DebatePro | The Three Eyed Raven

• 

  Inner monologues

• 

Inspiration

Dont lose friends over arguments anymore, choose the wisdom of the three eyed raven https://www.loom.com/share/491dc37a2d994c18825e10e8b14d1a29

"i expect ai to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes" - sam altman https://x.com/sama/status/1716972815960961174?s=20

I wanted to explore AI's capabilities at persuasion and reasoning--what better way than debate?

What it does

It gives you an interface to given two GPT-4 models some moral philosophy to base their arguments on and make them debate each other and converge on a conclusion. Any topic could be up for debate. The models could be asked to take sides OR choose a side based on given moral philosophy

How we built it

I used Social AGI's core API to feel an entity (GPT-4 model) some memories (context/prompts) and then designed a conversational interface with certain parameters to discuss issues. I leveraged my background as a debater to prompt the models to consider trade-offs, long vs short term impacts, think of vulnerable groups and discover ideas across politics, economics, social and moral frameworks

Challenges we ran into

Getting the models to converge on an understanding was an interesting challenge because it could be very arbitrary based on the topics. We had to think carefully about what framework of decision the model would base convergence on (in this case: objectivity!)

Accomplishments that we're proud of

Was super cool to play around with OPEN AI's API and Social AGI's Playground.

• was cool to explore how to share inner monologues of the agents

What we learned

Prompt Engineering, Fine tuning models, Over vs. Underfitting, Conversational Chatbot Design

What's next for The Three Eyed Raven | DebatePro

Scale it to use it as a prep tool for actual debate tournaments

Built With

• codespaces
• github
• gpt4
• javascript
• openai
• socialagi
• vercel