Building generative artificial intelligence requires more than just relying on huge models

The rapid rise of generative artificial intelligence (GenAI) has companies scrambling to find new and innovative ways to harness the power of this technology in business applications. Many enterprises believe that large language models (LLMs) have reshaped the way AI-driven business applications are built. All that is required is to feed data into the LLM model and it will do the job. However, things are not that easy

Research and advisory firm Forrester has released a new report highlighting that GenAI commercial applications will require more than A general LLM. Even the most carefully tuned and cue-trained LLM may not be sufficient to build and safely run GenAI-based applications. This simplistic approach does not allow organizations to use all their proprietary knowledge to work. It also presents other risks, including scalability, security and cost issues.

Forrester’s report examines how 15 of the largest service providers use GenAI to help more than 2,000 companies around the world write GenAI-powered business applications. The report's findings suggest that enterprises need to assemble a "layers, gates and pipes" architecture to run GenAI-based applications safely and efficiently.

"Layers, Gates, and Pipes" architecture leverages resources from many intelligent layers to tie together internal and external functionality. It also requires input and output control gates to protect people, the company, and the model itself. Additionally, it requires an application pipeline to prompt, embed, and orchestrate the intelligence layer to transform requests into outputs. Finally, a test and learn loop is needed to test and monitor results and make adjustments accordingly.

When digging deeper into the elements of the “layers, gates and pipes” architecture, the report states that the intelligence layer includes a wide range of capabilities, including general, embedded and specialized GenAI models.

Intelligent resources that organizations should create and manage themselves include software applications, AI/ML models, private GenAI models, structured and unstructured data, and people’s cues and behaviors. Sources of intelligence that organizations should obtain from vendors should include domain-specific GenAI models, public GenAI tools, and bundled GenAI models, such as SaaS applications.

Use input gates to block bad requests, false prompts, and dangerous searches. It can also turn vague requests into answerable prompts. Output gates help validate the output of issues based on aspects such as compliance needs and security

Application pipelines are used to connect all of this together, via API-first workflows. They help combine resources seamlessly from the intelligence layer for smooth end-to-end flow. The final element of architecture is testing through a feedback loop of testing. They help build trust, confidence and effectiveness in applications

According to the Forrester report, enterprises now have the opportunity to assemble applications from disparate parts because they can Build a complete architecture to support GenAI applications in the next few years. Only with proper attention can enterprises fully benefit from the power of GenAI business applications

The above is the detailed content of Building generative artificial intelligence requires more than just relying on huge models. For more information, please follow other related articles on the PHP Chinese website!