Update MetaGPT X Technical Report.md

src/blog/swebench/MetaGPT X Technical Report.md

@@ -67,7 +67,7 @@ We have developed multi-agent systems to address complex issue-solving tasks, in
 
 1. **Multi-agent collaboration**: Agents collaborate to declare tasks, tools, and symbols, solving them independently. This approach enhances issue resolution success and enables the handling of more complex problems.
 2. **SOP (Standard Operating Procedure) generation**: Agents autonomously gather problem-solving data to construct and refine SOPs. Agents automatically design and construct problem-solving workflows for different tasks by building a graph based on task scale, problem similarity, and prior problem-solving experience. They then use the graph solver to generate and iterate on SOPs, ensuring that each step is optimized for efficiency and effectiveness.
-3. **Trajectory fine-tuning based on feedback**: Throughout the issue-solving process, agents generate comprehensive trajectories based on tool usage decisions informed by environmental feedback, covering symbol retrieval, generation, and testing. These trajectories can be harnessed to refine LLM capabilities in complex software engineering scenarios, suggesting that further fine-tuning could enhance the agents' ability to autonomously plan and analyze challenges within software development contexts.
+3. **Fine-tuning based on trajectory and feedback**: Throughout the issue-solving process, agents generate comprehensive trajectories based on tool usage decisions informed by environmental feedback, covering symbol retrieval, generation, and testing. These trajectories can be harnessed to refine LLM capabilities in complex software engineering scenarios, suggesting that further fine-tuning could enhance the agents' ability to autonomously plan and analyze challenges within software development contexts.
 
 ## Acknowledgments