From TaskGen to AgentJo: Creating My Life Dream of Fast Learning and Adaptable Agents
36TaskGen is a framework that is a culmination of 5 years of thoughts during my PhD to build fast learning and adaptable agents. It uses a task-directed, memory-based mechanism to focus on tasks and learn from the environment, with knowledge sharing on a need-to-know basis.
AgentJo is the continuation of TaskGen, as we scale it to multiple memory abstraction spaces, multiple agent augmentations, adaptive learning, multi-agent learning and many more.
I am excited to see how AgentJo can scale and solve more real-world use cases!
I also demonstrate a live example of how learning can take place via reflection done after every subtask via a Rock, Scissors, Paper game. This improves on the baseline agent significantly and lets learning take place faster.
~~~
Repo: https://github.com/tanchongmin/agentjo
Slides: https://github.com/tanchongmin/agentjo/blob/main/resources/From_TaskGen_To_AgentJo.pdf
Related Work
TaskGen Paper Video: • TaskGen Overview: Open-Sourced LLM Ag...
TaskGen Paper: https://arxiv.org/abs/2407.15734v1
~~~
0:00 Introduction
5:30 Live Demo of AgentJo Reflection Wrapper
16:16 Chain of Thought and Reflection
17:53 StrictJSON
24:11 TaskGen
31:13 Current features of AgentJo
40:32 AgentJo Future Plans
1:03:20 Agent Wrappers
1:06:06 Memory Abstraction Spaces
1:10:28 Adaptive Learning
1:13:15 Multi-Agent Learning
1:26:22 Discussion + Discord Logo Design Competition
~~~
AI and ML enthusiast. Likes to think about the essences behind breakthroughs of AI and explain it in a simple and relatable way. Also, I am an avid game creator.
Discord: https://discord.gg/bzp87AHJy5
LinkedIn: https://www.linkedin.com/in/chong-min-tan-94652288/
Online AI blog: https://delvingintotech.wordpress.com/
Twitter: https://twitter.com/johntanchongmin
Try out my games here: https://simmer.io/@chongmin