Next Gen Tech , Quantum Computing

How Quantum Computing-Inspired Software Is Optimizing LLMs

Sam Altman, CEO of OpenAI, said the cost of training large language models from scratch can reach as much as $100 million, and that cost is forecasted to double every 10 months. That means it will soon be financially unviable for a private enterprise to train AI models. Multiverse Computing, a Spanish quantum computing software firm, aims to bring the cost down with its quantum-inspired software platform and optimization solutions.

Quantum computing is emerging as a transformative solution to reduce costs and meet the massive computational demands of the training process. Multiverse Computing, one of the largest quantum software companies in the world, aims to bring the cost down with its quantum-inspired software platform and optimization solutions.

"Our quantum-inspired software shrinks the size of the model while retaining accuracy. This makes the model smaller and more efficient, reducing operational costs and opens up more use cases in manufacturing, self-driving cars and remote operations that may not have a dedicated internet connection," said Enrique Lizaso Olmos, co-founder and CEO, Multiverse Computing. "Our optimization solutions can significantly improve the speed and precision of classical solutions."

Optimizing LLMs can reduce their size, saving money for enterprises that have to invest in expensive GPUs, memory and storage space for processing. The company claims its CompactifAI platform uses quantum-inspired tensor networks to reduce the number of parameters in a model, reducing the size of LLMs by more than 70%. This tool significantly reduces energy demands required to fine-tune and run large foundational models and is designed to reduce development costs making it easier to integrate these models into more digital services or in the edge.

Projects and Use Cases

Multiverse Computing is already applying its quantum and quantum-inspired solutions across various sectors. The company's current research focuses on improving AI efficiency with tensor networks and providing machine learning models with improved explainability. The company is also working with the German Aerospace Center to use quantum simulation to improve the detection capabilities of superconducting nanowire single-photon detectors.

Multiverse Computing has been applying its quantum-inspired algorithms in finance, artificial intelligence, energy, manufacturing, cybersecurity, defense and more. The company's software platform, Singularity, is designed to allow users without prior knowledge of quantum computing to use quantum algorithms.

"We have completed projects with a manufacturing company to reduce manufacturing defects and with banks to prove how Singularity can identify customers at risk of a downgrade in credit rating," Lizaso Olmos said.

Addressing Industry-Specific Challenges

In a blog post, Michel Kurek, CEO France, Multiverse Computing, cited the results of two proof-of-concept projects by Crédit Agricole CIB, Multiverse Computing and Pasqal. At the center of these two projects, he said, are real business problems: predicting changes in creditworthiness and pricing derivatives.

"These projects push the boundaries of what is possible with quantum and quantum-inspired solutions," Kurek wrote.

In 2021, Multiverse Computing announced 12.5 million euros in funding from the European Innovation Council Accelerator program. In April 2022, the company partnered with the Bank of Canada to explore how quantum computing can simulate cryptocurrency adoption. This research made Canada the first G7 country to explore the model of complex networks and cryptocurrencies through quantum computing. That July, Multiverse Computing partnered with Bosch to integrate quantum algorithms into digital twin simulation workflow to scale simulations more efficiently and improve the accuracy of defect detection. Later that year, BASF used Multiverse’s Singularity to develop models for foreign exchange trading optimization between U.S. and EU currencies.

AI for Green Transition

Multiverse is also leveraging its solutions to drive green transition by addressing key problems such as electrical distribution in renewable energies, battery design and hydrogen production - which cannot effectively be solved without new algorithms for optimization and AI.

"Our tensor network methods rival traditional deep learning models while providing better interpretability and explainability, opening up the 'black box' of AI," Lizaso Olmos said.

Lizaso Olmos noted "many opportunities" in the energy sector for quantum machine learning to improve utility analytics and for quantum solutions to ease the transition to green energy and help to scale smart grids.

Multiverse Computing is also working with quantum initiatives led by the U.K. and German governments, as these and other countries work with private companies to build quantum ecosystems.

Future Prospects

Lizaso Olmos said use cases include portfolio optimization and risk management, defect detection in manufacturing production lines, drug discovery, vehicle routing and weather simulation. He expects the financial industry to be among the early adopters along with the auto industry and energy sector.

"We have developed a platform with Moody’s Analytics to allow large financial companies to run quantum and quantum-inspired algorithms side by side. The platform is incorporated into existing production systems and allows users to compare the results of classical and quantum-inspired algorithms," Lizaso Olmos said.

In the realm of cybersecurity, Multiverse Computing is focused on advancing the detection of attacks with explainable methods enabled by quantum-inspired solutions.