Generative AI models such as ChatGPT are part of an ongoing evolution of AI capability. The development of natural language processing (NLP) began simply with what is known as a “bag of words.” A paragraph or block of text is tokenized, which means breaking the content into linguistic units. A count is calculated for each word. If the word “sad” appears ten times and the word “happy” appears once, then the content has a negative sentiment. A subsequent development was the ability to identify parts of speech, known as parts of speech tagging. This became essential for language translation. Understanding nouns, verbs, adjectives, and adverbs provides a way to interpret a sentence.
Recurrent neural network (RNN) algorithms allow software to evaluate a sentence from right to left and left to right and remember sequential dependencies. A recent software development is transformers. A generative pre-trained transformer (GPT) adds self-attention capability, which is identifying keywords or phrases. Parallel processing is also included and performs faster analysis. The pretraining part of GPT means it accesses a volume of content that has already been analyzed. At the core of a GPT model, such as ChatGPT, are machine learning models that predict the next word or phrase and determine how to respond to a query.
Many people may find AI capability a mystery, but it is based on statistical models and highly evolving algorithms that take advantage of logical sequences. ChatGPT does not provide perfect answers. Understanding how to interact with the software improves the accuracy and relevance of responses, which is why prompt engineering is becoming an important skill for project managers.
Community Champion
