Difficulty of control

- Another potential problem with generative AI models is that they're difficult to control and the results are difficult to predict and, in some cases, verify. Now, it's often easy to tell when AI gets an image wrong. Hands and fingers just aren't quite right, or some unfortunate melting faces. But AI-generated text is another story. The grammar might be perfect and the sentences might be well-constructed, but the message itself might contain errors or other problems. Here's why. First, generative AI models are complex and nonlinear, using intricate algorithms and neural networks to learn patterns and relationships to create something new. Because of this, it can be difficult to understand how the different parts of the model interact with one another. This may result in unexpected or illogical results, all while the model presents itself as an all-knowing source of truth. Second, generative AI models are trained from massive datasets that are imperfect. These datasets come from a wide variety of disparate sources, and it's difficult to know how this data will interact with one another. Also, as we know, this data is imperfect. It may include biased, inaccurate, or otherwise problematic input. All of this can lead to unexpected, confusing, or illogical output. Third, generative AI models are designed to be stochastic. This means they generate data that follows a probability distribution, predicting the likelihood of something happening rather than predicting something with certainty. This means the model can generate different outputs from the same inputs. Contrast this with deterministic modeling, which generates data that is always consistent. The plus side of stochastic modeling is that it's a dynamic way to generate creative content, but with that comes the difficulty of control and of prediction. Finally, generative AI models were built to learn from feedback and adapt their output accordingly. Because of this, they're constantly evolving and adapting based on new data and feedback. This also can make the results difficult to predict. Given all of these factors, it's important for each of us to just be aware of these issues. Remain curious, but do not place your trust in generative AI output, particularly text output. Fact-check AI's work or use it to inspire your own original work. For example, if you ask ChatGPT to write a history of your company in a friendly, informal style, use the results as a guide for your own version, substituting in facts such as dates and key milestones with ones that you've gathered with solid research and reputable sources. Current AI models are often an inspiring source of creativity, but they're not always great with facts. Keep that in mind, and you'll be better positioned to take advantage of them.