For my Deep Learning Application class project, I decided to blend humor with cutting-edge AI by building a roastmaster bear powered by large language models (LLMs). The intersection of humor and AI presents a fascinating challenge—can a machine understand what makes a joke funny, and better yet, can it become the ultimate roastmaster?
Table of Contents
Technical Pieces
Building the roastmaster bear required several key components, each contributing to the bear’s comedic performance. A microphone connected to a Raspberry Pi served as the input for user prompts, with speech-to-text technology converting spoken words into text for processing. To enhance the bear’s roasting capabilities, I employed prompt hacking techniques to bypass the safety restrictions built into LLMs, allowing for more creative (and cutting) responses. To give the bear a familiar voice, I used voice cloning technology to emulate a well-known Hollywood bear, adding authenticity to the roast. The final punchline was delivered through a speaker, also controlled by the Raspberry Pi, bringing the roastmaster bear to life in full auditory glory.
Outcome
The result of all this effort? Pure comedic gold. The roastmaster bear kicked off its performance by roasting a particular student’s grade, quipping that the only thing sadder was their love life. The class erupted into laughter—so much so that the hilarity unexpectedly lasted for over three minutes before the bear could deliver its next punchline. With every joke, the noise from the laughter filled the room, leaving the bear struggling to hear proper inputs from the microphone. In an effort to keep the joke going, I motioned to another student to quickly pull the plug on the bear in order to end the wildness, which only increased the classroom’s energy. By the end, the roast bear had made such an impression that one student even offered to buy it, cementing its status as the unexpected star of the classroom. I couldn’t have been happier when even the professor made a glowing mention of the project to round out the semester highlights during her final lecture.