New package announcement! Introducing the {friends} package 🙌
— Emil Hvitfeldt (@Emil_Hvitfeldt) August 27, 2020
Includes the entire transcript of all the 10 seasons of the beloved American sitcom Friends
Perfect for getting your feet wet with network analysis 🌐 and text analysis 📚https://t.co/0D8DagYqnt#rstats #tidytext pic.twitter.com/HUHC02zzUq
We're releasing a corpus of over 1 million snapshots of English-language privacy policies from over 130,000 websites spanning two decades, with an accompanying paper: https://t.co/oc1b8w1ERk
— Arvind Narayanan (@random_walker) August 24, 2020
By @RyanBMAmos, Günes Acar, @elenalucherini4, Mihir Kshirsagar, @jonathanmayer, and me. pic.twitter.com/7Kk9aLOOuh
The number of 'whale' words on every page of Moby-Dick. I loved the book. Absolutely loved it but still wished an editor would've shortened the book by about 20 percent. Unpopular opinion - sorry... Source: https://t.co/QO7KvjaS6b pic.twitter.com/NlgdgWgSmt
— Simon Kuestenmacher (@simongerman600) August 19, 2020
Check out the 3 courses on Tabular Data, NLP and Computer Vision from the awesome Amazon MLU team https://t.co/5RwvtRnyb4 - kudos to Bree, Rachel, Brent and the rest of the crew. Built on https://t.co/fuWTPxe9I1.
— Alex Smola (@smolix) August 18, 2020
11 state-of-the-art sentence embedding models are now included in the Hugging Face hub, thanks to @UKPLab 🚀
— Hugging Face (@huggingface) August 17, 2020
The model hub (linked) shows how to use these models both with the authors' sentence-transformers library as well as with native 🤗 Transformers.https://t.co/h6ljxWPS7B pic.twitter.com/w81dZucv5L
This is a great way to learn how to preprocess your own data with 🤗Transformers before using Trainer or a classic PyTorch/Keras training loop! https://t.co/vud0r6nzIO
— Sylvain Gugger (@GuggerSylvain) August 17, 2020
Check out REALM, a new open-source method for language model pre-training that uses a supplemental knowledge retriever that enables it to perform well on knowledge-intensive tasks without billions of parameters. Read more at: https://t.co/rwBGQVIVuz pic.twitter.com/Csl4fTvhHZ
— Google AI (@GoogleAI) August 12, 2020
Can't do all data science on the GPU they said...
— Joshua Patterson (@datametrician) August 11, 2020
you need CPU for string processing they said...
well...
Another step towards end to end #GPU acceleration with @RAPIDSai. Memory copies hurt perf... so do #NLP and text processing on the GPU! 🔥🔥🔥 I ❤️ this ecosystem! https://t.co/nzUiNseCLg
No labeled data? No problem.
— Hugging Face (@huggingface) August 11, 2020
The 🤗 Transformers master branch now includes a built-in pipeline for zero-shot text classification, to be included in the next release.
Try it out in the notebook here: https://t.co/31Z6LR1NjK pic.twitter.com/IugPUj3sa1
GPT-3 isn’t some super intelligent AI.
— Chris Albon (@chrisalbon) August 11, 2020
It is a language model desperately trying to say the right things so it fits in.
I'm not putting them on blast b/c they're a student, but I just ran across someone implying that GPT-3 correctly answered a medical question by using reasoning and underlying knowledge.
— Rachael Tatman (@rctatman) August 10, 2020
Language models simply do not do that. PLEASE don't use them for medical advice.
Turns out a lot of open-domain QA datasets have test set leakage. If you control for it, model performance drops by a mean absolute of 63%. Yikes! If we missed this for such a long time, I wonder if there are problems with other NLP datasets too. https://t.co/uPT2uYqou7
— Tim Dettmers (@Tim_Dettmers) August 7, 2020
