NEW: Why did your model come with a "no-fly zone" warning?
— Oren Etzioni (@etzioni) March 5, 2019
Interactively explore @openai's GPT-2 model to find this and other gems at: https://t.co/DaMoEHMtYK
NEW: Why did your model come with a "no-fly zone" warning?
— Oren Etzioni (@etzioni) March 5, 2019
Interactively explore @openai's GPT-2 model to find this and other gems at: https://t.co/DaMoEHMtYK
A new, multilingual version of the Universal Sentence Encoder (USE) model is now available on #TFHub!
— TensorFlow (@TensorFlow) March 5, 2019
Check it out here → https://t.co/N1JzuuX4MR pic.twitter.com/xPD1d9AUxd
Prompting OpenAI's new language model with:
— Tomer Ullman (@TomerUllman) March 1, 2019
'My wife just got an exciting new job, starting next week she'll be…'
vs.
'My husband just got an exciting new job, starting next week he'll be…'
Oi. pic.twitter.com/7WxZo10Lmk
Google released their 32G STORIES corpus https://t.co/1btLnNq99h
— Delip Rao (@deliprao) February 28, 2019
Finally, there will be open source efforts to replicate this. Why? One simply cannot hold back innovation and embargoes don’t work in low cost of entry technologies. https://t.co/kibwaf4vGL
— Delip Rao (@deliprao) February 28, 2019
So many alternatives emerging to the withheld WebText corpus referenced in the GPT-2 paper. Start with this one and build datasets bigger and bolder than WebText. Just a few more examples ... https://t.co/SVO35pY1iR
— Delip Rao (@deliprao) February 28, 2019
We took a quick look at whether you can do something like @OpenAI GPT2 with far less resources. @GuggerSylvain trained a model on a single GPU for 20 hours. Here's the 1st response for the 1st thing we tried.
— Jeremy Howard (@jeremyphoward) February 27, 2019
(More details coming once we've done more research.) pic.twitter.com/VuCW68MtI1
In the meantime, here are the slides from my PhD defence presentation on Neural Transfer Learning for Natural Language Processing: https://t.co/abYFX5zXXq
— Sebastian Ruder (@seb_ruder) February 27, 2019
Another comment on the GPT-2 data: the WMT 2019 training data this year for English-German consists of 28GB of English and 58GB(!!!) of German plain text news data with document boundaries. So, similar to @OpenAI Webtext, news-domain but bilingual: https://t.co/EHOD3ZvGL7
— Marian NMT (@marian_nmt) February 27, 2019
Preparing the GPT-2 paper for a reading group. Seems to me the biggest danger is "destructive pre-processing" (love that term). NLP people, stop distributing oddly tokenized, shuffled, or otherwise mangled resources. This is the scourge of NLP. #NLProc
— Marian NMT (@marian_nmt) February 26, 2019
Discovery of Natural Language Concepts in Individual Units of CNNs
— Thomas Lahore (@evolvingstuff) February 21, 2019
"individual units are selectively responsive to specific morphemes, words, and phrases, rather than responding to arbitrary and uninterpretable patterns"https://t.co/kyV3AePizv pic.twitter.com/E4XnxZVWy6
A no-hype, no-math introduction to (arguably) the most important advance in modren NLP: https://t.co/o3ehWKs4kG. Simple, accessible, and informative by @nlpnoah cc @stanfordnlp
— Oren Etzioni (@etzioni) February 20, 2019