• The system, called Pareto, improves the efficiency of the Al chip by simplifying its operation, thereby reducing power consumption.
  • Successfully tested on the Al model of the main platform, Recogni aims to work with hardware companies to realise a wider range of applications.

OUR TAKE
The new patented method, called Pareto, could make the chips it uses to train and run AI systems smaller, faster and cheaper to operate. Outperforming existing such innovations in running large AI models could have a significant impact on the Al technology industry by making it more efficient and cost-effective.
— Iydia Ding, BTW reporter

What happened

Al chip and software startup Recogni on Tuesday unveiled a new computational approach that makes the chips it uses to train and run artificial intelligence systems smaller, faster and cheaper to operate. Backed by BMW, Bosch and venture capital firm Mayfield, Recogni has developed specialised chips and software to enable AI reasoning – the process by which trained Al models make predictions or decisions about new, unseen data.

The new patented system, called Pareto, uses a logarithmic approach that outperforms existing methods for running large AI models, the company said. Gilles Backhus, co-founder and vice president of artificial intelligence at Recogni, said it’s a huge leap forward in AI computing for all the KPIs (key performance indicators) that affect the design of silicon hardware systems. The company has tested Pareto on Al models developed by Meta Platforms, Stability AI and others.

Also read: ZPE launches new console server and NVIDIA platform enhancing AI workload

Also read: GitHub CEO advocates for competition and open source in AI

Why it’s important

Current AI models, such as OpenAI’s GPT-4 and Google’s Gemini, require hundreds of thousands of power-hungry maths operations for simple prompts from chatbots like ChatGPT. And Recogni says its new system converts these multiplication operations into addition, significantly reducing power consumption while maintaining accuracy.

The system improves the efficiency of the Al chip by converting complex operations into simpler ones, thereby reducing power consumption. By successfully testing the Al model on major platforms, Recogni aims to work with hardware companies to enable wider adoption. Advances in chip technology and its rollout will affect the dynamics of the industry as a whole.Recogni has expressed its willingness to roll out its technology around the world and is talking to companies that are putting their hardware in data centres and delivering it to the world.

These kinds of innovative developments in the underlying hardware will have a significant impact on the entire tech industry by making Al technology more efficient and cost-effective, providing more efficient technology support to all areas of society.