Another thing I look forward to this year: presenting our #NLProc reproducibility survey at #EACL2021! A. Shimorina, @shubhamag199, @ehudreiter & I reviewed 67 papers, 506 individual reproductions. Just 16% of scores were the same. Where they differed, 60% were worse, 40% better.
Last chance today to vote for *ACL/EMNLP 2021 workshop proposals, e.g. the only workshop focused on human evaluation, with fantastic speakers @mohitban47@mmitchell_ai and @lspecia#NLProc
👉https://t.co/nAtsnEE4wl
👉https://t.co/5IJ39vMYHW
Combination of blinkered faith in algorithms and ignorance of how to design and test them underlies this spectacularly ill-conceived formula for calculating grades from pupil rankings - almost any kind of testing would have revealed extent of problems #ALevelProtests#alevels2020
.@RobKoeling and I have just released the Elsevier 40k OA CC-BY full-text corpus for NLP and ML research. 👨💻🎓📚
Arxiv: https://t.co/IfSuGCCsss
Dataset: https://t.co/3UUcaAaVYt
It is made up of full-text documents from across all academic disciplines (Dentistry to Agriculture)
Dominic Cummings moving into Downing St as de facto chief of staff to PM wearing an OpenAI t-shirt. Can't think of a safer pair of hands for testing usefulness of AI/ML in government. #ethicalai