GENERATIVE AI
GENAIS a revolution in the AI. Built upon OpenAI's latest innovation of ChatGPT 4.o.
Click below to view and obtain your own version of a robust, trainable AI. Make it your own AI.​
Click Here >
With all the hype about AI and Generative AI, let me take a moment to ensure everyone knows transition to an AI capability is not a PANACEA to make your life easier with the click of a button. It will take some work, commonly called ‘elbow grease’ for those of us who have done mechanical work or worked on a farm.
IMPORTANT ASPECT FOR IMPLEMENTING GENERATIVE AI
MANAGEMENT MUST UNDERSTAND
TO TREAT AI
AS A NEW EMPLOYEE
The question from management, if they have not already asked, is, “What is it going to take to get AI operational? Where Do I Start?”. To implement a new AI capability, first start with process review to determine repetitive workload activities.
There are several blogs pointing to a daunting question of where to start. The following article might help in making that decision. “Almost every business has a public-facing website or a customer call center. Unfortunately, many of these websites and agents have become a labyrinth for customers to navigate. Internal call center and customer service content is thus one of the prime places where organizations can begin applying generative AI. “ Stay up to speed on transformative trends in generative AI, Philip Moyer, Global VP, AI & Business Solutions at Google Cloud, April 29, 2023, https://cloud.google.com/blog/transform/prompt-choosing-generative-ai-use-cases/
FOUNDATIONAL MODEL
AI is not a capability to develop, put in a shoebox. Management must expect the need for continuous improvement for long term success, because AI while improving services and will reduce cost, it is a new IT capability/system.
Let us pause for a minute and gain an understanding of some basic principles. To begin, every reader by now has experienced the use of a search engine such as Google and Bing (from Microsoft) just to mention two. These are mainstay search tools and have created a wealth of data from their robots’ accomplishing web crawls on websites and returning the data (big) back to their regional operational servers.
AI in a conversational mode needs this big data to respond to questions asked by users. Google Maps is a classic big data AI model we have all used. Tools on the rise are those such as OpenAI (ChatGPT) or new AI tools.
CUSTOM MODELS - Unique to the Business
When dealing with unique model for a business, the AI responses are provided through an interactive building process for collecting and gathering both questions and responses – labeled as TRAINING AI. It is extremely important to understand this process of training a task AI model. ​
TRAINING THE MODEL
This is an exercise within itself. If the model is not trained properly, then the results will be negative. Just remember AI is software and the release of software requires testing and testing scenarios. Understanding software testing will enable the ability to grasp the needs for training the model. Training the AI model is in the same arena as testing software, with the knowledge the AI intelligent portion can expand. Just like testing software there are certain events that are critical to the training.
-
Each model can stand on its one, thus each model needs a High-Level Objective. (Note there are many scenarios for one model, thus many scenario-based objectives are required for one model to enable answers to the numerous inquiry/questions.)
-
Each model, due to its unique nature, will need a set of independent data, free of errors.
-
The model architecture is equally important. Based on the article in the drop down above make the selection for the intended outcome.
-
Training the model is where the work begins and requires both technical and functional resources to ensure the training and output meet expectations.
-
Never deploy the model or updates without end-to-end testing. Recall the need for automated testing to exercise the entire system? For example, if the input question is asking for Password assistance, expectation, therefore, would be the password strength requirements.
-
Stage the model in a staging area to add to the training until the entire model is trained with every, or as many scenarios as possible. Append these tested and trusted scenarios and build out the model. Once it is tested in the staging areas, signed off by leadership, then it is ready for deployment.
-
Deploy the model into operational use.