Inspiration
AI alignment is an important problem to solve.
We argue that there is 2 side for alignment. We need to be able to align AI agents with the creator needs, so that it can actually pursue it's goal, but we also need to align AI agents with the need of society so it does not destroy us :)
Luckily we already have a system to align autonomous agent in the real world (humans), the LAW!
So we decided to see if we could use our current law to try to align AI agent on the rules of society.
What it does
It's a langchain tool that you can use to allow agent to check if what they do is aligned with law.
We also build a front end to demonstrate our method result.
How we built it
We first chunked the law into enriched snippets. We then build embeding of this snippet. We then retrieve this embedding at a time of query and use them to enrich GPT answer.
Challenges we ran into
embedding are huge. We start to late to work (form 7pm Saturday) as we discard a previous idea
Accomplishments that we're proud of
What we learned
What's next for Law-aligned GPT
We will continue build it. We need to actually publish the Tool.
Log in or sign up for Devpost to join the conversation.