Promoted Content

Israeli startup CEOs explain how they aim to make AI better, cheaper and faster

OurCrowd investors grill Israeli tech leaders on the business outlook for technology that is sweeping the world – and the markets

Moshe Tanach, the Founder and CEO of Israeli artificial intelligence pioneer NeuReality, compares AI to a hobbled race car.

With its full potential unleashed. AI stands to be as revolutionary as the introduction of personal computing and the commercialization of the internet. McKinsey calls generative AI “the next productivity frontier” and projects it will add trillions of dollars annually to the global economy.
But for now, chip costs and infrastructure are major barriers to realizing the immense possibilities of the technology, Tanach says.

“It’s like you want to replace your car engine with a Formula One engine, but your car will not keep up with the power,” Tanach told ‘Investing in AI’,  an online event hosted by OurCrowd, the Jerusalem-based investing platform for startups.

The CEOs of four cutting-edge AI startups – NeuReality, One AI, Hailo and PolyN Technology – discussed the latest developments in AI technology before answering questions from a worldwide audience of investors. The event is now available to stream via the OurCrowd platform.

OurCrowd has also launched an AI Fund, enabling investors to participate in a range of promising startups while they are still operating as private companies.

NeuReality has set out to slash infrastructure costs and significantly boost performance by overhauling AI architecture to “make it easy, affordable and faster,” enabling AI to be deployed “everywhere it can make a difference,” Tanach says.

The company’s technology allows multiple AI chips to work in parallel to easily avoid system bottlenecks.

NeuReality’s AI-centric NR1 chip has passed quality assurance and moved to manufacturing – creating the world’s first AI-centric server on a chip. The chip demonstrated 10 times the performance at the same cost when compared to conventional CPU-centric systems.

NeuReality’s target market is cloud service providers or any big enterprise “that is using AI at scale and is suffering from the cost of the infrastructure and from the complexity of leveraging the next evolution processors for deep learning,” Tanach says.

Inefficient, inaccurate

Generative AI like Chat GPT or Google’s Bard are amazing tools for individuals, but they are inefficient for business. The materials they produce are often inaccurate. The systems themselves are not private, efficient, scalable or secure enough to be used by businesses.

One AI enables businesses to rapidly deploy cutting-edge AI capabilities in products and services. Its platform features AI pre-trained for common business uses. The technology is optimized for speed, scalability and cost-effectiveness, while helping to prevent biased or harmful content.

One AI’s OneAgent technology responds solely using up-to-date, internal data, with built-in fact-checking. And its language analytics provide structured insights on action items, business entities and metadata.

“There’s a very common misconception that AI is something that you just deploy and it just works,” says Amit Ben, One AI’s Co-founder and CEO, a veteran of the AI sector. “It is actually notoriously difficult to get AI to work in large scale and in high accuracy and in high-value deployments. Most companies don’t actually get to production with their AI.”

This is One AI’s first year of sales, and its revenue is growing fast.

“Our AI is dramatically more efficient both from performance and cost metrics,” Ben says.

Edge processing

Much of AI computing occurs in large data centers. But there is also a need for processing on the device itself – including semi-autonomous vehicles that need real-time traffic data, or facial recognition technology on a smartphone.

AI computing on devices use edge processing chips, but existing chips do not have enough power for high-level AI computing. Hailo has developed specialized AI accelerators for edge devices to meet the growing demand.

“We are focusing on the edge,” where devices are being deployed in scale, by the tens of thousands, if not millions, says Orr Danon, Hailo Founder and CEO. “We believe like many others this is the biggest source of growth in the market in the coming years.”

Danon expects Hailo’s revenue to grow dramatically this year.

The company’s revolutionary architecture allows smart devices to perform sophisticated deep learning tasks on the edge without the need for expensive large data centers that consume huge amounts of power.

Internet of things

The chips are designed for applications across an array of fields including security, retail, industrial automation, automotive, and the internet of things of connected devices, or IoT.

Danon predicts another wave of AI adoption and implementation in the next year or two, pushed from the cloud and proliferating at the edge.

“You’re already seeing industry leaders talking about it, and the applications are coming,” Danon says.
PolyN Technology, a fabless semiconductor company, is creating an even smaller edge processing chip to work with wearables and other small devices.

Using analog circuitry that imitates the brain, PolyN provides fast and easy conversion of trained neural networks into tiny AI silicon chips with ultra-low power consumption and low latency.

Lower cost

This allows AI computations to be performed directly on a device, without the need to send data to the cloud or a remote server.

The analog AI chips boast accuracy surpassing that of digital signal processing chips used in data centers powering generative AI platforms, at far lower cost. They also can be highly customized for any specific application.

“All other players represented in the market build fixed processor cores,” says Alexander Timofeev, PolyN’s CEO and Founder. “But when you create a neural net and train it, you don’t know what final size you will get. And for a fixed chip, it’s a problem. For some applications it’s too big, for some applications it’s too small. We don’t have this problem. Because we generate the core from the final model, we always fit it to the final model’s size. That is our fundamentally new idea.”

Goodyear is a key customer for the chip, which can be embedded in tires to detect surface conditions and the condition of the tire itself. PolyN is also collaborating with Infineon, a global leader in semiconductor solutions for energy management, smart mobility and communications. A major global TV brand has ordered a prototype and plans a launch.

PolyN is targeting “real players who know marketing requirements and can define product requirements,” Timofeev says.

To stream ‘Investing in AI’ on the OurCrowd platform, click here:Stream ‘Investing in AI’

read more: