Law?

[sethchandler]

It would be lovely to adapt this to legal research. The problem is the data tends to be behind walls or is incomplete. Ideas?

[ltjed]

There are a few things I can think of:

1. Use fine-tuned models that are fine-tuned on those closed-source data as the underlying model of certain agents in freephdlabor. I think those will probably be available soon from startups etc.
2. Depending on your individual use case, (for example, are you interested in law research or more like finding precedents for some particular cases?) there maybe online sources that are nevertheless useful, and you can build tools that interact with those sources.
3. Worst case scenario, you can also make humans with such knowledge almost serve as an agent in the multiagent system, providing relevant information to the rest of the systems when needed, this can still be useful and automate a lot of legal work I think.
   Let me know what you think!