A team incubated by BAAI has also set its sights on RSI

What happened: A team incubated by BAAI, Tongyan Lingzhi, is developing Recursive Self-Improving Personal AI (RSI), combining a shared base model with lightweight Personal LoRA parameters and an Evaluator that filters feedback before updates. Key numbers: Paged LoRA tests used Qwen2.5-72B-Instruct on 8 NVIDIA H800 GPUs with a 5.79GB Why it matters: The company argues enterprises should accumulate proprietary capability in model parameters rather than rely on retrieval, and says its RSI core route and product architecture are defined while the full automated loop remains in development.

A team incubated by BAAI, Tongyan Lingzhi, is developing Recursive Self-Improving Personal AI (RSI), combining a shared base model with lightweight Personal LoRA parameters and an Evaluator that filters feedback before updates.

The company argues enterprises should accumulate proprietary capability in model parameters rather than rely on retrieval, and says its RSI core route and product architecture are defined while the full automated loop remains in development.