The Spanish startup Multiverse Computing takes a key step to democratize advanced AI. With its CompactifAI technology, it launches a self-service API portal and a demonstration application. Its goal is to offer compressed versions of large language models, enabling their local execution on enterprise hardware without relying on the cloud. This promises greater control, privacy, and a drastic reduction in computational costs.
CompactifAI: compression without loss of critical performance 🤖
The technical core is CompactifAI, a method that drastically reduces the size and requirements of AI models from large labs. Its API portal allows access to versions like HyperNova 60B, which maintains performance similar to the original with lower latency. The demonstration app includes the Gilda model, operating offline on devices with sufficient resources. This efficiency is crucial for intensive workflows like assisted programming, where local inference eliminates cloud costs and latency, accelerating processes and protecting sensitive data by not leaving the device.
Towards enterprise sovereignty in AI 🚀
This initiative signals a profound change: the transition from cloud-centralized AI to distributed and local AI. It reduces dependence on infrastructures of big tech companies, granting autonomy and technological sovereignty to enterprises. By making access to powerful models cheaper and portable, it could reduce the digital divide in AI, allowing more organizations to implement advanced solutions privately and efficiently, redefining business models in the digital era.
Can AI model compression for local execution, like Multiverse's CompactifAI, balance the scalability of artificial intelligence with technological sovereignty and data privacy? 💡
(PS: trying to ban a nickname on the internet is like trying to cover the sun with a finger... but digital)