions using large language models like OpenAI's ChatGPT, Google's Lambda, or Meta's Llama can pick up on biases if trained on historical data or other data that may have embedded biases around race, gender, and other demographic factors. This can lead to biased output such as gendered language in job descriptions and postings. Fifth, while LLM-based GenAI solutions can produce human level text, they can often produce bland or average output. Given the aforementioned issues, it's critical that the output of any GenAI solution is reviewed for accuracy, bias, and quality before making use of it. Now when it comes to working with third party or internal developers of GenAI solutions, it's important to ask them some tough questions so you can make informed decisions. One, do you use our data to train your general model for others to benefit from the learnings? Two, how can your solution be customized for our company? Three, how do you train your solution for our specific use cases? Four, how do you control for hallucinations? Five, how do you make sure output is accurate? Six, how do you support explainability? Seven, how do you make sure output is fair and avoids bias? Eight, how does the model learn over time? Nine, how do you ensure you're compliant with all applicable AI, automation, and data regulations? And 10, can you provide case studies or references from similar clients or users? As you can see, there are some issues and concerns associated with GenAI solutions. While the opportunities presented by GenAI are exciting, keep in mind the limitations and risks and be sure to ask the tough questions of solution providers to make sure you're confident that the risks are appropriately mitigated.