absorb.md

Yejin Choi

Chronological feed of everything captured from Yejin Choi.

paper / yejinchoinka / Jun 10

SurfDesign: Surface-Conditioned Framework for Advanced Protein Design

SurfDesign is a new protein design framework that leverages continuous geometric manifold representations of molecular surfaces, integrated with protein language models. It employs surface-based equivariant message passing to capture detailed surface geometry and physicochemical properties. This approach surpasses existing methods in de novo binder and enzyme design by explicitly considering surface information, establishing a foundation for principled functional protein design.

ai-designprotein-engineeringcomputational-biologymachine-learningsurface-based-modelingequivariant-networks
“SurfDesign is a novel protein design framework that integrates molecular surface geometry with protein language models.”
paper / yejinchoinka / Jun 10

Scaling Participation: Modular AI Systems Built by Diverse Contributors Outperform Monolithic LLMs

This work introduces "scaling participation," a paradigm shift from monolithic AI to modular, bottom-up systems built by diverse contributors. These participatory AI systems, composed of small, specialized models, outperform centralized LLMs by up to 15.4% on various tasks. They demonstrate benefits from contributor diversity, improve individual model performance, and exhibit emergent capabilities, offering a technical foundation for a more open and collaborative AI future.

ai-systemsdistributed-aiparticipatory-aimodular-aiai-collaborationllm-alternatives
“Participatory AI systems outperform monolithic LLMs by up to 15.4% across 15 tasks, including reasoning and factuality.”