Oh, I see. This was my error, actually -- he wrote "Black Kite", which I ended up turning into {{noping|Black}} Kite because I formatted them with a multi-line selection. Looks like I missed one! Good eye on that. jp×g07:32, 1 August 2022 (UTC)[reply]
An interesting experiment, and it's cool how well the language model managed to extract and summarize information. As for the imitation of Holmes, I'm slightly less impressed. It's difficult to stylize information summaries, particularly since factual summaries in Us Supreme Court decisions tend to be rather dry. GPT-3 also seemed to be trained primarily on web data and books, so I wonder whether its training data even had enough Holmes for a coherent style to coalesce. With this in mind, it would be interesting to see how it performs with more contemporary legal scholars whose work would be more available in a corpus of online text, for example, Amanda Frost a professor of law who writes the Academic Round-up for SCOTUSblog. — Wug·a·po·des20:27, 2 August 2022 (UTC)[reply]
← Back to Arbitration report