AWS announced a new tool they created called PartyRock which allows a next-to-no experience way to play with Generative AI in ways that create actual, working applications.
Recently, I got to test it it and wanted to provide my thoughts on this new tool. So let's start from the top and work our way down to what this is about.
What is Generative AI
Generative AI, at it's core, is the ability to use complex machine learning techniques to create, or generate, content such as words, audio, or pictures. Behind the scenes, models are created based on what are known as Large Language Models, or LLMs, that create potential interaction between given words, ideas, and phrases.
Based on how those models work and tweaks made by the data scientists that curate them, they can make logical connections. These models are often trained on huge sets of data and phrases, sometimes entire subreddits or Github documentation. The amount of data used in public GenAI LLMs (such as ChatGPT, Google Bard, or Claude) is expansive and the models are curated by some of the best data scientists out there. The continual improvement of the models is important to the continued growth of these GenAI platforms.
Getting into using GenAI is pretty easy - you can jump on ChatGPT or Bard or Claude and start asking questions or making requests. Feel free to go ask it to write a pop song about a ham sandwich or tell you about Peter the Great like you're a 3rd grader. Those are, however, premade chat tools. What if you wanted to make your own tools using GenAI?
That's where PartyRock comes in.
PartyRock takes a prompt and creates an application with a simple front end GUI. For example:
Then clicking generate and waiting about a minute or two, you get a full fledged app that runs on Amazon Bedrock
To get to it, first go to PartyRock, create and account and login. There you can create your own apps from descriptions.
If you want, you can start by looking at the app I built for this post - https://partyrock.aws/u/mjh/a4MH28e1E/Blog-Post-Generator
I actually used this app I built to create this blog post's outline and generate the cover image. Since it can create images and text, I figured this was a good example of how to make it work well.
When you go to run it, you simply add in the topic and it does the rest
This includes creating a picture that you can download
and the outline, that you can copy somewhere else
The example one here:
Here is a detailed outline for a blog post on how spices are impacted by where you are from: Introduction - Spices add flavor and aroma to dishes, but the spices used vary greatly around the world. The spices commonly found in cuisine are largely influenced by geography and culture. Body - Herbs and Spices by Region - Asia - Common spices: ginger, garlic, sesame, chiles, Szechuan pepper, star anise, cinnamon, cardamom, cumin, turmeric - Cuisines that use these spices heavily: Chinese, Indian, Thai, Vietnamese - Europe - Common spices: parsley, thyme, sage, rosemary, oregano, bay leaf, fennel, paprika - Cuisines that use these spices: French, Italian, Spanish, Greek - Middle East - Common spices: cumin, cinnamon, allspice, sumac, saffron, mint - Cuisines that use these spices heavily: Moroccan, Persian, Turkish - Africa - Common spices: piri piri, berbere, ras el hanout, dukkah - Cuisines that use these spices heavily: North African, Ethiopian - Americas - Common spices: chiles, cumin, coriander, cinnamon, allspice, vanilla - Cuisines that use these spices heavily: Mexican, Caribbean, South American - How Geography and Climate Impact Spice Availability - Tropical climates allow cultivation of spices like pepper, cinnamon, ginger - Arid climates enable growth of spices like cumin, coriander, and saffron - Trade enabled greater spice availability and integration into cuisines over time - How Culture and History Shape Spice Usage - Trading along spice routes introduced new spices into cuisines - Immigrant communities bring spice preferences with them - Traditional dishes dictate certain spice combinations Conclusion - Geography and climate where you live limits spice availability, while culture and history shape spice preferences in cuisine. This leads to distinct regional spice profiles around the world. Understanding these differences provides insight into culture and cuisine.
In a given box, you can actually tweak exactly how it works. For instance, if I wanted to modify my Outline box, I'd click the little switch button
and get a box like this:
From here, I could change my model type, for instance, from Claude v2 to Claude Instant v1 or AWS Titan Text Express
I'll leave mine on Claude v2 because after a bit of experimenting, I like the outputs best for that. However, you should absolutely experiment and make sure the outputs are good for you.
Does it work?
Well, I used it to write some of this blog post - I ignored most of the outline, but I used the cover image it generated. It did give a good outline, however, I am not one to write to outlines well.
Why use it?
It's good practice. You aren't going to build you enterprise application on PartyRock but it's a good way to test things like how Temperature works with a given model. It also gives a sense of the true power of it, even if you are using it to generate bedtime stories or give you example IaC code.
Overall, this is a fun playground tool to experience GenAI without worrying about the extra pieces necessary to make it work. It's a very good stepping stone into the world of GenAI and will get your brain going on all the possibilities!