This blog is a doorway to understanding how these advanced models are reshaping our future, piece by intricate piece. Join me in uncovering the secrets of these extraordinary AI architectures, and discover how they’re forging a path to a world where the possibilities are as vast as the universe itself. Get ready to elevate your understanding of AI — a journey awaits that’s as thrilling as it is enlightening.

Hey guys it’s Adrian here. If you appreciate my content consider hitting the like button or sharing this article. It’s the only way the algorithm really notices me.

Supervised learning

What if I told you that Gen AI means nothing when compared to supervised learning?

Generative AI like ChatGPT is just a flash in the pan. Recently, Andrew Ng recently stated in a talk that the future of AI is going to be built on the back of supervised learning.

In his talk he referred to supervised learning as “labeling things”. In other words teams of engineers and support testers provide both the input and the output to the AI. Their goal is to give the model an understanding of the patterns in relation to predictions.

While to benefits of this architecture are easy to see there are downsides. Supervised learning requires massive amounts of data. All of which needs to be labeled by humans or algorithms they code to label.

Unsupervised learning

This small change in architecture changed everything for me and my teams. The transition to unsupervised learning.

While supervised learning provides great predictions and categorization it’s clear that it has drawbacks. Time and effort.

One way to improve the process of training the model is to adopt unsupervised learning. The only difference?

The data is unlabeled.

This means that the machine learning algorithm operates in such a way that it recognizes patterns on its own. Without needing to be given examples.

When you hear stories about video games beating human players after 2 million training rounds it means and unsupervised algo played enough to find a way to win.

Reinforcement learning

Think about it as the carrot and the stick but even more intense.

Reinforcement is learning from all the other models because it can optimize its own algorithms and improve its own pattern finding.

The architectures we have reviewed before allow solely for the recognition of patterns. Now we are moving into the world of self-learning.

When an AI can find its own ways to improve it can find things that its creators had not thought above. Setting the foundation for self-learning.

Deep learning

When you hear about neural nets and AI taking over the world, deep learning is what they are referring to.

Deep learning is an evolution, and really the pinnacle, of modern of artificial intelligence. Building on top of the architectures we have cover up to this point.

Models of this architecture construct complex graphs of interconnceted nodes to understand the relation between millions of points of data. Helping it make advanced predictions that outperform human minds.

Humans can only process a few dozen points of data at one time. An example is planning your day or remembering a movie plot.

Deep learning models can remember millions of points of data to draw conclusions across a wide range of topics and disciplines.