Generative AI is driving a gold rush for performance tools

Generative AI work and investment have blossomed in recent months, accelerated by ChatGPT and other LLMs introduced in November and afterwards.

Some might even call what is happening more of an explosion than a blossoming or perhaps a mini-revolution, if not something bigger. Dare we call the trend another technology revolution, or have the cynics completely taken over?

“We are only at the beginning of that [generative AI] revolution and you can never underestimate where you are in the cycle of that revolution. It’s kind of what happened with the internet,” said Yonatan Geifman, founder and CEO of Deci, a startup based in Tel Aviv.   

The company, founded more than three years ago, now has 60 workers, with 50 data scientists and others working on deep-learning software out of Israel.  A push right now is on to hire sales and marketing staff in North America and Europe to keep up with strong demand, Geifman said in an interview with Fierce Electronics via Teams.

In simple terms, Deci is helping companies build and optimize their AI models faster and with better performance. The company boasts an 80% shorter development process with 30% lower development costs per model on average and a quintupling of inference acceleration with one-fifth the cost.

Some customers and partners on its web site are biggies: Intel, Ring Central, HPE, Adobe, Applied Materials, Sight Diagnostics and more. Deci has raised $55 million from investors thus far.

“Usually, our customers are already working with AI and have a use case they are trying to solve,” he said. Since the recent LLM publicity and craze, startups are showing up in addition to the bigger development houses. “Some are very small companies and sometimes scale is not the main issue but just to make it work with reasonable performance.” Others are larger and have services where they want to add features.

Customers may want writing assistance or have image generation needs with image editing and video generation. “There are many, many new apps that are really interesting and creative and we see them starting to take them to production,” Geifman said.

At Deci, the current challenge is to keep up with demand, part of which has come from the dubious macroeconomic climate. “With the impact of tech layoffs, some companies are trying to get as much as they can with smaller teams and when facing that, they prefer to buy infrastructure products compared to something they can do inhouse,” he said. 

Competition is very fragmented, he added. While not taking the same approach, Octoml is a potential competitor, he said, but some analysts even put MATLAB and IBM Watson Studio into a similar grouping.

Deci’s subscription pricing starts at $100,000 and is broken into three offerings: standard, professional and enterprise, according to its website. The software Deci produces is AutoNAC, short for Automated Neural Architecture Construction, a proprietary technology that is a Neural Architecture Search algorithm.

A year ago, Deci announced it had achieved “breakthrough deep learning performance” using CPUs instead of the more traditional approach of using GPUs. The company said its AutoNAC automatically generated a new family of models for image classification, called DeciNets, for Intel’s Cascade Lake CPU.

While some ChatGPT competitors have emerged that have given users widely-publicized errors and questionable insights, the future for generative AI looks promising for Deci and, potentially, many competitors. In just one example of the kind of applications that are emerging, Microsoft unveiled on March 6 Dynamics 365 Copilot for Ai-powered assistance across business functions such as sales, services, marketing, operations and supply chain.

RELATED: Nvidia’s happy to share AI stage with ChatGPT after strong Q4