Recently, Silicon Valley displayed a rare mixed emotion towards achieving AGI. This became more apparent during the recent banter between Meta’s Yann LeCun and xAI’s Elon Musk.
However, unlike its contemporaries like OpenAI or Antropic, Google DeepMind or even xAI, the Canadian startup Cohere’s AI ambitions are not focused on AGI, at least for now.
“We remain concentrated on designing AI solutions that deliver better workforce and customer experiences for businesses today rather than pursuing abstract concepts like AGI,” Saurabh Baji, SVP of engineering, Cohere, told AIM.
Some experts predict it will arrive next year, others say by 2029, and some think it will never happen. Some claimed that AGI has already been achieved multiple times in 2024 with projects like Devin and Claude-3 Opus, leading to the belief that future achievements might be overlooked.
Cohere offers models in three categories: Embed, Command, and Rerank, each designed for specific use cases and customizable.
At CouldWorld 2023, co-founder and chief executive officer Aidan Gomez stated that the company is focusing more on embedding models, which are expected to perform twice as well as competitors on varied and noisy datasets.
Unlike generative models trained on public internet data, embedding models are trained on enterprise data, retrieving information from specific data sources.
Enterprise Needs is Cohere’s Focus
In April, the startup launched Command-R, a language model for enterprise use with a 128k context window and support for ten languages, performing tasks like RAG and tool integration.
Building on its success, they introduced Command R+, which ranked 6th on the Arena leaderboard, matching GPT-4-0314 based on over 13,000 human votes. It is regarded as one of the best open models.
“Command R+ is our most powerful model to date targeted for use cases needing complex reasoning and is highly performant for tool use and building AI agents,” said Baji. “We are seeing growing demand from customers for scalable models that customers can use to bring AI applications into large-scale production,” he added.
This focus on enterprise-critical features allows Cohere to cater to both large, and complex enterprises like Oracle, Accenture, and McKinsey and fast-growing startups like Borderless AI and AtomicWork, helping them streamline operations and boost productivity.
For example, Oracle integrates Cohere’s models into applications for finance, supply chain, HR, sales, marketing, and customer service, while AtomicWork uses Command R+ and Rerank models to boost IT support efficiency.
The company recently introduced fine-tuning for Command R, which surpasses larger, more expensive models.
The company’s technology enhances business processes in financial services, technology, and retail through enterprise search, copy generation, and AI assistants. It offers flexible deployment options, including on-premises solutions for highly regulated industries, ensuring data privacy and security.
Up Next
“At Cohere, we are continually working to iterate and improve our models to adapt to the unique needs of our enterprise customers. We want to ensure that our products solve real-world business problems today and are designed to excel in an enterprise environment,” said Baji.
The company is currently working on making the latest Command R models widely available on platforms like Microsoft Azure, Amazon Bedrock and Oracle Cloud Infrastructure (OCI).
Looking ahead, Cohere wants to continue to innovate, focusing on refining its models to better meet enterprise customers’ needs. It also unveiled the Cohere Toolkit, which aims to speed up the development of generative AI applications.
By the end of March, the company generated $35 million in annualized revenue, up from $13 million last year. The startup recently raised $450 million from investors, including NVIDIA, Salesforce Ventures, Cisco, and PSP Investments, pushing the valuation to $5 billion.
Based in Seattle, Washington, Baji has been with Cohere for about two years. “With Cohere’s world-class team, singular focus and clear strategy, it is well positioned to lead the effort to accelerate wider enterprise adoption,” concluded Baji.
