GenAI Build, FineTune, RAG...and the so many confusing options to roll out!

This is article 2 in the AI for Business Leaders.

Generative AI models use neural networks to identify the patterns and structures within existing data to generate new and original content on patterns and insights learned from a massive training dataset. Examples of Generative AI models include:

·?ChatGPT:?An AI language model developed by OpenAI that can answer questions and generate human-like responses from text prompts.

·?DALL-E 3:?Another AI model by OpenAI that can create images and artwork from text prompts.

·?Sora: ?Sora is an AI model from OpenAI that can create realistic and imaginative scenes from text instructions.

·?Google Gemini:?Google’s generative AI chatbot and can answer questions and generate text from prompts.

·?Claude : By Anthropic which collaborates closely with Amazon, it is another generative AI-based model described as a “friendly, enthusiastic colleague or personal assistant.

·?Midjourney:? This Gen AI model interprets text prompts to produce images and artwork, similar to DALL-E.

·?GitHub Copilot:?An AI-powered coding tool that suggests software code completion.

·?Llama:?Meta’s open-source large language model can be used to create conversational AI models for chatbots and virtual assistants.

·?Grok: Another Generative AI virtual assistant from Elon Musk’ new generative AI venture xAI.

·?BLOOM: From Hugging Face, is like ChatGPT but has been trained on 46 different languages and 13 programming languages.

·?Cohere:? Provides pre-built LLMs (Large Language Models) to perform common tasks on text input, such as: summarizing, classification, and finding the similarities in content aka. natural language processing (NLP).

Generative AI platforms are built on top of large language models (LLMs) and foundation models:

·?LLMs?are deep learning models that consume and train on massive datasets to create new combinations of text that mimic natural language based on its training data.

A key factor in how LLMs work is the way they represent words. Earlier forms of machine learning used a numerical table to represent each word. But this form of representation could not recognize relationships between words such as words with similar meanings. This limitation was overcome in LLMs using a technique commonly referred to as word embeddings, to consider context of the text and relationships between words in the text.

· Foundation models describe machine learning models trained on a broad spectrum of generalized and unlabeled data and capable of performing a wide variety of general tasks such as understanding language, generating text and images, and conversing in natural language. ?

A unique feature of foundation models is their adaptability. These models can perform a wide range of disparate tasks with a high degree of accuracy based on input prompts. Some tasks include natural language processing (NLP), question answering, and image classification. The size and general-purpose nature of FMs make them different from traditional ML models, which typically perform specific tasks, like analyzing text for sentiment, classifying images, and forecasting trends.

So, what do we mean when discuss implementation of GenAI??

There are the main modes when adopting generative AI.

1. Off-the-shelf:?Use an existing foundational model directly by inputting prompts (e.g. interacting with ChatGPT).? For example, ask the model to create a job description for a software engineer or suggest alternative subject lines for marketing emails.

2.Prompt Engineering:? This is currently the most common approach. Use connectors (i.e., APIs) to build applications leverage an existing foundational model.? For example, using the OpenAI API to develop an internal knowledge management solution with a virtual assistant for employee support.? This approach when done right enables better security and intellectual property protection for an enterprise.

3.Retrieval Augmented Generation (RAG): Combining an LLM with external knowledge retrieval. Retrieval Augmentation Generation (RAG) is an architecture that augments the capabilities of a Large Language Model (LLM) like ChatGPT by adding an information retrieval system that provides grounding data. Adding an information retrieval system gives you control over grounding data used by an LLM when it formulates a response. For an enterprise solution, RAG architecture means that you can constrain generative AI to?your enterprise content?sourced from vectorized documents, images, audio, and video.

4. Fine Tuning: Organizations can introduce significant tuning and customization to an existing model to meet their unique needs. ?For example, adding a new data layer focused on customizing the model to better understand the context and nuances of the cyber security data terminology to make the model more relevant to cyber security applications.?

5. Build Your Own:?Building a completely new foundational model is probably not feasible for most organizations due to the sheer data and computational power required to pre-train the model from scratch.

The first three options of GenAI adoption offer the benefit of having the models pre-trained. This allows for use cases that can be done using off-the-shelf tools or using prompt engineering.? This means that for use cases where GenAI can provide value, the time of value can be reduced from many months to a matter of weeks if not days. ?

This is a key point to note in terms of the difference between using traditional machine learning versus generative artificial intelligence; GenAI leapfrogs the effort to implement specific use cases that Machine learning would have been able to deliver but with significant amount of effort in terms of training massive amount of data, the model development and related specialized expensive computational requirements. ?This is the case since with GenAI, the pre-training of the data and the pre-requisite technology infrastructure to build the model is already done.??

This is akin to hiring a CPA to do accounting for your organization compared to hiring a complete novice to do the same job and spending the time and effort to get them up to speed on all accounting basics and foundational knowledge of accounting, then how to apply to your organization. ??

Hopefully this gave you some ideas on the various options to roll out GenAI!

The next article will double click on these deployment approaches and related considerations.. Then another article on AI, Privacy and Cyber Security. Stay tuned!