Merge pull request #7 from yushengsu-thu/yushengsu-thu-patch-1

Update about.md

_pages/about.md

@@ -43,9 +43,9 @@ To achieve this goal, LLMs should be *comprehend instructions and efficiently ad
 
 Recently, I am more focus on:
 
-* <b>Autonomous Agent.</b> Developing agents (based on LLMs) that can autonomously interact with the external environment (or humans) to self-improve and drive long-horizon decision-making, thereby accomplishing more complex tasks in the real world ([AgentVerse](https://arxiv.org/abs/2308.10848), [XAgent](https://blog.x-agent.net/about/), [Tool Leaning](https://arxiv.org/abs/2304.08354), [ChatDev](https://arxiv.org/abs/2307.07924)) <small> Note that: Currently, I am exploring ways to transform an assistant agent into an autonomous agent. </small> [[Resaerch Talk 2](https://drive.google.com/file/d/1kxnBD_Q-zX5CD7raNSqd9IflbES8HTkW/view?usp=sharing)]
+* <b>Autonomous Agent </b> Developing agents (based on LLMs) that can autonomously interact with the external environment (or humans) to self-improve and drive long-horizon decision-making, thereby accomplishing more complex tasks in the real world ([AgentVerse](https://arxiv.org/abs/2308.10848), [XAgent](https://blog.x-agent.net/about/), [Tool Leaning](https://arxiv.org/abs/2304.08354), [ChatDev](https://arxiv.org/abs/2307.07924)) <small> Note that: Currently, I am exploring ways to transform an assistant agent into an autonomous agent. </small> [[Resaerch Talk 2](https://drive.google.com/file/d/1kxnBD_Q-zX5CD7raNSqd9IflbES8HTkW/view?usp=sharing)]
 
-* <b>AI Alignment - scalable oversight.</b> Aligning AI models to act in accordance with human values is challenging: these values are taught by humans who make mistakes, harbor biases, and have a complex that is hard to completely specify. AI models often learn to exploit even minor imperfections in the specified objective, a tendency known as reward hacking. To achieve the alignment goal, a core open problem is a scalable oversight. Thus, I study how to explain, understand, and evaluate ways to better control how AI models extend  our supervision over tasks that are beyond our direct oversight, thereby improving alignment and their ability to generalize. ([Model Emotion](https://arxiv.org/abs/2302.09582), [Chateval](https://arxiv.org/abs/2308.07201)) 
+* <b>AI Alignment - scalable oversight </b> Aligning AI models to act in accordance with human values is challenging: these values are taught by humans who make mistakes, harbor biases, and have a complex that is hard to completely specify. AI models often learn to exploit even minor imperfections in the specified objective, a tendency known as reward hacking. To achieve the alignment goal, a core open problem is a scalable oversight. Thus, I study how to explain, understand, and evaluate ways to better control how AI models extend  our supervision over tasks that are beyond our direct oversight, thereby improving alignment and their ability to generalize. ([Model Emotion](https://arxiv.org/abs/2302.09582), [Chateval](https://arxiv.org/abs/2308.07201))