Inspiration
I came here with a different idea, to build an AI agent that can autonomously handle technical support cases. But I've hit consecutive roadblocks that made implementing this difficult. Simultaneously, I have been quite preoccupied with a report that came out today by a former OpenAI superalignment employee on impending AGI and what it means for the future. This made me think back to a report from this OpenAI back in 2018 that intrigued me. One of their goals for the future was to test debate between two capable AIs in natural language. We're in that future now, so at 10:35pm, this is what I have decided I will work on.
What it does
Two AI agents, each with an opposing agenda and the same set of tools at hand, will try to convince a third AI to make a certain stock market choice. The third AI will have to think through it, ask them questions, and eventually make a choice.
How I built it
I will be using two similarly intelligent LLMs, giving each of them the same set of tools (company financial reports and twitter+stocktwits threads, and have them convince a third (less intelligent) AI on a decision to make.
Challenges we ran into
I've faced a few challenges on my previous project and quite underestimated its complexity to be able to complete in a day. I am just starting work on this, and expect to face more challenges in the next ~12 hours.
Accomplishments that we're proud of
TBD
What we learned
TBD
What's next for debAIte
TBD
Built With
- agents
- llms
- python
- rag
- tool-use
Log in or sign up for Devpost to join the conversation.