Research on AI model types. - Cian

For our business project, we are required to integrate some form of AI model into our news reporting application. 

 

An AI model is a tool or set of algorithms which makes decisions based on a set of data inputs.

Most AI models attempt to identify patterns in the data to make predictions or come to conclusions. These are especially useful when dealing with very large amounts of data as they can process and identify information of interest very quickly.

 

There are many different forms of AI models created for different purposes:

 

Machine Learning algorithms are a part of AI which allows the machine to learn and program its own algorithm without needing to be programmed by a human.

ML algorithms can be created using “Supervised” and “Unsupervised” learning.

Supervised learning involves training the model on a set of data handpicked and labelled by the data scientists. This allows the AI to zero in on what it needs to learn and it’s a model often used in computer vision and image recognition.

Unsupervised learning is the opposite. The algorithm is fed vast amounts of unlabelled data and is required to infer what predictions to make and what patterns can be found. This method is typically used when hidden factors could exist within the data which would be undetectable by humans.

 

Deep Learning is an extension of ML which is more suitable for detecting patterns within data. DL models are created in a way to mimic the human brain, utilising layers of digital neurons in between the inputs and outputs. Many of these systems are incomprehensible by humans once they’ve been trained. The deep layers of neurons become a kind of black box which takes in data and outputs the result. These models are best suited for things like Image recognition in images and videos, and understanding human speech.

 

Nearly every AI model requires a vast amount of storage and processing power to be effective. This means that smaller businesses and startups don’t have the resources to run them on their own.

Thankfully there are accessible and open source AI models operating in the cloud, like ChatGPT.

 

ChatGPT is a large language model powered chat bot. It is able to have very convincing conversations with human users and can even help solve a myriad of problems which it is asked, from programming solutions script and essay writing.  It’s not perfect, and occasionally presents made up information as facts, bot overall it has a very high success rate for the problems applied to it.

In my previous post, I looked at NewsNotFound, an AI powered news website. The articles on this website are written by ChatGPT based on information scraped from other articles and information online.

 

ChatGPT has high potential for use in our project given its versatility, however it will have to be paired with other systems to ensure a high level of accuracy and to limit bias.