Some areas that could be improved with ChatGPT include the information they provide, as sometimes they hallucinate something that's not real. Sometimes they comment positively without being asked, and I have questioned ChatGPT that I don't want the prompt to agree with me but rather to challenge me. If I provide a slide and ask to compare it to McKinsey or other big companies and scale it from zero to 10, they might say it's an eight and suggest improvements to reach nine or ten. However, these evaluations aren't always realistic because if you're not a specialist on the subject, you might create something that's not accurate. This becomes problematic because we are working with real clients, not experiments. When working on real cases, I have to be the one to validate that the information is accurate. Sometimes they provide information that isn't real. You have to be the gate to validate that this information is correct before proceeding. For simpler tasks adjusting communication style or making text more formal, they perform perfectly. However, for complex cases, I don't use ChatGPT as the source of truth because it's not always accurate. If you delegate it to make decisions in your place, it could create significant problems to solve.
