Amidst growing geopolitical influences shaping the AI landscape, Red Hat is championing an open-source, collaborative approach focused on small language models (SLMs) to address concerns surrounding transparency, environmental impact, and data sovereignty. Julio Guijarro, CTO for EMEA at Red Hat, emphasizes the critical need for AI education and understanding, particularly highlighting the risks associated with ‘black box’ closed-source LLMs, including language support limitations, data sovereignty issues, and compromised trust.
Red Hat advocates for deploying SLMs locally or in hybrid cloud environments on standard hardware. This approach offers greater efficiency and performance for specific tasks while minimizing computational resource requirements. Organizations gain enhanced control over business-critical data and can adapt swiftly to evolving information.
Guijarro points out the potential for LLMs to become quickly outdated and incur substantial hidden costs due to their iterative development. By running models on-premise, organizations can limit expenses to infrastructure, avoiding per-query charges. Red Hat is actively optimizing models for standard hardware, streamlining them for specific use cases. Their support for vLLM, an inference engine project, further facilitates efficient model interaction across diverse environments.
Leveraging local, user-relevant data enables tailored AI outcomes, making it accessible to diverse language communities. Addressing latency issues and establishing trust remain paramount. Red Hat strongly promotes open platforms, tools, and models to foster transparency and collaboration, aiming to democratize AI by empowering users to replicate, fine-tune, and deploy models.
Red Hat’s acquisition of Neural Magic and the collaborative release of InstructLab with IBM Research underscore their commitment to scaling AI, improving inference performance, and enabling non-data scientists to leverage their business expertise in AI application development.
Despite potential economic headwinds in the AI market, Red Hat envisions a future where AI is use case-specific, open source, and universally accessible. As CEO Matt Hicks affirms, “The future of AI is open.”
Photo by Luis Quintero on Pexels
Photos provided by Pexels