Generative AI (or GenAI) is the hottest industry buzzword, and shows no sign of going away. At a high level, it allows you to use AI to generate new types of content including text and images. At it's heart are huge machine learning models, trained on vast amounts of data. For those who work in this space, explaining the differences between prompt engineering and prompt tuning, or describing zero, one-shot and few-shot inference, or even weighing up the merits of various tuning techniques may be normal. However, for most, this is a bewildering area and it is confusing to know where to start.
AWS has a history of coming up with fun ways to start a journey into complex topics. You only need to look at Deep Racer and the reinforcement machine learning topics it introduces. They have done it again for GenAI with PartyRock!
PartyRock is a new site from AWS that allows you to build an AI-generated app in a playground powered by Amazon Bedrock. With no AWS account to setup and manage, and no coding skills required, it is an amazing way to try out imaginative ideas, and see what GenAI could do for you.
Lacking in any imagination of my own, I enlisted the help of a 10 year old, and challenged them to build their first ever GenAI powered app.
There are three different ways of creating an app with PartyRock. You can build one from scratch starting with an empty app, you can remix an existing app, or you can start with a prompt. With the goal of creating a story around a medieval knight fighting different creatures in strange locations, we entered a few short sentences and clicked the button to generate our app
Within seconds, PartyRock had automatically generated the app using a number of widgets:
There are a total of 5 widgets displayed. The very top widget is there to display static text about the purpose of the app. The next two entitled Opponent and Location are for User Input. This allows someone using the app to enter information that can be used as prompts in other widgets. Finally, the story is created using a Text Generation widget, and the image is created using an Image Generation widget.
Once the app is created, it is simple to click into each widget and try changing the prompts to see what effect this has. On a Text Generation or Chatbot widget, you can also go into advanced settings to increase the temperature or Top P. These are actually inference parameters. Lowering the values results in more factual text, whilst increasing them results is more creative and imaginative text.
You also have the option to change the underlying model. This allows you to note any changes moving from Claude as the default LLM to Jurassic-2 or another available model.
This was all there was too it. After the first app, we moved onto look at language translation capabilities:
Next stop will be to try out the Chatbot capabilities
It's fair to see this got a huge thumbs-up. It was great to see curious and inquisitive minds amazed by how apps could be generated and images magically created from just a few words. If you want to have some fun exploring AI concepts, this is definitely a great place to start.