Augrented plans to use GPT-3 to help explain legal notices to tenants, and help tenants' rights attorneys draft legal filings. Full disclosure: it's my own startup!
I tried setting up a "chatbot" to answer questions about tenant rights in California using a few human-written questions and answers as examples. The results were promising: none of the answers is (according to the attorneys I talked to) incorrect, although they are quite brief. Of course, GPT-3 can "expand on" responses to order, so this may turn out to be less of an issue than it appears from a static screenshot.
In what became one of the most widely-circulated examples of what GPT-3 can achieve in the legal field, I showed that the model can re-write common tenant complaints in legal language, even including relevant statutes.
Full disclosure: the second-to-last one is slightly incorrect (it's the wrong paragraph of the Rent Ordinance). Thanks to Adam David Long, Esq., for pointing that out! However, I'm told it's highly unlikely this would be material in an eviction case, for instance–and based on the 200 or so eviction files I've processed to train the neural networks we're using to classify and respond to eviction notices, this is a small defect by landlord attorney standards!
Finally, here's an example of GPT-3 rephrasing relatively complex legal language in terms a second grader could understand:
While there is a lot of work to be done to ensure that tools designed for use by the public deliver consistently accurate results–we'll be deploying a "conventional" chatbot for tenants soon, for instance, which won't rely on GPT-3–the potential for tools which can "translate" to and from legal language to improve access to civil justice is immense.
I hope we'll be able to bring tools based on GPT-3 to tenants' rights organizations in time for the coming eviction crisis coming in the wake of the COVID-19 pandemic.