Winners
Hall of Fame
AI development at a human level, aimed at improving society, necessitates a strong, widespread community. Addressing complex contemporary issues requires collective effort, especially given the urgency of global crises.
Therefore, community challenges are crucial for AI progress. Here, we celebrate Lab42 challenge winners in a Hall of Fame, showcasing their accomplishments and character as top achievers.
Lab42 Competitions:
Shared 1st Place
Jack Cole
December 22, 2023
together with:
Mohamed Osman
"Many were skeptical of my LLM approach"
Jack Cole, a tenacious contender in the world of AI, is driven by a desire to advance human-level AI to address the world's pressing challenges. His passion for tackling difficult problems led him to the field of AI, where he thrives in its complexities and challenges.
Winning the ARCathon 2023, Jack credits his success to unwavering perseverance, especially when facing discouraging outcomes. He recounts enduring a prolonged period with no significant progress: "I went probably 6 months or more with a score of 0 on the private test set." His belief in the potential of an LLM approach, despite skepticism, underscores the importance of persistence and faith in one's ideas. Jack also highlights the value of community engagement, emphasizing the significance of sharing ideas and receiving feedback from others working on ARC.
Discussing the journey to solving a substantial portion of ARC, Jack believes achieving scores above 50% is feasible with enhanced strategies like synthetic data generation, data augmentation, and improved NLP training across various task types. He also sees potential in refining test-time tuning procedures, retriever-based systems, and iterative-improvement multi-model systems.
Currently, Jack is dedicated to pushing the boundaries of ARC performance while exploring a broader spectrum of tasks. His projects are focused on understanding generalization, meta-learning, and the limits of transformer models. With the ambitious goal of surpassing the ensemble approach using a single NLP model, Jack is committed to advancing model capabilities, stating, "I will continue training the existing models with an aim towards beating the ensemble approach with a single NLP model." His current best performance stands at 23% on Kaggle, showcasing his relentless pursuit of AI development and innovation.
Shared 1st Place
Somayyeh Gholami
December 22, 2023
together with:
Mehran Kazeminia
"The Synergy of diverse Methodologies"
Somayyeh Gholami, an ardent enthusiast of AI, is driven by her passion for leveraging technology to address significant global challenges, including healthcare, climate change, and inequality. Her interest in ARC stems from a fascination with understanding the initial steps of machine abstraction and reasoning, motivated by the potential impact of AI on people's lives.
As the proud winner of the competitive ARCathon 2023, Somayyeh attributes her success to the innovative fusion of diverse approaches and methodologies. She emphasizes the value of integrating individual techniques, which, when combined, yielded superior results. Another critical strategy was thematic task division, focusing efforts on specific themes for more efficient solutions. She advises the community to embrace a variety of approaches and tailor them to specific task clusters, as this could unveil unique pathways to conquering the ARC challenge. "The synergy of diverse methodologies was a key factor in winning the ARCathon 2023."
Discussing the path to solving 50-80% of ARC, Somayyeh believes the key lies in a three-stage approach involving hypothesis generation, testing, and application. She notes that while humans excel at the initial understanding stage, machines potentially outperform in the testing and application stages due to their computational power. The implementation of around 500 to 800 fundamental functions, such as image rotation, resizing, color alteration, and more, could enable machines to handle tasks comparable to human intelligence. She envisions that once these basic keys are established, machine learning will evolve to generate hypotheses and compile lists of fundamental functions for each new task, paving the future for ARC advancements.
1st Place
Santosh Thoduka
December 17
"Simple Behaviors, Bigger Goals: Fascination of Swarm Intelligence"
Santosh Thoduka finds the intersection of programming and real-world application through his studies and research in robotics particularly engaging. His interest was piqued by the Swarm Intelligence Cup, seeing it as an intriguing programming challenge. Thoduka is fascinated by the concept of simple behaviors of multiple agents accomplishing larger goals. He is optimistic about the future of multi-robot systems, especially with recent advancements in perception and locomotion, and believes that events like the Swarm Intelligence Cup foster bottom-up intelligence.
Currently, his work focuses on computer vision for robotics, specifically enabling robots to recognize errors during task execution. While acknowledging the current progress in learning-based AI approaches, Thoduka anticipates that the future of AI will likely involve a balance between learning and knowledge/logic-based reasoning.
1st Place
David Gamez
31. March 2023
"A new Generation of intelligent Machines"
David Gamez, Senior Lecturer at the Department of Computer Science at Middlesex University, explains his draw to the AI field: "Artificial intelligence is an opportunity to model natural thinking and intelligence, which can help us to understand ourselves. It is also one of the most challenging, fun, and interesting areas of computer science!"
His recent accolade, winning the Essay Competition on Virtual Worlds, was no coincidence. Gamez reveals: "I entered the competition because the title question was closely aligned with my research […]". Currently, Gamez's research is focused on a groundbreaking project: "I am currently working on a new universal measure of intelligence, which will enable us to compare the intelligence of humans, animals, and AIs on a single scale," he shares. His vision for AI's future is the creation of large multimodal models, "In the medium term, I think that people will build better large multimodal models that work across natural language, video, sound, emotions, and other inputs. When these models are embodied in robots and connected to sensors, we will see a new generation of intelligent cognitive machines".
As for human-level AI, Gamez holds a realistic view: "Right now, we have systems that are more intelligent than humans in the environments that they are designed for. For example, AIs are more intelligent than humans in the game of Go. In my opinion, we are a long way from AIs that can match human intelligence in our natural and social world."
1st Place
Simon Ouellette
31. January 2023
"Don't follow the Hype"
After years of programming and consulting in industrial optimization, financial services and business information management in Montréal and New York, Simon began his AI career. His expertise and experience led him to AI-powered agricultural technology and financial trading.
With such a multifaceted background, it was no surprise that Simon earned first place in the Essay Challenge. His interest in truly understanding intelligence, "one of the most important mysteries of the universe," and his skepticism of current hype in the AI field led him to thoroughly research and consider the principles of intelligence. In Simon's view, today's AI does not exhibit intelligence, so we need to address the big problems first.
Simon is now focusing his efforts on building a virtual world that allows algorithms to learn abstract reasoning and ARC solving. This endeavor includes both ARCathon 2023 and our latest Essay Challenge on Virtual Worlds.
1st Place
Michael Hodel
13. January, 2023
"ARC is rather addictive"
Michael is a local talent from Zurich, Switzerland, pursuing his Master’s degree in Computer Science at ETH. What brought him into AI is a mix of his fascination for technological advancement, his liking for science fiction, and his interest in mathematics and programming. He believes that much more advanced AI will be the most significant invention ever.
We asked him what factors made him win the competitive ARCathon 2022: “My programming experience was definitely very useful when working on my solution approach for ARC. However, I consider my time investment to have been the biggest driver, something that, fortunately enough, came easy due to finding ARC to be a very fascinating problem and great fun to work on. Even though it takes some work to get off the ground, I can only encourage anyone curious to dive into it, it is rather addictive, and the odds are great that the experience will be joyful.”
In that spirit, Michael is currently focused on continuing to work on ARC alongside his studies at ETH, hoping to contribute to eventually solving it. He is looking forward to participating in the ARCathon 2023 and curious to see what further approaches the community will come up with.
For Humankind
Villa Fontana Obere Strasse 22B 7270 Davos, Switzerland
Powered by 