Welcome to the Current Capabilities and Limitations page. This will provide you with a starting point to explore further discussions on the capabilities and limitations of GenAI tools. 

GenAI tools continue to improve as they train which means that their capabilities and limitations will change over time and may vary from tool to tool depending on how it was trained. During this project, several key capabilities and limitations of ChatGPT emerged.

Please explore the links below to find specific information you are looking for on this page.

Capabilities

• Converses with the user and answers questions
• Produces easy to understand text in plain English and other languages (accuracy depends on language)
• Produces grammatically correct and generally well-structured text
• Summarises or distils key points from a text
• Solves math problems (not always accurate)
• Writes code
• Generates graphs/charts/images (sometimes a plug-in is needed, depending on the tool)
• Produces audio, video, and audiovisual content (dependent on tool)

Limitations

• Hallucinations (inaccurate or made-up information)
• Biases based on training data
• Produces surface-level or general information without expert prompting (does not think for itself)
• Does not always provide citations or references (or produces fake ones)
• Does not always respond to prompts accurately and can require several specific prompts to refine to the desired output
• Limited to training data (many tools are only trained on data up to a certain date so are not aware of up-to-date or specific information)
• Code produced is accurate but is the most straightforward and basic version of the code so it is not robust or secure and would not stand up in many real-world applications
• Images and video content can appear unnatural

Remember, GenAI is a tool. What it produces, and the accuracy of this content, is largely dependent on how you critically engage with it or direct it.