Can AI Startup Cohere’s New Embed V3 Boost AI Productivity For Enterprises?

The demand for GenAI (Generative Artificial Intelligence) and Large Language Model (LLM) applications has spiked in the last few months. It’s no surprise too, given the capabilities they bring to the business table in reducing time and effort!

However, there have been a few issues with such technologies, especially ones that produce inaccurate or outdated results. With such barriers present, businesses are always on the lookout for improvements. We bet businesses will be excited at Cohere’s new announcement!

What’s it all about? Let’s explore:
 

What Is Cohere’s Latest Announcement About?

 

• In an announcement made on the company’s website, Cohere, a Canadian technology company focusing on artificial intelligence for enterprises and in large language models, announced the launch of Embed v3, the updated version of its embeddings model. 

• Embedding models AKA embeddings are mathematical representations of objects like text, audio, images, etc. that are used by machine learning models and semantic search algorithms. These objects or values are based on defining characteristics and enable machines to understand the connection between words and objects, so that they can find similar objects. Essentially, embeddings form the founding blocks of artificial intelligence!

• An excerpt from the announcement reads, “We're excited to introduce Embed v3, our latest and most advanced embeddings model. Embed v3 offers state-of-the-art performance per trusted MTEB and BEIR benchmarks.”

• Furthermore, embeddings are also used in RAG (Retrieval Augmented Generation), which is a crucial element in large language models in the enterprise sector. RAG helps by inserting an information retrieval system which feeds and offers control over data used by LLMs to formulate responses.

• RAG also enables the use of data that wasn’t part of the original training data. However, for organizations to make use of RAG, they must create embeddings of their documents, which must be stored in a vector database.

• Cohere will be releasing new English and multilingual Embed versions with either 1024 or 384 dimensions. Additionally, all models of the Cohere can be accessed via the company’s APIs.

 

How Will Embed V3 Help Businesses?

 

• Cohere is here to help create a future without a language barrier between humans and machines. 

• This is what Embed v3 looks to achieve, as it competes with OpenAI’s Ada and others, through superior performance and enhanced data compression, as per the company.

• The idea is to enable businesses to explore, generate, search and activate information in more intuitive and natural ways.

• As per the announcement, Cohere says, “With Embed v3, developers can immediately improve search applications that engage with real-world, noisy data and improve retrievals for retrieval-augmentation generation (RAG) systems.”

• Essentially, Embed’s updated RAG will help solve the “hallucinations” or false information generation issue LLMs have, which is a result of outdated information.

• This is further enhanced by Embed v3’s ability to evaluate and assess queries to match documents and rank higher quality documents on top.

• Additionally, Embed v3 comes with a compression-aware training method, which helps reduce costs when running custom vector databases as it allows users to handle billions of embeddings without increasing cloud infrastructure expenses.

• Embed v3 improves Cohere’s reranking feature, which allows “search applications to sort existing search results based on semantic similarities” and was added by the company to its API a few months ago.

Do you think Cohere’s Embed v3 can spark stronger competition in the GenAI and LLM space? Do you think competitors such as OpenAI and Anthropic will respond with an updated LLM soon?

Let us know in the comments below!