🛑Limitations

Limitations of generative AI in developing applications

The limitations in this section can be generally applied to all LLMs.

The following limitations may be found in various generative tools:

• Ambiguity

• Truth

  • Time Accuracy

  • Scope

• Visualizations

Visualization limitations are specific to large language models in generative AI. However, large language models can be augmented with machine learning image analysis or generative image tooling. For example, As of GPT-4, the platform supporst the ability to consume and understand images and output responses from it.

As ChatGPT is this most capable and accessible in the LLM space, this course focuses on ChatGPT so we will specifically discuss the limitations and weaknesses of this platform, which may or may not be the same as other platforms.

AI can be wrong about programming concepts. In the same way GPT is not completely accurate with all code generation, we should be critical of the code generated and how it is implemented. Always thoroughly review them for accuracy and cognitive complexity. This will help build competency up in various patterns and concepts.

These limitations are far from prohibitive, but are something that must be considered during the development process. Having a good understanding of these limitations provides developers focus areas and context that make the development process and application more efficient.