AI or NAI: Data, AI, and LLMs
AI or NAI
Data, Artificial Intelligence, and Large Language Models
jesi
INTRODUCTION
Machine learning, artificial intelligence (AI), and in general developing and using mathematical models of real life matters is where the current economies and businesses? are. In my view, AI has been a subject of intrigue for us human beings, ever since humanity itself has begun to exist and generation after generation, we have been slowly edging toward the elusive and mysterious natural intelligence (NI) through continuous learning and understanding of nature and matters that impact our daily lives. In many ways, prediction comes naturally to all of us, though sometimes it's a gamble, and sometimes it's a serious life matter, and yet we do it naturally daily. Think of complicated life decisions we make based on our “gut-feelings” alone or taking an umbrella to work based on what we heard in the news media, and so on; in all these, we rely on some shape or form of machine learning and AI concepts.
AI or NAI and Large Language Models
Some cohorts and I, while working on a project at CSUEB noticed that intelligent decision is an outcome of experience, which in turn is based on accurate, relevant, and “fresh” data. We have noticed a strong correlation, though unsurprising, between results and input data features. We used Large Language Models (LLMs), Google’s Bard and OpenAI’s Chat GPT-4 in particular, extensively and found them to be very useful in dealing with topics that are truly non-social or cultural in nature. Based on the past 5 months of research and practice around LLMs and predictive AI allow me to posit that
BASICS OF PREDICTIVE AI AND LLMs
In our quest to understand and find the outcomes of prediction, My cohorts and I selected sports betting parameters as the outcome (dependent) variables and all sports data as the independent variables.?
领英推荐
The general approach involves, gathering, cleaning, preprocessing data, treating the right input data features through analytical methods such as neural networks (Scikit, Keras, Tensorflow, etc.), and finally predicting using the latest input data. Our model worked well under linear conditions, however the accuracy isn’t 100%, which is the risk involved with the AI models.?
LLMs on the other hand, employ generative prediction, which is to create usable results based on the given input parameters. When we asked for sports concepts and code snippets from Google’s Bard and OpenAI’s Chat GPT, they both were able to provide usable answers, but yet needed additional work on our side to fit the results of LLMs into our project. However, this is exactly the benefit of LLMs; id est, they acted as enablers and thus saved time and increased efficiency.?
GOOD DATA “may” PRODUCE AI, else RESULT IS NAI
Input data features become the critical component of a predictive model, though it is not the case for generative models. Predictive models are definitive and generative models are artistic, in a way. We used the generative power of the AI through LLMs to create a predictive model on sports betting, with relative ease. Interestingly, 75% of time is spent on data related tasks such as gathering, parsing, cleaning, grouping, and filtering data to create various input features for creating a model and predicting.
One could use predictive AI in personal, enterprise (workplaces), and commercial (Google, OpenAI, etc) situations, but the quality of outcome depends on the quality of feedback loop involving inputs, predictive model, predicted outcome, and real life outcome. If the calibrated data between the predicted outcome and reality isn’t fed-back to freshen up the input datasource (shown in the picture above), the prediction becomes noisy, and thus becoming a Noisy AI (NAI). It’s worth noting that this noise based on irrelevant and non continuous data is an added noise over the inherent inaccuracy related to predictive models.?
CONCLUSION
Certain outcomes are like science experiments and mathematical models, and are? monotonous and boring, unlike real life. Until a certain result has occurred, there exists an uncertain amount of risk in everything we deal with.? The uncontrollable nature of variables in and around us make life wonderful and unpredictable, and that is the beauty of life. If generative AI models continue to focus on the creative part of life that enables us to enjoy life by being an enabler of human power, without intertwining with the divisive aspects of our lives, LLMs and AI will continue to be a huge part of our daily lives.
REFERENCES