Announcing OPT-IML: a new language model from Meta AI with 175B parameters, fine-tuned on 2,000 language tasks — openly available soon under a noncommercial license for research use cases. Research paper & more details on GitHub
@aiatmeta
-
ImageNetX: Identifying Vision System Failures at Scale
By
–
Even today’s best #deeplearning vision systems can fail when pose/lighting/background vary. Our work on ImageNetX is one of the first large scale efforts to pinpoint mistake types of in AI computer vision systems. Explore the dataset
-
Acknowledgment of Paper Contributors and Authors
By
–
Thank you to @ThomasScialom
, @omerlevy_
, @timo_schick and @OHonovich for their work on this paper. -
Scaling Natural Language Instructions to 240K Diverse Examples
By
–
New paper — In ‘Unnatural Instructions’ we propose a new method to automatically generate natural language instructions, allowing us to scale up to 240K diverse instructions & train models that rival the performance of contemporary instruction-tuned models. Read on ArXiv
-
Meta AI Newsletter Year-End Recap and NeurIPS 2022 Highlights
By
–
A new issue of the Meta AI newsletter is heading to your inbox next week! We're wrapping up the year with a look at some of our biggest breakthroughs, research you may have missed and a recap of what we shared at #NeurIPS2022. Subscribe
-
ROSCOE: New Metrics Suite for Evaluating Step-by-Step Reasoning
By
–
ROSCOE is a first-of-its-kind suite of metrics for scoring step-by-step reasoning. By publishing this study we hope to provide a foundation that enables scalable systematic evaluation and benchmarking of new language models. See the paper on arXiv https://
arxiv.org/abs/2212.07919 -
Thanking researchers for advancing AI study and progress
By
–
Thank you to @OlgaNLP
, @moyapchen
, @spencerpoff
, @LukeZettlemoyer
, @CMacota99
, Maryam F. and @real_asli for all of their work to move this space forward with this study! -
CICERO Achieves Top 10% Performance in Online Diplomacy League
By
–
If you guessed ‘double the average score of the human players’ you are correct! CICERO’s score ranked it in the top 10% of participants who played more than one game in the online league. Read about CICERO’s gameplay
-
CICERO AI Diplomacy Game: Win Rate Against Human Players
By
–
Meta AI’s CICERO plays Diplomacy with humans, like humans. And more often than not, wins. Recently, CICERO played Diplomacy games online against 82 competitors. Can you guess CICERO’s win rate compared to that of the human players?
-
Meta AI Unveils data2vec 2.0: 16x Faster Self-Supervised Learning
By
–
Announcing data2vec 2.0, a new general self-supervised algorithm built by Meta AI for speech, vision & text that can train models 16x faster than the most popular existing algorithm for images while achieving the same accuracy. Read more & get the open source code