Revolutionizing AI on Smartphones: Spain's Multiverse Computing Secures $215M to Miniaturize Deep Learning Models

A Quantum Leap for AI: How a Spanish Startup is Shrinking Models Without Sacrificing Performance

Multiverse Computing's Breakthrough Compression Technology

In a bold move that could revolutionize the AI industry, a Spanish startup has secured $215 million in funding based on a groundbreaking claim: they can reduce large language models by 95% without compromising performance.

Multiverse Computing's CompactifAI technology uses quantum physics concepts to condense AI models to smartphone-friendly sizes. Their compressed Llama-2 7B model reportedly runs 25% faster while using 70% fewer parameters, with only a 2-3% accuracy drop.

Unlike traditional compression methods, CompactifAI employs tensor networks - mathematical structures originally developed by physicists to track particle interactions. This innovative approach preserves meaningful correlations while eliminating redundancy, achieving reductions where conventional techniques fail.

Tackling AI's Elephant-Sized Problem

Multiverse's innovation addresses a critical challenge in AI development. The industry's trend toward increasingly large models has created a barrier where only tech giants can afford the specialized infrastructure needed to run them.

"We can now profile neural networks to eliminate billions of spurious correlations and truly optimize AI models," explains Román Orús, Multiverse's chief scientific officer.

Their selective compression technique focuses on less critical deeper layers, achieving performance metrics that surpass traditional methods. After a short retraining period, compressed models reportedly operate twice as fast at a fraction of the original cost.

Quantum-Inspired Edge AI and Beyond

This breakthrough extends beyond cost efficiency. With backing from HP Tech Ventures, the technology could enable sophisticated AI models to run locally on devices rather than requiring cloud infrastructure.

"Multiverse's approach can deliver enhanced performance, personalization, privacy and cost efficiency to businesses of all sizes," notes Tuan Tran, HP's President of Technology and Innovation.

As AI competition intensifies, this quantum-inspired compression technology may pave the way for the next generation of accessible, high-performance artificial intelligence applications.