Here’s a discussion on our recent paper with an additional experiment that uses GPT-4 to beat past GPT-4 standards. In our recent paper, "Reflexion: An Autonomous Agent with Dynamic Memory and Self-Reflection," we introduce a framework that allows AI agents to emulate human-like self-reflection and evaluate its performance on the
Thanks so much for all the amazing work you are doing! For curiosity, did you try running GPT-4 on ALFWorld and HotpotQA with and without Reflexion? The charts in the paper were all with GPT3.0 right?
The work they are doing is amazing. Do you intend that this agent can be implemented in a framework such as langchain in the future?
I read through everything.
I barely understood anything.
Yet, I couldn't stop reading.
Frig'n smart kids.
Love sharing a planet with you
Thank you for your hard work
It's ironic that if I open this page in Edge, and ask Bing Chat to summarise it, it says: "the web page context is empty".
Thanks so much for all the amazing work you are doing! For curiosity, did you try running GPT-4 on ALFWorld and HotpotQA with and without Reflexion? The charts in the paper were all with GPT3.0 right?