AAAI 2026 Bowling Happy Hour @ 2077AI Foundation
🎳 AAAI 2026 Bowling Happy Hour Hosted by 2077AI Foundation Community！
Join us for a bowling happy hour designed for top-tier AI researchers, including professors and PhD students from leading laboratories. This event blends relaxed social interaction with deep technical conversations—right on the bowling lanes.
📅 Date: Saturday, Jan. 24
🕒 Time: 4:30 – 7:30 PM
📍 Location: 5-min walk from Merlion
(with world’s most durable high performance bowling lane system)
To foster meaningful exchange, bowling lanes will be organized by research themes, allowing participants from different backgrounds to freely rotate, connect, and engage in focused discussions throughout the event.
😋 Beautiful food and unlimited drinks will be served. Come inspired and enjoy! 🍖🥩🍻🍹
🌟 About 2077AI Foundation
2077AI Foundation is a 501(c)(3) non-profit, open-source research organization led by world-class AI scientists and researchers. Over the past year, 2077AI has grown into a global research ecosystem, connecting researchers, developers, and engineers from leading universities and frontier organizations.
Our mission is to advance an open, collaborative, and safe path toward AGI by building rigorous benchmarks, large-scale open datasets, and shared research infrastructure. We believe meaningful progress in AI should be transparent, reproducible, and community-driven, rather than developed behind closed doors.
Beyond open-source research, 2077AI actively fosters cross-institutional collaboration and supports researchers at different career stages through fellowships, scholarships, and research funding programs, enabling the next generation of scientists to pursue ambitious, high-impact ideas.
💼 Follow us on LinkedIn (We also offer research intern positions！)
🧷 Subscribe to our Twitter/X
💻 Visit our website 2077AI
In 2025, 2077AI made foundational contributions across a broad range of AI research areas, including multimodal perception and document understanding, formal mathematical reasoning, agents and embodied AI, and audio generation. Our benchmarks and datasets have been widely adopted by both academia and industry, with systems such as Google Gemini 3 Pro, DeepSeek-OCR, and Apple research projects citing and using our work.
🔬 Representative Work (2025):
OmniDocBench (CVPR '25): A comprehensive benchmark for diverse document parsing, covering everything from handwritten notes to densely typeset newspapers. Cited by Gemini 3 Pro & DeepSeek OCR as a core evaluation benchmark.
PIN (Paired and interleaved documents): We scaled up multimodal learning with a new data format and released PIN-200M (~200 million documents) to improve visual and textual integration. Our team scaled the dataset to PIN-200M with over 102k downloads.
EditReward: To fix the lack of reliable reward models for image editing, we built a human-aligned reward model trained on over 200k expert-annotated preference pairs. This was used by Apple's "Pico-Banana 400k" and many other data efforts from frontier AI labs.
SuperGPQA (NeurIPS '25): We launched a rigorous benchmark evaluating LLMs across 285 graduate-level disciplines using a novel Human-LLM collaborative filtering mechanism.
VeriGUI: A verifiable long-chain dataset for GUI agents, designed to help generalist agents handle complex, long-horizon tasks across desktops and the web. It has been proven to be a solid foundational contribution to the agent browser ecosystem.
As we move into 2026, 2077AI remains committed to open-source transparency, rigorous evaluation, and deep reasoning research, continuing to build a future of AI powered by collaboration—one benchmark, one dataset, and one commit at a time.