HACE coevolves specialist models and LLM agents
Original title: Heterogeneous Agentic Coevolution
The method beat evaluated self-improving baselines on recommendation, protein function, and crystal stability tasks.
Researchers propose Heterogeneous Agentic Coevolution (HACE), which jointly evolves specialized foundation models and LLM agents. Specialists improve through supervised learning while agents train with on-policy reinforced self-training guided by verified gains. Evolutionary search adapts how the models exchange information. HACE outperformed the evaluated self-improving methods on primary metrics in all three tested domains.
- Domains are recommendation systems, protein function prediction, and crystal stability prediction.
- Agents are trained only when verified performance gains appear on the current specialist.
- Evolutionary search optimizes the interaction harness as capabilities change.
- The paper reports no named model releases or public weights.
From the original article
1 min read at huggingface.co
Recent advances in self-improving LLM agents have demonstrated the potential of AI systems that drive their own intelligence. However, many domain-specific applications rely on specialized foundation models that possess strong domain expertise but remain dependent on human-designed improvements.
By Wenqian Ye, Ziwei Guan, Zi Yang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.