I am a partner at a 13 attorney law firm in CA and lecture on the use of AI in the law and pitfalls to avoid. We have about 35 years of documents on our system and it would seem to me that if we could upload them to a closed system we could then have AI help draft things for us. How can I go about doing this as our files are confidential and cannot be put on the web somewhere or used for training by another system? Thanks!
The best way would be to self host a vector database
and open source LLM. The vector database will enable semantic search over your document base and the llm will write the draft. This method reduces hallucinations but does not eliminate them entirely
Thanks for this. Really appreciate your taking the time although was not sure I completely understood it so I asked ChatGPT for help and then at the end it added the below as to how to go about it. Wonder if you agree and whether you think someone with rudimentary knowledge (like me) could make all that happen or best to hire a professional?
Next Steps for Implementation
Software Setup:
Install a vector database (e.g., Pinecone, Weaviate, or Chroma).
Deploy an open-source LLM (e.g., LLaMA 2, GPT-J, or Falcon).
Customization:
Fine-tune the LLM using your legal filings to improve accuracy and relevance.
Testing:
Evaluate outputs for accuracy and reliability, keeping an eye on potential hallucinations.
Integration:
Connect the system to your internal tools (e.g., document management software).
1 Like
It’s a bit challenging but if you want the challenge and experience it’s totally doable. There should be a lot of tutorials online about building an AI chat with document base.
I can help you if you get stuck
Much appreciated. I will give it a go with a very small batch of publicly available docs so just in case I royally screw it up, no harm no foul. And then will likely reach out to you…