Tianqiao & Chrissy Chen Institute’s OMNE Framework Achieves Top Ranking on GAIA Leaderboard

Leveraging extensive insights into brain function and memory, the internal AI team at the Tianqiao & Chrissy Chen Institute (TCCI) has made significant strides in artificial intelligence with their OMNE Multiagent Framework. This innovative framework has secured the leading position on the GAIA (General AI Assistants) benchmark leaderboard, co-launched by Meta AI, Hugging Face, and AutoGPT from Hugging Face. OMNE surpassed offerings from esteemed organizations such as Microsoft Research, marking a notable achievement rooted in years of TCCI’s brain research. By equipping agents with Long-Term Memory (LTM) capabilities, OMNE enhances the depth of thinking and decision-making processes of Large Language Models (LLMs) in tackling complex problems.

This accomplishment represents a key milestone for TCCI’s AI team, following the institute’s founder, former tech executive Tianqiao Chen, announcing the “All-In AI Strategy” last year.

Currently, OMNE boasts a remarkable success rate of 40.53%, outpacing submissions from major players like Meta, Microsoft, Hugging Face, Princeton University, the University of Hong Kong, the British AI Safety Research Institute, and Baichuan. In comparison, GPT-4 with plugins achieved a success rate of just 15%. GAIA is recognized as one of the most challenging datasets for multi-agent intelligence, and achieving the top spot on its leaderboard underscores TCCI’s deep expertise and commitment to innovation.

OMNE operates as a multi-agent collaboration framework grounded in long-term memory. Each agent within this framework shares an independent system structure and can autonomously learn and comprehend a comprehensive world model, thus gaining an independent understanding of its environment. This LTM-based collaboration system allows the AI to adapt in real-time to behavioral changes, optimize task planning and execution, and foster personalized and efficient self-evolution.

The breakthrough achieved with OMNE integrates a long-term memory mechanism that significantly narrows the search space of Monte Carlo Tree Search (MCTS), enhancing decision-making abilities for complex challenges. By incorporating more efficient logical reasoning, OMNE not only elevates the intelligence of individual agents but also considerably boosts the capabilities of the multi-agent system through optimized collaboration mechanisms. This advancement draws inspiration from the columnar structure of the human cerebral cortex, which serves as a fundamental unit for cognitive and behavioral functions. By enhancing collaboration among individual intelligences and agents, the AI model can gradually develop cognitive abilities, construct internal representation models, and ultimately elevate the system’s overall intelligence.

“We are extremely proud to see OMNE take the top position on the GAIA leaderboard,” remarked the head of TCCI’s AI team. “This achievement showcases the immense potential of leveraging long-term memory to facilitate AI self-evolution and address real-world challenges. We believe that advancing research in Long-Term Memory and AI self-evolution is vital for the continued development and practical application of AI technologies.”

Founded in 2016 by Tianqiao Chen and Chrissy Luo with a commitment of $1 billion, the Tianqiao & Chrissy Chen Institute aims to advance brain science. The institute’s vision is to enhance human experiences by understanding how our brains perceive, learn, and interact with the world. With a strong emphasis on artificial intelligence, the organization seeks to accelerate the pace of scientific research.

SEE ALSO: ByteDance Plans to Establish An AI R＆D Center in Europe