The question of whether it "grasps or understands" anything is irrelevant when you're dealing with a transfer learning model.https://medium.com/analytics-vidhya/openai-gpt-3-language-models-are-few-shot-learners-82531b3d3122 …
-
Oh it *is* close in basic law. What you have to remember is, there's an enormous gap between rephrasing snippets and assembling an actual filing, brief, etc. It isn't close to doing an end-to-end job. But as a time-saving drafting aid for attorneys? Feedback is + from them.
This is precisely a use case for which I would vigorously oppose the use of models such as this. GTP-3 may give the illusion of legal intelligence but a) only a trained lawyer could discriminate truth vs BS and b) there is no explainability.
-
-
That's why I see the use case, at least initially, as a "robot associate" not a public-facing service. There are also a lot of developments in deep learning for assessing legal arguments etc. The results from an adversarial scoring-generation-refinement process would be better.
-
Even then, IMHO relegating issues of material human consequence with no degree of transparency or explainability is dangerous. This is precisely what
@safiyanoble wrote in Algorithms of Oppression.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.