Published: 29th April 2024

Any early adoption of an emerging technology that promises a huge market advantage comes with risks. Generative AI promises organizations the potential for significant market differentiation, dramatic cost reduction, and a slew of other pluses, such as improved CX, but its safe implementation is by no means a given. Its dangers include potential resource overrun, customer-facing misfires, and significant PR fallout.

Recent mainstream media coverage of Canada Air’s AI bot mis-step and a New York lawyer’s submission of hallucinated case law show that, at least in the public’s perception, running effective AI instances leaves a great deal to be desired.

Perhaps the disparity between the technology’s potential and its real-world worst-case outcomes is down to the nature of decision-making in large organizations. In fact, an Innovation Catalysts study published this year found that 81% of business decision makers believe there are reasons to exclude the IT department from strategic business decision making. Sure, the IT function has a responsibility to investigate and advocate for technology’s benefits, but it could be argued that there may be broader enterprise concerns that need to be addressed which involves all stakeholders, including IT.

The advantages of deploying AI in workflows for data processing, creativity, and operations are well-known, although every use case varies according to the organization and its approach. But harnessing the technology in production where results are based on local data means considering safeguards around intellectual property and legal & compliance issues, plus the need to embed transparency into the solution. This transparency is important to satisfy regulatory authorities concerned over issues like data processing and sovereignty, as well as customers’ and service users’ concerns about privacy and data practice.

Trustworthy generative AI is a phrase that encompasses a set of smart services and practices that ensure safe operation: trustworthy, legally compliant, and transparent. Building those necessary elements is not a simple undertaking and represents a significant addition to the normal overheads associated with machine learning (compute and storage), and includes extra processes like query & response validation, bias monitoring, data sanitization, and provenance checking.

Source: Shutterstock

Some vendors offer pre-trained models that can form some of the basis of a GenAI solution. But until now, there has been nothing on the market where a solution includes data security, use and development guardrails, manageability, and vendor support. In short, those elements that are mandatory to transform what’s essentially experimental (and therefore has potential risk) into a reliable production-ready platform – which is what Dell and NVIDIA can now offer.

The AI industry is doing its best to address many of the needs of larger organizations that are concerned about some of the potential misfirings that a premature rollout of GenAI could create. NIST’s Artificial Intelligence Safety Institute Consortium (AISIC), for example, has come about to create safe and trustworthy artificial intelligence and comprises more than 200 bodies. It produces empirically backed standards for AI measurement and policy, so organizations leveraging GenAI have guides to safe and legal AI deployment.

NVIDIA, a key member of AISIC, now offers NeMo Guardrails which is designed to support enterprise data security and governance standards, acting as a two-way arbiter between user queries and AI responses.

In enterprise use cases, working with internal data also brings challenges with regard to an organization’s intellectual property. Without proper safeguards, any GenAI instance represents a potential danger to an organization’s ongoing viability. It’s with that challenge and those detailed above that Dell and NVIDIA have partnered to offer a GenAI system that boasts topical, safety and security features, producing the closest to a production-ready, drop-in GenAI solution currently available on the market.

Dell Technologies’ Generative AI Solutions encompass best-of-breed infrastructure designed to greatly simplify the adoption of generative AI for organizations that need the power of machine learning technologies to leverage the value of their digital assets without compromising their ethos, data, customers, or third parties.

Based around the Dell PowerEdge XE9680 GPU-accelerated server, it’s designed for generative AI training, model customization and large-scale inferencing. It comes with NVIDIA AI Enterprise software, which allows rapid deployment of production-ready models in local, hybrid, and remote computing topologies.

The Dell Generative AI Solution range is highly scalable, with hardware that can be expanded according to need, with eight NVIDIA H100 or A100 GPUs fully interconnected with NVLink. The air-cooled 6U devices offer any variation of local and remote deployment at a lower TCO than equitable processing power from other vendors.

With NVIDIA AI Workbench, developers can experience easy GPU environment setup and the freedom to work, manage, and collaborate across workstation and data center platforms regardless of skill-level.

The combination of hardware and software designed from the ground up for generative AI development and deployment, comes with guardrails, data governance, and security baked in. Together, the two mean that organizations can deploy powerful AI-based applications safely and responsibly.

Source: Shutterstock

Building trustworthy generative AI means greater buy-in from business decision-leaders outside IT, as many of their rightly-held concerns around the technology are addressed: transparent development and use, safeguarded IP and customer-facing responses, statutory compliance, and best-in-class operating costs.

To find out more about how the Dell Generative AI Solution portfolio takes machine learning to a fully-viable production setting, contact your nearest representative.

Dell Technologies: https://www.dell.com/en-sg/dt/solutions/artificial-intelligence/generative-ai.htm

NVIDIA: https://www.nvidia.com/en-sg/ai-data-science/generative-ai/