What’s New in NLP? #5 Summarize Beta, Top NLP Papers, and More!

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

Greetings to all the developers out there! I'm thrilled to bring you some exciting news from Cohere. First off, we're proud to?launch Cohere Summarize Beta, a new endpoint for text summarization. This powerful new endpoint allows you to condense essential information from incredibly large documents, with customizable settings and the ability to process up to 50,000 characters.?Try it out for yourself?and see how it can transform the way you work!

But that's not all—we're also bringing you the?top NLP papers of January 2023. As enthusiasts of language AI, it's crucial to stay up-to-date with the latest breakthroughs and advancements. We've curated a selection of the best research for you to explore based on submissions from C4AI's community. From language models to text generation and summarization, these papers showcase the cutting edge of NLP.

And finally, we're excited to share our?generative AI blog series?with you. In these articles, we explore the importance of these powerful models and discuss how we can better harness their power to create a new generation of language AI systems. Whether you're interested in text generation, creating custom models, or the present and future of generative AI, this series has something for you.

We want to make Cohere your go-to resource for language AI, and we'd love to hear your feedback. Please feel free to?email us?with any thoughts or suggestions—we're always striving to improve and provide you with the best possible experience.

What's New in NLP?

Here's the latest scoop on the latest in NLP:

A new paper proposes a?Hindsight Instruction Relabeling (HIR) algorithm for aligning language models with instructions. It involves a different approach, relabeling feedback as instructions to train the model for better alignment in a supervised way. The paper evaluates the performance of HIR on 12 challenging BigBench reasoning tasks and shows that it outperforms the baseline algorithms and is comparable to or even surpasses supervised fine-tuning.??

The latest research in augmented language models (ALMs) is the focus of a survey that explores how LMs can be augmented with reasoning skills and the ability to use tools, such as calling external modules like a code interpreter. ALMs can use various external modules to expand their context processing ability, thus departing from the pure language modeling paradigm. The survey concludes that this new research direction has the potential to address common limitations of traditional LMs, such as interpretability, consistency, and scalability issues. Check out the?paper on arXiv?for more details.

Facebook?just introduced LLaMA, which is a collection of foundation language models ranging from 7B to 65B parameters. They trained these models on trillions of tokens using publicly available datasets. LLaMA-13B shows that it can outperform GPT-3 (175B) on most benchmarks, and LLaMA-65B is competitive with the leading models out there.

Last but not least, BERTopic just released version 0.14, which includes the ability to fine-tune topic keywords and labels with models from providers like Cohere! You can also use models for part-of-speech tagging, text generation, zero-shot classification, and more! Check out the?Twitter announcement by creator Maarten Grootendorst?for all the details on this exciting update, and start exploring the?new features today. For an introduction to BERTopic,?watch Maarten’s presentation on the Cohere YouTube channel.

Cohere Community Section

Contribution by Muhtasham#4630.?Join the Cohere community and share yours.

Meet Our Co:mmunity Champion

We’re stoked to introduce you to our Co:mmunity Champion: Elle Neal#0726!

Remember to say???to Elle on Discord and check out her Medium post?3 Easy No-Code Steps to Build Your AI Integrated App using Retool and Cohere API, where she shares how to bring your AI ideas to life without a single line of code!

Coolest Co:mmunity Projects in February?

Here’s a rundown of some of the coolest projects coming from our co:mmunity in February:

1. Harish Garg created a?Valentine’s Day Gift Ideas Generator?using Cohere.
2. Chris Wallenwein built a?browser extension that summarizes highlighted?text?using our API.
3. Ibrahim El Chami built a?YouTube video summarizer?using Cohere’s Generate.?He?broke it down in his Medium article, check it out!!?
4. Misbah#5820 shipped?Ask Qran, an AI-powered search & references from the Quran. It uses Cohere’s multilingual embedding model for data and queries.??
5. FunShot#2838 launched?SymbiotAI, a marketplace of prompts where you can create and share your prompt with other people. The marketplace is integrated with Cohere’s Generate.
6. marmiteCloud#5923 shipped?AnyQuestions, a grounded QA answering tool using Cohere’s Generate and co.chat endpoints.?

Hungry for more demos? Make sure to join co:lab friday #13 for a live demo session. Details below!

co:lab friday #13: community demos showcase?

Mar 31, Friday at 12pm EST

Register here

What's New on the Blog

What is Semantic Search?

In this article, learn how to create a semantic search model using embeddings and similarity to efficiently search documents with a query.

What Is Attention in Language Models?

Self-attention models help language models overcome the challenge of ambiguous words, which have multiple meanings depending on the context. By using the context of the sentence, self-attention models help the model choose the correct meaning of the word.

Sentence Transformers and Embedding Evaluation

In this talk, Nils Reimers, the creator of Sentence-BERT and an expert in NLP, discusses Sentence Transformers and his experience in developing this widely used tool. In this informative conversation, learn about evaluating embeddings through works like MTEB & BEIR.

Read more

Get Started

We want to make NLP useful and accessible to anyone who needs it. Whether you're a beginner or an expert, we're here to help.

Register Now

Updates From Cohere For AI??

• We’re welcoming?Sasha Rush?(Associate Professor at Cornell University, Researcher at HuggingFace) as our next guest in the Cohere For AI?Fireside Chats series! Join us on?March 15th at 11am ET?to hear about Sasha’s research journey:?sign up here.?
• C4AI heads to Montevideo, Uruguay March 6-10 for?Khipu 2023.?Sara Hooker, Head of Cohere for AI, is a keynote speaker and C4AI Scholar?Luiza Pozzobon?is presenting her research! Keep an eye on our?Twitter?to see where you can say hi to the C4AI team during the conference and keep up with our latest research papers on our?website!?
• Join our Community! The C4AI Community includes ML-minded members from?100 countries?and counting! We want to keep growing our?fantastic events and programs: complete the?community application form to join in.?

Join Cohere's Research Community

Hosted Events

Future of Data & AI virtual conference: Large Language Models for Real-World Applications by Software Engineer, Hemant Jain

??? 2 Mar, Thursday - 3pm PT / 6pm ET

Sign up?here

Talking Language #5:?ML Explainability & Language Model UI by Jay Alammar and Hima Lakkaraju

??? 9 Mar, Thursday - 8am PT / 11am ET

Sign up?here

Multilingual Search Hackathon: Cohere x Qdrant

??? 10 - 17 Mar

Sign up?here

Multilingual & Cross-Lingual Search by Nils Reimers

??? 23 Mar, Thursday - 8am PT / 11am ET

Sign up?here

Co:lab friday 3.0 - Community x Cohere

??? 31 Mar, Friday - 9am PT / 12pm ET

Sign up?here

NLP JumpStart Series | Vector Search and Embeddings in Organizations: Harnessing Language Models and Vector Indexing, in partnership with Weavite. Hosted virtually on March 30 at 10:00 am PT.?Reserve your Spot!?

Join the Cohere co:mmunity on Discord!

Want to explore more demos and collaborate with other NLP practitioners??Join in on the conversation on our Discord