I'll just draw folks attention to the long running AAMAS conference series.
Of course, it's quite academic in nature, but it may be that some useful approaches could be picked up from this resource for LLM driven approaches.
"In the ever-evolving landscape of Natural Lan- guage Generation (NLG) evaluation, a noteworthy paradigm shift is underway as researchers increas- ingly turn their attention towards fine-tuning open- source language models (e.g., LLaMA), in lieu of traditional closed-based LLMs like ChatGPT and GPT-4. This transformative shift is propelled by a thorough exploration of key perspectives, including the expenses associated with API calls, the robust- ness of prompting, and the pivotal consideration of domain adaptability."
This paper was written by an LLM. Probably Claude-3.
Not mentioned: learning from demonstration. This is the approach we are taking at https://github.com/OpenAdaptAI/OpenAdapt.
Finding it quite difficult to decide which platform to bet on. Autogen langchain and langgraph seem to be main contenders. And then people seem to custom roll them too
perfect timing! I'm just building myself an assistant via telegram and for now went with the multi-agent collaboration via supervisor pattern.
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
[flagged]
Not mentioned in the paper, but I have been experimenting with behavior trees for LLM agents, and have had a lot of success: https://richardkelley.io/dendron/tutorial_intro/