Get the best of
The Daily Bundle in your
inbox every week

Generative AIs: Size Does Matter

By Arooj Shakeel
October 5, 2023

Get the best of The Daily Bundle in your inbox every week

How are small generative AIs reshaping the world of software development? When small generative AIs are compared to large ones, it’s like comparing a pocket knife to a machete. Small AIs are efficient and ideal for quick, precise tasks. Large AIs excel at producing complex content, but they require more resources and may require more work to integrate. Both small and large-scale AIs have advantages. The specific requirements of your AI system determine the option you select. Small generative AIs are ideal for making quick decisions and performing specific tasks, whereas large ones tackle more difficult problems. 

Although it is important, your generative AI model’s size is not the only factor. There are applications for both small and large models. Begin small, experiment, collaborate, and always double-check the results. Accept the potential of generative AI and allow creativity to flourish!

How Generative AI, such as ChatGPT, Is Already Changing Businesses 

With a market value of USD 8 billion and an anticipated CAGR of 34.6% by 2030, the global generative AI market is poised to undergo a turning point. With over 85 million jobs expected to be unfilled by then, more intelligent operations based on AI and automation are required to deliver the efficiency, effectiveness, and experiences that business leaders and stakeholders expect. Generative AI offers a compelling opportunity to supplement employee efforts and increase enterprise productivity. However, as C-suite executives investigate generative AI solutions, they are discovering new questions, such as: Which use cases will provide the most value to my business? Which AI technology is best suited to my requirements? Is it safe? Is it long-term? 

After years of working with foundation models, IBM Consulting, IBM Technology, and IBM Research have developed a solid understanding of what it takes to derive value from responsibly deploying AI across the enterprise. 

Differences between existing enterprise AI and new generative AI Capabilities 

In enterprises, Generative AI, as the name implies, generates images, music, speech, code, video, or text while interpreting and manipulating pre-existing data. Generative AI is not new; the machine-learning techniques that power it have evolved over the last decade. The most recent method is based on a neural network architecture known as “transformers.” Large foundation models emerged from the combination of transformer architecture and unsupervised learning, outperforming existing benchmarks capable of handling multiple data modalities. 

Due to the fact that they act as the basis for the creation of smaller, more intricate models, these big models are referred to as foundational models. By building upon a foundation model, we can develop more specialized and complex models suited to particular use cases or domains. Early models, such as the GPT-3, BERT, T5, and DALL-E, demonstrated what was possible: enter a short prompt, and the system generates an entire essay, or a complex image, based on your parameters. Large Language Models (LLMs) were explicitly trained on large amounts of text data for NLP tasks and had a large number of parameters, typically in excess of 100 million. They make it easier to process and generate natural language text for a variety of tasks. 

Each model has advantages and disadvantages, and which one to use depends on the specific NLP task at hand as well as the characteristics of the data being analyzed. Choosing the best LLM for a specific job requires knowledge of LLMs. BERT is used for task classification, question answering, and named entity recognition. 

It is designed to understand bidirectional relationships between words in a sentence. GPT, on the other hand, is a unidirectional transformer-based model that is used primarily for text generation tasks such as language translation, summarization, and content creation. T5 is also a transformer-based model, but unlike BERT and GPT, it is trained using a text-to-text approach and can be fine-tuned for a variety of natural language processing tasks.

Acceleration and Shorter Time to Value 

Because they are pre-trained on massive amounts of data, these foundation models significantly accelerate the AI development lifecycle, allowing businesses to focus on fine-tuning for their specific use cases. Enterprises are able to reduce the time to value from months to weeks by using foundation models rather than creating unique NLP models for each domain. In client engagements, IBM Consulting has seen a 70% reduction in time to value for NLP use cases such as call center transcript summarization, analyzing reviews, and more. 

Responsible Implementation of Foundation Models 

Given the cost of training and maintaining foundation models, enterprises must decide how to incorporate and deploy them for their use cases. There are use case-specific considerations and decision points regarding cost, effort, data privacy, intellectual property, and security. One or more deployment options can be used within an enterprise while balancing these decision points. Foundation models will dramatically accelerate AI adoption in business by lowering labeling requirements, making it easier for businesses to experiment with AI, build efficient AI-driven automation and applications, and deploy AI in a broader range of mission-critical situations. IBM Consulting aims to bring the power of foundation models to every enterprise in a frictionless hybrid-cloud environment. 

Notify of
Inline Feedbacks
View all comments

How useful was this post?

Click on a star to rate it!

Back to

Get the best of The Daily Bundle in your inbox every week