Welcome to our Day 2 recap of the Advent of GenAI Hackathon! This event, hosted on the Intel Developer Cloud (IDC) and powered by Intel GPUs, brings together the brightest minds in AI - from developers to startup enthusiasts and students. Focused on exploring Generative AI and LLMs, the hackathon challenges participants with daily tasks, offering a unique combination of learning, competition, and creativity. Let's dive into the highlights and accomplishments of this exciting event.
Challenge 2: Creative Storytelling using LLMs
On Day 2, December 6, 2023, of the Advent of GenAI Hackathon, participants tackled the task of developing a chatbot using models from the Intel Developer Cloud (OpenLlama, Camel, Zephyr, Intel Neural Chat, Llama2 variants) or the Prediction Guard’s LLM API. The objective was to creatively tell a story through these chatbots, with the evaluation focusing on session transcripts showcasing the chatbot's responsiveness and creativity.
Key Learnings:
- Prompt Engineering for LLMs: Participants learned to craft effective prompts for Large Language Models.
- Fine-tuning: They explored fine-tuning LLMs to enhance chatbot performance for specific storytelling tasks.
- Utilizing LLM API Services: The use of Prediction Guard’s LLM API services was a critical part of the challenge, focusing on its efficient application.
- Memory Management in Chat Interfaces: Managing memory in chat interfaces was a crucial skill, as highlighted in the Simple LLM Inference notebook.
Approach and Techniques:
Participants developed storylines for their chatbots, ranging from scripted narratives to interactive sessions with creative user responses. Emphasis was on model efficiency.
Participants using local models referred to the LLM fine-tuning notebook, particularly for the 4-bit model loading method via Intel's BigDL library. Those using Prediction Guard’s LLM APIs focused on efficient model loading and application. The chatbots showcased creativity in storytelling and adeptness in handling unexpected inputs while staying true to the story's theme or character.
The challenge illustrated the trade-off between the detailed customization of fine-tuning and the quicker, creative demands of prompt engineering.
Documentation and Insights:
Participants documented their development process, including model choice, storyline creation, chatbot responses, and faced challenges. The session transcripts they provided highlighted the chatbots' storytelling abilities and responsiveness.
The Top Submissions of Challenge 2
A big congratulations to all the participants! The community cast their votes, and it was a time of celebration for the top submissions. The skill demonstrated in weaving compelling narratives through chatbots was truly impressive. Here's a spotlight on the exceptional storyteller bots that stood out:
Story Maker Agent by Samy Gauquelin
Samy Gauquelin created a custom Agent that is configured to write stories. The Agent also utilizes a custom tool that he developed to enhance its creativity when required.
Samy Gauquelin reflected on their hackathon: "Challenge 2 focused on the LLM's creativity in a chat. I aimed to create an interactive solution, emulating that process. I envisioned an agent using a tool to boost its creativity. After establishing the architecture, I delved into the creative tool's details. The tool had two phases. First, I explored the subject by asking questions for a deeper understanding. The second phase used this insight to generate ideas. These ideas, processed by the global agent, formed the final response. I also added a memory system for users to craft their own stories during interactions."
Pen and Paper Chatbot by Vincent Müller
Vincent Müller chose to create a Pen and Paper Chatbot that guides a player through a fantasy role play game.
It is Vincent Müller's first attempt to visit a random place in the real world.
Here is an example of a fight in which Vincent Müller tried to use an atypical weapon for a fantasy role-playing game:
Vincent Müller commented on his project for the hackathon challenge: "I chose a simple approach, utilizing only prompt engineering and combining system and user messages. It was exciting to achieve usable results without the need for fine-tuning. The Prediction Guard API made it easy to employ different models, thanks to its clear syntax and straightforward setup."
Chat Bot by Aaryan Mudgal
Prompt:
Describe the abandoned factory and the efforts people make to revive it.
Output:
The story unfolds as the group arrives at the desolate factory. Upon entering, a sense of abandonment permeates the air, accompanied by the echoes of a bygone industrial era. The once-thriving facility now stands silent, its machinery rusted and production lines frozen in time. Undeterred by the daunting task, the team initiates their revitalization mission, tearing down dilapidated structures and removing outdated equipment. Yet, as they delve into the restoration process, they encounter unforeseen challenges, revealing the factory's eerie history.
Prompt:
As challenges arise, the team perseveres despite the odds.
Output:
Faced with unexpected hurdles and unforeseen complications, the team refuses to retreat from their mission. External pressures mount, and doubts creep in, but the determined individuals press on, driven by a collective desire to resurrect the factory. The resilience of the team becomes evident as they collaborate to overcome each obstacle, displaying unwavering commitment even in the face of adversity.
Prompt:
Share the surprising discoveries that motivate the team to persist.
Output:
Amid the labyrinthine corridors and hidden recesses, the team stumbles upon a forgotten archive room. Dust-covered blueprints and documents reveal the factory's illustrious past and the groundbreaking innovations once conceived within its walls. Inspired by the rich history and potential for innovation, the team redoubles its efforts. As they delve deeper into the archives, a cache of overlooked prototypes is uncovered, each holding the promise of transforming the factory into a hub of cutting-edge technology.
Prompt:
Despite setbacks, the team remains dedicated, working towards a brighter future for the factory.
Output:
Undeterred by setbacks and fueled by the newfound knowledge of the factory's illustrious past, the team forges ahead with renewed vigor. They implement innovative solutions, restoring the factory to its former glory while integrating modern technologies. The once-silent assembly lines hum back to life, echoing a revitalized spirit. The community witnesses the remarkable transformation, and the factory, once abandoned and forgotten, emerges as a beacon of industry and progress, showcasing the indomitable spirit of those who refused to let it fade into obscurity.
Aaryan Mudgal commented on the hackathon challenge: "I'm delighted that the story resonated! The narrative unfolds with mystery and intrigue, drawing readers into the eerie atmosphere of the abandoned factory. The team's perseverance in overcoming challenges adds determination and resilience, making the story engaging and inspiring. Surprising discoveries and the team's commitment to uncovering the factory's history create a compelling narrative arc. It's a tale of restoration, innovation, and the transformative power of collective effort. I hope you enjoyed the journey through the abandoned factory's revival!"
Honorable Mentions
We are excited to honor the additional outstanding projects by Aditya Krishna RS, Tomáš Barczi, Gaurav Chaudhari, Vishal Bankar, Johannes Widter, Kumari Pallavi, and Amit Sadhu.
The AI Odyssey Continues: Exciting Challenges Await!
Kudos to All Participants! Your submissions were nothing short of spectacular, displaying remarkable creativity and skill. Your talents and imagination shone brightly in each entry.
Keep fueling your creativity and inspiration as we continue to explore the wonders of AI together. The next hurdle is Challenge 3. Stay tuned for more exciting updates from our hackathon!
Are you an emerging AI developer or startup founder? Elevate your tech journey with the Intel® Liftoff Program for Startups.
I'm a proud team member of the Intel® Liftoff for Startups, an innovative, free virtual program dedicated to accelerating the growth of early-stage AI startups.