Chenguang Wang(@ChenguangWang) 's Twitter Profile Photo

🔥Really excited about our work on creating a new dataset to test LLMs' ability to follow human social norms. Our Agent, a multiagent framework, can improve current social understanding.

📃arxiv.org/abs/2404.02491
Stay tuned for the code and data!

🔥Really excited about our #NAACL2024 work on creating a new dataset #Social to test LLMs' ability to follow human social norms. Our #SocialAgent, a multiagent framework, can improve current social understanding.

📃arxiv.org/abs/2404.02491
Stay tuned for the code and data!
account_circle
aivnte(@aivnte) 's Twitter Profile Photo

The experiment ran GPT 3.5-Turbo and LLaMA2-70B-Chat as well as a proposed agent - “SocialAgent” that ran on both models.

Here are the results:

The experiment ran GPT 3.5-Turbo and LLaMA2-70B-Chat as well as a proposed agent - “SocialAgent” that ran on both models. 

Here are the results:
account_circle
aivnte(@aivnte) 's Twitter Profile Photo

SocialAgent used a constellation of three agents. ‘Programming Agent’, ‘Retrieval Agent’ and ‘Reasoning Agent’. These three agents provided SocialAgent with the necessary tools to browse the internet, complete calculations and implement chain of thought reasoning.

account_circle
aivnte(@aivnte) 's Twitter Profile Photo

“SocialAgent pushes the standard models on par with human performance. For instance, SocialAgentGPT3.5−Turbo outperforms humans by an average of 0.8%. These results are significant although more advancements are needed to compete with human experts.”

account_circle
SocialAGENT-(@SocialAGENT_AI) 's Twitter Profile Photo

Einkommen aufbaust. Lass dir dieses etablierte und effektive System nicht entgehen.
copecart.com/products/d4fde…

account_circle
SocialAGENT-(@SocialAGENT_AI) 's Twitter Profile Photo

Mit unserem Produkt erhältst du eine einfache, aber sehr effektive Anleitung, wie du ein höheres Einkommen auf Instagram generieren kannst. Du erhältst 1 zu 1 Unterstützung und alle notwendigen Informationen für eine erfolgreiche Umsetzung unserer Methoden. Wir bieten eine

account_circle