" a claim such as Tim Roth was born in <MASK> should have predicted "1961", but predicted "London"." - At risk of stating the obvious... what's the problem with that autocompletion? There are clearly multiple possible valid completions here, even if some of them are facts no normal human would bring up. But both city of birth and birthdate are valid things in normal conversation as well as syntactically. SnowFire (talk) 17:49, 1 July 2020 (UTC)[reply]
CCing Matthew Sumpter who wrote this review and has read the actual paper, but my understanding is that this ambiguity becomes a problem if one wants to use the method to e.g. fact-check the claim "Tim Roth was born in 1944". Regards, HaeB (talk) 02:21, 2 July 2020 (UTC)[reply]
Facebook research about automated Wikipedia-based fact-checking using language models
Forgive me, but I couldn’t help shake my head when I read this. Facebook is the leading source of coronavirus misinformation in the world. Perhaps these "researchers" should focus their efforts on their employer. Viriditas (talk) 22:35, 2 July 2020 (UTC)[reply]
← Back to Recent research