Wasalu Jaco, who you might know as Lupe Fiasco, has been lending his artistic expertise to the realm of artificial intelligence. As both a rapper and an academic, Fiasco delves into the transformative influence of technology on music, particularly rap, in a recentvideo on Google’s YouTube channel. His focus isn’t on a system that generates lyrics autonomously, but one that mirrors his creative process, prompting users to view any word or phrase from many different perspectives. This pursuit led to a fruitful collaboration with Google and the PaLM 2 API, giving birth to TextFX, a suite of AI-powered writing tools for rappers.
TextFX isthe brainchild of Google Lab Sessions, an initiative dedicated to exploring how AI can amplify human creativity. To develop this toolkit, the team began by studying Lupe’s lyrical and linguistic techniques. They were fascinated by his knack for dismantling language and putting it back together in novel ways, often crafting phrases that sound identical to a given word but carry different meanings. For instance, “expressway” could morph into “express whey”, possibly implying speedy delivery of a dairy byproduct.
This led the Google team to contemplate how AI can aid Lupe, and others, to explore the creative possibilities inherent in text and language. They turned to large language models such as Text Bison for their language-related applications, especially their capacity for text generation and few-shot learning, allowing them to train the LLM on a small set of examples.
A crucial step was constructing prompts that primed the LLM to behave in specific ways. After iterations, they arrived at a formula for a prompt: first, describe the linguistic operation, followed by examples, and end with an incomplete input-output pair, awaiting the user’s input.
The end product, TextFX, houses an array of prompts designed to provoke creative possibilities from any given word, phrase, or concept. These include techniques like simile creation, word explosion into similar-sounding phrases, and unexpected scene generation. Users can interactively build and experiment with LLM prompts viaMakerSuite, a platform specifically designed for this purpose.
The team behind TextFX believes this experiment underlines the potential of AI to boost human creativity rather than supplant it. This exploration extends beyond creative writing, prompting contemplation onthe collaborative roles AI can take in other creative domains. The importance of creators in imagining these collaborations is highlighted by Fiasco’s own involvement and the invaluable perspective he brought to the table.
You can try TextFX for yourself attextfx.withgoogle.com. If you’re interested in the behind-the-scenes of how TextFX was built, the code has been open-sourced and isavailable for perusal.
