Oracle recently announced its ambitious goal of becoming a $65-billion company by 2026. To achieve this, it is betting big on Cohere, its billion-dollar generative AI baby, alongside multi-cloud strategy and healthcare initiatives. It even raised $270 million at a valuation of $2.2 billion in a series C round from Oracle, NVIDIA and Salesforce.
Unlike OpenAI, Anthropic, Google DeepMind, and Meta AI, which are all over the place, Cohere seems to be focused on meeting the enterprise needs. “Cohere has been oriented toward solving business problems, not toward the consumer space,” said Greg Palvik, senior vice president, OCI, at the Oracle CloudWorld 2023 at Las Vegas.
Ironically, the marriage between cloud providers and generative AI startups seems to be the new norm nowadays. It all started with Microsoft partnering with OpenAI. Recently, AWS announced a $4 billion investment in Anthropic. The amount was substantial enough for everyone to take notice, and many saw this alliance in the same light as Microsoft’s partnership with Azure and OpenAI.
However, very little is known about Cohere and Oracle. Surprisingly, they are moving ahead swiftly and silently without making much noise and delighting enterprise customers with generative updates across its products, solutions and services. This includes Oracle Fusion Cloud Applications Suite, Oracle NetSuite, and industry applications such as Oracle Cerner, alongside bringing Vector Store to MySQL HeatWave, which boasts generative AI capabilities.
Cohere is all about enterprise
Cohere is unlike any other generative AI startups out there. At CouldWorld 2023, Cohere co-founder Aidan Gomez spoke in great detail about how he sees Cohere as different from the rest of the competition. According to him, there are two types of models: generative models and embedding models.
Generative models are usually trained on publicly available data on the internet, whereas embedding models are typically trained on enterprise data where the model retrieves information from specific sources of data points. Cohere is focussing on building both, but it places a stronger emphasis on ’embedding modes’. He added that the new embeddings models will perform twice as well, compared to the competition on datasets that are heterogeneous and noisy.
Gomez believes that embedding models, combined with RAG (Retrieval, Augmented Generation), is going to solve most of the problems of enterprises. RAG is a relatively new AI technique that can improve the quality of generative AI by allowing LLMs to tap additional data resources without retraining. Interestingly, Patrick Lewis, who coined the term RAG while working at Meta, now works at Cohere. “He’s now at Cohere leading our RAG efforts alongside Sebastian Hofstätter. We’re super fortunate to have him,” said Gomez. Hofstätter has also worked on RAG during his PhD internship at Google Research.
Furthermore, he said that Cohere is coming up with new embedding models. “I am really excited to kind of pre-announce our new embedding models,” said Gomez, explaining how it is crucial for RAG – i.e. when the model makes that query to a database, the response is going to be of higher quality, and does not require training of a model as it updates the knowledge in real-time.
Oracle and Cohere Mean Business
In order to execute its plans, Cohere could not have found a better partner than Oracle. “We are cloud agnostic, but very closely partnered with Oracle,” said Martin Kon, COO, Cohere, at Oracle CloudWorld 2023, Las Vegas. Cohere service will form the basis for generative AI capabilities embedded across Oracle’s suite of SaaS applications, including Oracle Fusion Cloud Applications Suite, Oracle NetSuite, and industry applications such as Oracle Cerner.
Built on OCI, in collaboration with Cohere, the OCI Generative AI service will enable users to integrate LLMs in their own applications through an available API. Once generally available, this service and Cohere models will work seamlessly with AI Vector Search, a feature of Oracle Database 23c that provides retrieval augmented generation (RAG).
Just like how Microsoft leveraged generative AI capabilities from OpenAI, Oracle is planning to do the same with Cohere, but with so much clarity and coherence. The key difference is that while OpenAI targets both consumers and enterprises, Oracle and Cohere are exclusively focused on serving enterprise customers.