Core Viewpoint - The article discusses the introduction of AMix-1, a new protein foundation model developed by Tsinghua University's Intelligent Industry Research Institute and Shanghai Artificial Intelligence Laboratory, which marks a significant advancement in protein modeling akin to the transition from BERT to GPT in NLP [1][2][10]. Group 1: Technological Advancements - AMix-1 utilizes a systematic methodology based on Scaling Law, Emergent Ability, In-Context Learning, and Test-time Scaling to create a versatile protein foundation model [1][2][11]. - The model has demonstrated the ability to autonomously learn and design new proteins, significantly enhancing its capabilities as training progresses [3][4][18]. - AMix-1 has achieved a 50-fold increase in the activity of optimized variant proteins through rigorous wet lab testing [6][40]. Group 2: Model Capabilities - AMix-1 exhibits four key "superpowers": predictable growth, emergent understanding of protein structures, in-context learning for design, and test-time scaling for enhanced performance with increased validation resources [8][17][30]. - The model's emergent ability allows it to develop a "structural perception" capability autonomously as training loss decreases, leading to a qualitative leap in understanding protein folding and spatial structures [19][21]. - In-context learning enables AMix-1 to generate new proteins based on examples without requiring additional training, streamlining the design process [22][27]. Group 3: Experimental Validation - The EvoAMix-1 method allows for sustainable expansion of model capabilities during testing, demonstrating strong performance across various tasks [31][33]. - The model's ability to iteratively improve protein designs through a feedback loop from testing results showcases its potential for continuous optimization [41][43]. - A virtual biological laboratory has been developed to support the protein generation and evolution work facilitated by AMix-1, making protein modification as simple as interacting with a conversational AI [44][46].