Introduction
In starting my project, which heavily involves AI, it was really important to look at all the different AI options available. I spent time comparing different AI language and image models, as well as the services that offer them. The main goal was to find the best ones for my project's needs and for iO.
Abstract
In my research I compared popular language models like GPT-4, LLaMA 2 70B, Code LLaMA, and more. I checked how well they understood different themes and HTML/CSS, as well as how fast they were, how much they cost, and if they can be used commercially. I also compared popular image models like DALL-E 3, Stable Diffusion XL, OpenJourney, and more, looking at the same factors.
I also looked at AI service providers like OpenAI, Azure, AWS, and more, to see what their limits are, how they protect data, and how secure they are. Using all this information, I made a simple and clear overview showing notable points of the different models and services.
Conclusion
From my research, I created the following overview showing abnormalities in the models and services.
After discussing with the Technology Director, I went with GPT-4 and DALL-E 3, and chose Azure as the service provider for the project. Both GPT-4 and DALL-E 3 did well in all the tests, and they're both from OpenAI, a company that takes privacy and security very seriously. These things are important to iO too, as they have several ISO certificates. Azure was chosen because it passed all the tests, it's big on privacy and security, and it's already used within iO.