Highlights:

  • Toronto-based Cohere offers generative AI models via application programming interfaces stored in the cloud.
  • The startup offers the option to fine-tune its models by training them on unique datasets for businesses with complex requirements.

Cohere Inc., a startup in generative AI, revealed that it has secured USD 270 million in funding from a group of illustrious tech investors.

The venture capital arm of Salesforce.com Inc., Nvidia Corp., Salesforce Inc.’s venture capital arm, and Oracle Corp. all participated in the Series C round of funding. SentinelOne Inc., a publicly traded cybersecurity business, also contributed. More than a dozen others backed, including Inovia Capital, which headed the round, joined the four companies.

Toronto-based Cohere offers generative AI models via application programming interfaces stored in the cloud. Chief Executive Officer Aidan Gomez, a former Google LLC AI researcher, oversees it. While working with the search giant, Gomez co-authored the scholarly paper that introduced the Transformers concept, one of the most essential machine learning discoveries in recent memory.

Transformers are a sort of neural network designed specifically for text processing. They are also substantially more accurate than prior models in understanding and producing language and other activities like writing code. Many of the most sophisticated large language models in the market, such as OpenAI LLC’s GPT-4, are powered by this technology.

A neural network searches the surrounding text for contextual cues to establish a sentence’s meaning. The vanishing gradient problem, which existed before the release of Transformers, caused neural networks to lose a large amount of the gathered contextual cues. They could only be so accurate.

On the other hand, Transformers can recall much contextual information accurately. Additionally, they rank the importance of each contextual piece of information. They can comprehend text and code more accurately than prior models.

Gomez reported, “It’s still super early in the technology’s trajectory. I think in the next year, we’re gonna see some major changes. You’ve already seen the very first hints of that with stuff like Bing Chat, where you augment these dialog models with an external knowledge base. Now the models can be kept up to the date to the millisecond because they can search the web.”

Gomez released Cohere with Nick Frosst and Ivan Zhang two years after the Transformers paper’s release. The business had already garnered around USD 183 million from investors before the current fundraising round.

A neural network named Command is the highlight of Cohere’s lineup of generative AI models. It is designed to produce text in response to user-provided cues. Command has the ability to create marketing content, such as product descriptions, as well as document summaries and customer service chatbots.

Cohere also provides several models that are more focused. Another collection of models is intended to fuel online search engines, while one set is targeted at text summarizing tasks. Additionally, it provides neural networks tailored to transform words into embeddings, a type of data that AI systems can handle more quickly than raw text.

The startup offers the option to fine-tune its models by training them on unique datasets for businesses with complex requirements. Users can select from a variety of deployment options with Cohere. The startup claims its models may be used internally by businesses and on all the leading public clouds.

“Language is everywhere. These large language models can be deployed all over the place, into every single industry sector,” Gomez stated.

According to the most recent fundraising round, Cohere is worth USD 2.2 billion. The funding comes just a few weeks after Spark Capital-led USD 450 million investment in Anthropic, a startup creating sophisticated generative AI models. The business was reportedly valued at USD 4.1 billion after the raise.