Generative AI Model or Tool?

It is important to understand the difference between generative AI ‘models’ and ‘tools’.

Models require technical expertise to interact with and are often used by researchers or developers. They represent the raw capability of AI to learn and generate content. The model is the neural network, the artificial intelligence ‘engine‘.

The most common large language models are:

• ‘GPT-3.5’, released in November 2022, developed by OpenAI.
• ‘GPT-4’, also from OpenAI, was released in March 2023 and is more powerful than GPT-3.5.
• ‘GPT-4o’, also from OpenAI, was released in May 2024 and is more powerful than GPT-3.5. Free tier GPT-4o is available to all users on a limited basis.
• ‘Pro 1.0’ is Google’s standard model, released in February 2024.
• Other models that take in text and produce images are based on ‘diffusion’ models e.g., OpenAI’s DALL-E .
• There are other models that can produce other outputs such as video e.g. OpenAI’s Sora – a text to video tool

Tools are designed for end-users, often with intuitive interfaces. They translate the model’s capabilities into practical applications. For example, a writing assistance tool using a generative AI model can help students create and refine written content. The tool or chatbot interface is the software that you use. For example:

• ChatGPT Free tier (tool) uses the GPT 4o model.
• ChatGPT Plus (tool) gives access to the GPT-4o model with increased functionalities for paying customers.
• Microsoft Copilot (formerly known as Bing Chat) (tool) uses the GPT-4 model when put in ‘creative mode’, and uses the GPT-3.5 model otherwise, along with other models.
• Google Gemini (formerly known as Bard) (tool) currently uses Pro 1.0.
• DALL-E (tool) uses a diffusion model to generate images.

Attribution: Adapted from AI in Education University of Sydney licensed under a Creative Commons BY-NC 4.0 licence

Page updated 20th May 2024