Limitations of AI
Using Artificial Intelligence for content writing is not perfect and it is important to talk about the limitations of AI. We often get asked by users for certain features which are simply not possible so it makes sense to go through some of the limitations so you can better understand what is and what isn't possible and also the limitations of AI.
The AI performs remarkably well for text generation. It will be able to see the information you have given it and predict the next characters in the sequence. This is the main purpose of this AI technology and what it excels at. The AI cannot see the contents of a website from a URL. It cannot look at images and understand them. It cannot create images itself. There are of course other methods and ways to get this sort of information and to do these sort of capabilities outside of using the AI but for the most part, it would involve a custom solution.
The way that the AI works with predicting the next likely token in a sequence means that it is not trained to recite facts as it is merely trying to predict the next token. The way that some of the models are trained with a higher degree of wildness than others makes it so that facts are less likely to come out in any coherent sense. Expect wild outputs and always expect to fact check anything that an AI outputs. It just makes sense to do so if you are sharing the output with others to avoid the discrepancies that can happen.
All the NLP models are trained up to a specific point in time. Anything past this point in time is unknown. A good example of this is to ask any NLP model about covid-19. You are likely to get a nonsense answer. The model was not trained on data which has mentions of covid-19 and therefore you can't expect it to be an expert. OpenAI's GPT-3 for example has a cut off date of October 2019. This is always worth keeping in mind when you are using AI for content generation. It will not have knowledge of current events. It will not know the price of bitcoin. It will answer you of course, but it will give you what it thinks the answer should be and not necessarily the truth.
Last modified 2yr ago