AI21 nabs $53 million more to challenge OpenAI in enterprise generative AI

AI21, a company that competes with OpenAI to serve enterprise companies with generative AI systems, has chosen an opportune time to announce more funding.

While competitor OpenAI is reeling from internal chaos and revolt, and may fall apart after the majority of its employees have supported a letter calling upon that company’s board to resign, AI21 has raised $53 million more for its series C round of funding, at a valuation of $1.4 billion, and brings the round to a total of $208 million. The company has now raised $336 million.

OpenAI is one of a handful of other companies, including Cohere and Anthropic, that offer large language model (LLM)-driven AI systems to the enterprise that compete with OpenAI, but which are not open-sourced. 

And with OpenAI looking very unstable, enterprise companies looking for safety may we decide to hesitate before continuing with OpenAI products, and consider other options.

Companies need choice, robustness and safety, said Yoav Shoham, co-founder and co-CEO in an interview with VentureBeat. He said A121 works closely with enterprise companies, offering a sort of “white-glove” service to them to ensure they have what they need, and that’s not something OpenAI can do.

The company has 250 employees, and is based in Tel Aviv. The latest investors include Intel Capital and Comcast Ventures. Those investors join numerous other investors, including big names like Nvidia and Google.

My initial interview with Shoham took place a few days before the weekend’s events that threw OpenAI into turmoil. He declined to comment on what the OpenAI mess means for his company when I closed the loop with him yesterday. But in his comments even before the OpenAI blowup happened, Shoham expressed confidence about his company’s position in the market versus OpenAI.

AI21 also said it had appointed the former US Ambassador to Israel and former leader of Morgan Stanely, Tom Nides, as a board member. 

OpenAI is the main company AI21 sees when competing for deals, Shoham said, but it’s generally not seeing anyone else. His company does well against OpenAI when it does see them, he added.

OpenAI did well by creating excitement with its general-purpose LLMs like ChatGPT, which perform well in chat format, and make it easy for hundreds of millions of people to play with. But the chat interface is usually not the best way to work with large language models, Shoham said.

ChatGPT for the enterprise ChatGPT is not a model, he said. It is an application under which there is a model that calls for a chat modality. “It is sometimes right, sometimes totally not what you want.”

By sticking with general purpose offering, OpenAI is “betting on brute force” and lags at doing specific tasks very well, but that specificity is usually what enterprise companies need. He said OpenAI is not set up business-wise to work closely with companies: “It’s just not their model,” Shoham said. OpenAI works at scale, serving more than 100 million users, he said.

Instead, enterprise companies require robustly programmed applications for deployment, which requires a lot more work. He said scripting tools like Langchain that developers have adopted to pull in OpenAI’s LLMs via an API “point in the right direction, but true enterprise AI requires full programs to be written.” He added: “In two years, I bet we won’t be talking about LLMs, but about ‘AI systems,’ Shoham said. 

He said OpenAI’s big price drop for its services was the most significant announcement from OpenAI’s DevDay event two weeks ago. It showed the company’s willingness to lose money to grab market share — which isn’t necessarily a bad strategy, Shoham said.

AI21, by contrast, is building both general-purpose LLMs and task-specific large language models, with algorithms around them that make them robust for those tasks, Shoham said. A121’s summarization model, for example, was ranked better than GPT-4, ChatGPT and Claude by a large margin, when tested by a large financial institution, Shoham said, declining to name the company. 

The difficulty of working with LLMs in the enterprise has meant that those companies have been slower to adopt generative AI. However, enterprise interest “has exploded over the last six months,” he said. “Companies have moved from sporadic experimentation to mass experimentation….Within a year you’re going to see real money.”

Author: Matt Marshall
Source: Venturebeat
Reviewed By: Editorial Team