Allen Institute for AI (@allen_ai)

Allen Institute for AI @allen_ai

AI for the Common Good. › Join us: https://t.co/3jnLzLT2x0 › Get our newsletter: https://t.co/0FMF3m22VC › Build an AI startup: https://t.co/B7nibqAS4G

Tweets by Allen Institute for AI

How could we build AI systems to reuse existing models rather than learn a completely new model each time? This is the question @TusharKhot, Kyle Richardson, @DanielKhashabi, and Ashish Sabharwal seek to answer in their papers, described in 2 parts: https://t.co/1n4DqrJIkk

Shared by Allen Institute for AI at 5/13/2022

Have you ever wished you could consume more research to help you in your scientific endeavors? In their new paper, @Hoper_Tom, @_DougDowney, @etzioni, @dsweld and @erichorvitz envision #AI systems that can assist by presenting relevant information for key tasks.

Shared by Allen Institute for AI at 5/9/2022

AI2's computer vision team PRIOR just released a new benchmark to challenge the research community to build robust, general-purpose vision models and push the field forward in this important area. 👁 Learn more: https://t.co/JZabO2OJTG

Shared by Allen Institute for AI at 5/6/2022

Test your AI knowledge, explore a new multi-modal model, access @SemanticScholar's open research corpus, and lots more in our latest edition of the AI2 Newsletter: https://t.co/OlHbit3k2m

Shared by Allen Institute for AI at 5/5/2022

In their #NAACL2022 paper, @alexandraxron, @mattthemathman and @JesseDodge make language modeling more efficient using a hierarchy of domain adaptation. Learn more on our blog, or access their paper below! https://t.co/VacuEzEmBb

Shared by Allen Institute for AI at 5/4/2022

In their #NAACL2022 paper, AI2ers @arnaik19, @SergeyFeldman, @lucyluwang and @Hoper_Tom, together with @sravmd, build an NLP system that searches medical literature for information that improves clinical outcome prediction of ICU patients.

Shared by Allen Institute for AI at 5/2/2022

Transformer-based language models are mysterious... 🧐 To help understand their inner workings, AI2 researcher @megamor2 & colleagues just introduced LM-Debugger, an open-source interactive tool for inspecting and intervening in LM predictions. Learn more: https://t.co/tOGNaOXzcr

Shared by Allen Institute for AI at 4/29/2022

Transformer-based language models are mysterious... 🧐 To help understand their inner workings, AI2 intern @megamor2 and colleagues just introduced LM-Debugger, an open-source interactive tool for inspecting and intervening in LM predictions. Learn more: https://t.co/tOGNaOXzcr

Shared by Allen Institute for AI at 4/29/2022

Natural Instructions v2 is a collection of 1600+ diverse language tasks and their expert-written instructions, designed by the @ai2_allennlp team as a powerful new benchmark for a wide variety of #NLProc tasks.

Shared by Allen Institute for AI at 4/20/2022

"AI processes things differently from the way that humans do, but there are principles that are difficult to avoid if we want robust AI systems. Multimodality is definitely one.” AI2 researcher @rown spoke w/ @ryonck about QA model Merlot for @geekwire: https://t.co/BXYQrU79Hl

Shared by Allen Institute for AI at 4/18/2022

ICYMI: The newest 'Incubator Insights' newsletter is out and it's full of interesting developments in applied AI including search tech with https://t.co/nvTQumKYOy, few-shot entity extraction, and open-source large models. Check it out & subscribe: https://t.co/VEtTpDdn3y

Shared by Allen Institute for AI at 4/12/2022

Check out AI2's own multimodal video question answering model MERLOT Reserve: https://t.co/VjjYPi5RMG

Shared by Allen Institute for AI at 3/25/2022

“It is fascinating that the tools of artificial intelligence, specifically deep learning, do seem to be the natural tools to study this other kind of ‘A.I.’—animal intelligence.” - AI2's @etzioni on the way AI is illuminating animal communication in this great article by @mims:

Shared by Allen Institute for AI at 3/21/2022

Our #ComputerVision team PRIOR continues to raise the bar in the realm of #EmbodiedAI - learn more about their updated Rearrangement Challenge featuring a new dataset and faster training for #CVPR2022 in this coverage from @anthony_alford via @InfoQ: https://t.co/Xhkm0VA1uJ

Shared by Allen Institute for AI at 3/17/2022

Tomorrow at 11am PST, hear AI2's Peter Clark and Niket Tandon of the Aristo team talk about Macaw, the team's powerful new question-answering system that can outperform GPT-3! Free to attend, don't miss it. https://t.co/7Hm9DEvotn

Shared by Allen Institute for AI at 3/7/2022

"That's not what I meant!" 😒 Introducing a new way of teaching complex language models like GPT-3 to better understand human instructions using user feedback, avoiding costly retraining: https://t.co/Jx77Q0BdMm

Shared by Allen Institute for AI at 2/28/2022

We are delighted by the recognition of AI2’s researchers in this year’s @NeurIPSConf Outstanding Paper Awards! Congratulations to @Swabhz @KrishnaPillutla @rown @jwthickstun @wellecks @YejinChoinka and Zaid Harchaoui whose work will be presented Dec. 7. https://t.co/FixXLFsw4g

