What's New in NLP? #3

Welcome to the December edition of Cohere’s monthly newsletter, where we share the latest updates in the world of LLMs.

Hola ??,

We wish you a happy and peaceful holiday season and a prosperous 2023. Cohere’s journey would not be the same without you. We are incredibly grateful that you are a part of it.?

News and Announcements

Welcome to Cohere’s monthly newsletter! Our recent launch of our multilingual model, the first to be trained on authentic language data, is highlighted in this edition, as is our recent welcome of Martin Kon, previous CFO of Google’s YouTube division, as our President & COO, a highlight of the latest research in language AI, as well as the exciting events we'll be attending.?

What’s New in NLP?

We had lots of fun meeting researchers and developers at EMNLP 2022. Hearty congrats to the authors of EMNLP's Best Paper Awards—Best Long Paper, Abstract Visual Reasoning with Tangram Shapes, which introduces a challenging new benchmark for language and vision while making a compelling connection to widely studied psychological research and Best Paper Awards—Best Short Paper, Topic-Regularized Authorship Representation Learning, this paper introduces a new approach to improving generalization abilities to the challenge of authorship attribution and improves the accuracy of author recognition even with writing changes.?

NeurIPS 2022 also wrapped up a few days ago. There were many exciting papers on Language Models, Diffusion Models, Procedural Generation and more, and if you haven’t, check out Cohere’s accepted papers at NeurIPS 2022.

Baker Lab shared a deep-learning method for protein design called RoseTTAFold Diffusion. With minimal input, it turns prompts (“create a molecule that binds X”) into new pproteins that fold and function in the lab.

DeepMind’s AlphaCode made breakthroughs by beating slightly over 50% of coders in a competitive programming competition. AlphaCode takes a “data-only” approach. It learns by observing buckets of existing code and is eventually able to flexibly deconstruct and combine “words” and “phrases”—in this case, snippets of code—to solve new problems.

According to Forbes, artificial intelligence will continue to impact society significantly. Toronto is becoming one of the world's leading AI hubs in 2022. Thanks to the startups capturing headlines, including Cohere, as language models become a central theme in the world of AI.

Researchers from NVIDIA announced Magic3D, an AI model that can generate 3D models from text descriptions. After entering a prompt such as, "A blue poison-dart frog sitting on a water lily," Magic3D generates a 3D mesh model, complete with colored texture, in about 40 minutes. With modifications, the resulting model can be used in video games or CGI art scenes.

Top Demos of the Year

How We Built an AI-Powered Magic the Gathering Card Generator

This is a story of how we created Urza’s AI, a website that uses Artificial Intelligence (AI) to generate Magic cards, and how over 38 thousand others joined in the fun.

Combing For Insight in 10,000 Hacker News Posts With Text Clustering

This demo shows how to build a map of the top 10,000 Hacker News posts of all time. By visualizing it using the embeddings of the titles and kept browsing it and slicing till I zoomed in on some of tech’s favorite areas of Hacker News.

Cohere Sandbox Open-Source Libraries

Get building with Sandbox, Cohere open-source libraries:

• Conversant: A framework for building conversational agents on top of the Cohere API, with a hands-on demo on how to use generative language models in conversational settings and build those interactions.
• Route Generation: Build a functional chatbot that recognizes users' intent from descriptions, maps incoming user messages, and accelerates its training by leveraging Cohere's models to enable zero-shot learning.
• Grounded QA:? A powerful, contextualized, factual question-answering Discord bot that uses embeddings, text generation, and web search.
• Topically: A suite of tools that help you use the best of topic modeling to make sense of text collections (messages, articles, emails, news headlines, etc.) using large language models.
• Toy Semantic Search: A simple semantic search engine built with the Cohere API. The search algorithm here is fairly straightforward; it uses embeddings to find the paragraph that matches the question's representation. In text sources, a concrete paragraph containing the answer is most likely to produce the best results.

Community Contribution of The Month

Contribution by Shishir Joshi. Join the Cohere community and share yours.

Attention Is All You Need is one of the most cited papers in natural language processing (NLP) that introduces the Transformer architecture. The paper was authored by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Cohere Co-founder and CEO, Aidan Gomez, et al.?

What’s New on the Blog

Bonjour. ?????. Guten tag. Hola. Cohere's Multilingual Text Understanding Model is Now Available

Humans speak over 7100 languages, yet the majority of language models only support the English language. Cohere’s mission is to solve that by empowering our developers with technology that possesses the power of language. That’s why we’re introducing our first multilingual text understanding model that supports over 100 languages and delivers 3X better performance than existing open-source models.?

Generative AI with Cohere: Part 1

Generative AI is a type of artificial intelligence that focuses on creating or generating new content or data. This can be in the form of language, images, videos, and more.In this multi-part guide, we will go through everything that you need to know about generative AI with Cohere’s large language models (LLMs).

Top NLP Papers—November 2022

In this roundup, we highlight some exciting papers on natural language processing, our work from Cohere For AI and Cohere’s technical staff, along with our involvement at NeurIPS 2022. Topics for this month include different prompting methods for understanding dialogue and humor, use cases like summarization and essay scoring, and what language models learn beyond language.??

Read More

Updates From Cohere For AI

• C4AI traveled the world to share our research, organize workshops, and connect with our community at NeurIPS and EMNLP! Organizing the Broadening Research Collaborations workshop at NeurIPS was a particularly special moment to share the values of expanding research access and inclusion at the heart of our lab. Keep up with our latest publications here.
• We celebrated our first six months as a lab by looking back at the moments that shaped C4AI. Check out the highlights!
• Join our Community! The Cohere For AI community is made up of brilliant research and engineering talent from 90+ countries across the globe. We just reached the 900-member milestone and want to keep growing our fantastic events and programs! Complete the community application form to join in.?

Upcoming Events

• Foundation Model Virtual Summit, Snorkel AI. January 17, 2023 | 9:00AM - 2:00 PM PDT. Join us for a free virtual summit that will bring together speakers from research and industry to discuss enterprise applications, best practices, and the latest research in the field of foundation models (FMs) and large language models (LLMs). Cohere’s Jay Alammar will be speaking. Register now

Like what you’re reading? Share it with your friends and colleagues.