Our report is now on arXiv:
— Miles Brundage (@Miles_Brundage) August 27, 2019
“Release Strategies and the Social Impacts of Language Models,” @IreneSolaiman et al.: https://t.co/IynZTMpjdH
Our report is now on arXiv:
— Miles Brundage (@Miles_Brundage) August 27, 2019
“Release Strategies and the Social Impacts of Language Models,” @IreneSolaiman et al.: https://t.co/IynZTMpjdH
Facebook AI researchers are releasing a new feature for the fastText library which provides hyper-parameter autotuning for more efficient text classifiers. https://t.co/gINIHhViTO pic.twitter.com/TV84W4skav
— Facebook AI (@facebookai) August 26, 2019
You can play with it here (there are only a few Bert architectures up for now):https://t.co/3auAZKWKfs
— Thomas Wolf (@Thom_Wolf) August 26, 2019
narrowing the gap between autoregressive and non-autoregressive (iterative refinement) NMT one paper at a time: this time with continuous latent variables and deterministic inference strategy by @raphaelshu, @jasonleeinf and me (and rejected)https://t.co/pPlU5kStB6
— Kyunghyun Cho (@kchonyc) August 23, 2019
This replication project trained a 1.5B parameter “OpenGPT-2” model on OpenWebTextCorpus, a 38GB dataset similar to the original, and showed comparable results to original GPT-2 on various benchmarks. 👏🏼https://t.co/m4ZMB8RmdShttps://t.co/ZrqJ0IuHbw https://t.co/o3KBv5VXKJ pic.twitter.com/pGN0p00DBR
— hardmaru (@hardmaru) August 23, 2019
This analysis uses a straightforward @mcmc_stan model, implemented in #rstats with brms and #tidybayes.
— Julia Silge (@juliasilge) August 22, 2019
Will I become a Bayesian?!?!????? 🤔https://t.co/smlixbT37U
‼️ 1.5B parameter GPT-2 model released, but not by OpenAI https://t.co/8tgjUWxjZo
— Mark 🦑. Riedl (@mark_riedl) August 22, 2019
Presenting LXMERT at @EMNLP2019 --> https://t.co/T9SeONSlFO (prnc. 'leksmert'). Top3 in GQA & VQA challenges (May2019), Rank1 in VizWiz, & v.strong generalzn to NLVR2 (22% abs jump)! Awesome effort by @HaoTan5!
— Mohit Bansal (@mohitban47) August 21, 2019
CODE+MODELS all public: https://t.co/JWbjEWbhXS; pls use+share! 1/2 pic.twitter.com/WvxRirYGoB
This new tool from for systematic error analysis from @uwdata looks like it will be really useful!
— Christopher Manning (@chrmanning) August 20, 2019
(I fear the common standard of 100 sampled errors in QA papers is my fault – we intended to do 200, but I never got around to doing my half….) https://t.co/0Py7IpOFRD
Today, OpenAI released GPT-2 774M (English) and Facebook released XLM pre-trained models for 100 languages. Looks like a glut of #NLProc resources for everyone freely accessible. What a wonderful time to live!https://t.co/1OMd3D5xDohttps://t.co/lyC2eKvp3J
— Delip Rao (@deliprao) August 20, 2019
OpenAI: "GPT-2 synthetic text samples almost as convincing (72%) as real articles from the New York Times (83%)"
— Chip Huyen (@chipro) August 20, 2019
Me: "Whoa people still find the media 83% convincing?" https://t.co/tSoEIP7yXT
GPT-2 6-month follow-up: we're releasing the 774M parameter model, an open-source legal doc organizations can use to form model-sharing partnerships, and a technical report about our experience coordinating to form new publication norms: https://t.co/wnLwmRt9jT
— OpenAI (@OpenAI) August 20, 2019