Shared by Allen Institute for AI at 12/1/2021

“The thing about AI learning from just raw text…is it ends up reflecting a lot of human problems and biases.” AI2 researcher @YejinChoinka spoke with @ryonck at @geekwire about how our machine ethics model Delphi was built and where it might go from here. https://t.co/Vvu0Lv7I9e

Shared by Allen Institute for AI at 11/29/2021

AI2 CEO @etzioni has been appointed to @UiPath’s new AI Advisor Board! Read more about it here: https://t.co/pXQYtsAB7O #AI2 #ArtificialIntelligence

Shared by Allen Institute for AI at 11/18/2021

Our Featured AI2er series is back with Bailey Kuehl from the @SemanticScholar team! Learn more about her work on the blog.

Shared by Allen Institute for AI at 11/11/2021

@WhyLabs also announced the full availability of their SaaS platform for monitoring data and model health. #ML engineers can explore it (free and self-serve!) right here: https://t.co/ySpuRsybZG

Shared by Allen Institute for AI at 11/4/2021

Today on the AI2 Blog: An overview of our model Delphi as a nascent step toward machine ethics, a look into the stress testing of this research prototype, the lessons learned, and reflection on why working on imperfect solutions is better than inaction: https://t.co/ObjUwdVD5N

Shared by Allen Institute for AI at 11/4/2021

“We have to teach AI ethical values because AI interacts with humans. And to do that, it needs to be aware of what values humans have.” – AI2's @YejinChoinka, interviewed by @PoppyNoor for @guardian on our model Delphi and the new field of machine ethics. https://t.co/D7ao5Et7gN

Shared by Allen Institute for AI at 11/2/2021

AI2's Aristo team presents Macaw, an open-source, versatile, generative question-answering system with strong zero-shot performance on a wide range of question types. Macaw outperforms #GPT3 by >10%, even though Macaw is an order of magnitude smaller! https://t.co/WaszkZSFpY

Shared by Allen Institute for AI at 10/15/2021

Learn how to use CheckLists with AllenNLP! Easily construct suites of tests defined by inputs and expected outputs to check for robustness across several general linguistic capabilities. https://t.co/ISKBMlNGP7 #allennlp #nlproc @ai2_allennlp

Shared by Allen Institute for AI at 10/11/2021

How did we scale up the AllenNLP library to train 11 billion+ parameter models on a single node? AI2's @epwalsh10 from the @ai2_allennlp team wrote a great deep dive about exactly this, today on the AI2 Blog: https://t.co/vl3DKF91WF #NLProc #AllenNLP

Shared by Allen Institute for AI at 10/7/2021

Google's new Multitask Unified Model (MUM) was trained to make inferences using huge amounts of data scraped from the web, something our researchers have shown can introduce problematic biases – more on this new tech from @kharijohnson via @wired: https://t.co/J2fxO3s2Xf

Shared by Allen Institute for AI at 10/1/2021

Happening soon at #CVPR2021, 7pm PT! AI2 is working on the future of #EmbodiedAI with #AI agents that can successfully navigate realistic, visually complex home-like environments and rearrange the objects inside. Visit our presentation to learn more!

Shared by Allen Institute for AI at 6/23/2021

Researchers at USC and AI2 PRIOR propose new large-scale dataset, VidSitu, for understanding and representing related events in a video using visual semantic role labeling. #CVPR2021 Time: Jun 22, 8am - 10.30am PST Paper: https://t.co/LNphPhpzuz Video: https://t.co/evCsPDmxI2

Shared by Allen Institute for AI at 6/21/2021

Learn about the approaches researchers including AI2's @YejinChoinka and @JesseDodge are taking to address the huge problem of the racist, sexist, & otherwise toxic language that today's powerful language models can produce – by @kharijohnson via @WIRED: https://t.co/TFZRXdwRE1

Shared by Allen Institute for AI at 6/17/2021

Lots of #NLProc researchers are interested in the C4 dataset, which the @ai2_allennlp team made available for direct download in March. The team partnered with the mT5 authors, and we’re happy to announce the multilingual C4 is now available too!

Shared by Allen Institute for AI at 6/16/2021

Humans do multimodal reasoning across time to understand events contextually, a talent even the most advanced #AI struggles with. @Kyle_L_Wiggers covers a new model out of @uwcse & AI2 that can learn this kind of multimodal knowledge by watching YouTube: https://t.co/GxUu8DZDwh

Shared by Allen Institute for AI at 6/16/2021

"Led by @tanmay2099, researchers at @allen_ai and UIUC designed a general-purpose vision architecture and built a system, GPV-I, that can perform visual question answering, image captioning, object localization, and image classification." More in the latest issue of The Batch:

Shared by Allen Institute for AI at 6/4/2021

The next session of the #mbodied AI Lecture Series @ PRIOR is happening at 11am PST TODAY! Hear from @SongShuran from @Columbia: Self-Adaptive Manipulation: Rethinking Embodiments in Embodied Intelligence Link to join via Zoom: https://t.co/xcJ3Ld2izR

Shared by Allen Institute for AI at 6/3/2021