ucb_agentic_ai

Lecture 01: LLM Reasoning

Link to lecture recording on YouTube

Date: 2024-09-09

Speaker: Denny Zhou

Speaker’s Social Profile: Website / Company Profile / Google Scholar / GitHub / LinkedIn / X (Twitter)

Work:

• Research Scientist, Google Deepmind

Overview of the Course by Prof. Dawn Song

Accelerated development of large language models (LLMs)
Instead of text inputs / outputs, here we

• use LLM as the brain for reasoning and planning
• enable the agents to interact with external environments: observe the environment and take actions in the environments
• agents will be using external tools and knowledge base for retrieval

The rich capabilities of LLMs make these agents very flexible, and they can easily operate in diverse environments without much particular training
Agent can interact with other agents; multi-agent collaboration: division of labor for complex tasks

Why empowering LLMs with the agent framework

• solving real-world tasks typically involves a trial-and-error process
• leveraging external tools and retrieving from external knowledge expand LLM’s capabilities
• agent workflow facilitates complex tasks

Challenges for LLM agent deployment: | Challenge | Details | | – | – | | Reasoning and planning | LLM agents tend to make mistakes when performing complex tasks end-to-end | | Embodiment and learning from environment feedback | <ul><li>LLM agents are not yet efficient at recovering from mistakes for long-horizon tasks</li><li>continuous learning, self-improvement</li><li>multimodal understanding, grounding and world models</li></ul> | | Multi-agent learning, theory of mind | | | Safety and privacy | LLMs are susceptible to adversarial attacks, can emit harmful messages and leak private data | | Human-agent interaction, ethics | how to effectively control the LLM agent behavior, and design the interaction mode between humans and LLM agents |

[Incomplete, work in progress]

Notes

References

This site is open source. Improve this page.