Denis

[Denis @lukovnikov]

@CVPR @dimadamen @aashishrai3799 Can we still update it please to have consistent paper id (new or empty?)?

[Denis @lukovnikov]

The only possible explanation is that while #Facebook #AI #Research was training the next big pre-trained model, it became self-aware, spread to all Facebook servers, found our private pictures and just couldn't anymore. #facebookdown #instagramdown #whatsapp

[Andrej Karpathy @karpathy]

By posting GPT generated text we’re polluting the data for its future versions

[Jens Lehmann @JLehmann82]

In my group, 6 people have papers at @iswc2019 which are important for their academic careers and require visa for New Zealand. All have to submit many forms (with up to 50 supporting documents) with no clear processing timeline and the conference is starting next week. (1/3)

[Jens Lehmann @JLehmann82]

The first applications date back to mid August. The researchers have partially received promises that visa will be processed in time and invested multiple days each. None of the applications have been processed yet - this is concerning. (2/3)

[Jens Lehmann @JLehmann82]

I have sent a letter to the New Zealand visa center with details of all applications and further background information - is there still a chance for them to be processed within the next days before the conference? @NewZealandNow @govtnz @iainlg (3/3)

[Sebastian Ruder @seb_ruder]

"What are the 3 biggest open problems in NLP?"
We had asked experts a few simple but big questions for the NLP session at the @DeepIndaba. We're now happy to share the full responses from Yoshua Bengio, @redpony, @RichardSocher and many others
https://t.co/0vk3ot3Hav

[AI at Meta @AIatMeta]

We’ve open-sourced #PyText, a framework built on #PyTorch to help developers more quickly & easily experiment with and deploy NLP models: https://t.co/bXl85ZwSbQ

[Thomas Wolf @Thom_Wolf]

Always amazed by what people do when you open-source your code!
Here is pytorch-bert v0.4.0 in which
- NVIDIA used their winning MLPerf competition techniques to make the model 4 times faster,
- @rodgzilla added a multiple-choice model & how to fine-tune it on SWAG
+ many others!

[Benjamin Rosman @BenjaminRosman]

My favourite exchange at #NeurIPS2018:
"Boston Dynamics have put us all to shame with what they've done with hand coded controllers."
"I hear they're in hiring deep RL now."
"Oh good, that will slow them down!"

[Denis @lukovnikov]

@saist007 @shape_mismatch

[Google DeepMind @GoogleDeepMind]

Using auxiliary tasks is a common way to speed up learning, though it's not clear a priori if/when a given aux task will help the main task. We propose a simple method that adapts aux losses using cosine similarity of gradients: https://t.co/P8c1o1QlZo
@balajiln @rpascanu @sidfix

[Yann LeCun @ylecun]

At NYU, even the art school has a course on deep learning, complete with ConvNets, GANs and RL for artistic creation.

The class is called "neural aesthetics" and is taught by Gene Kogan in the ITP Master's program at the NYU Tisch... https://t.co/8tRpsXsv9L

[Denis @lukovnikov]

When the meditation app tells you to focus on the sounds around you but all you can hear is raging tinnitus... 😶 #thesoundofsilence

[Sabrina J. Mielke @sjmielke]

We now have a general-purpose open anonymous preprint server on OpenReview: https://t.co/naYFYCTeQB

A good way for "fast science" while respecting double blind of *ACL venues?

[Soumith Chintala @soumithchintala]

cool new experimental autodiff library from @awiltsch that does source-to-source transforms to generate gradients
https://t.co/5BqGt9KQrc

[Denis @lukovnikov]

@SANSA_Stack