A Newbies Introduction to AI & ML

My entry level journey into AI concepts & open source ML models

Nov 30, 2023

Hello humans!

Welcome back to issue #5 of Waves and Code Weekly 🏄.

This past year the world has been ablaze with this latest buzzword, AI, and it has evoked some feelings of uncertainty and fear in much of the world’s population. I’ve always found that the best way to combat fear is with knowledge and so begins my journey into the world of Large Language Models (LLM).

This past week I attended an AI & machine learning(ML) event. One of the key speakers was Julien Simon - “Chief Evangelist” at Hugging Face (a platform that provides a library of pre-trained ML models). The event was my first real exposure to AI & ML so I thought I’d share my experience and learnings.

What exactly is AI, ML, LLM, and Hugging Face? 🤖

Take whatever I'm about to say at face value because I'm no expert in AI or ML. That being said, I have some nuggets of information and learnings that I’d love to share.

To start, let’s familiarise ourselves with some terminology:

Artificial Intelligence (AI): Refers to machines programmed to simulate human intelligence. It uses a broad range of techniques, however recent use of the term AI has been used to describe Generative AI. Generative AI is a subcategory of AI that focuses on creating new content/images/text etc based on patterns learned from existing data.
Machine Learning (ML): Training machines to learn from sets of data without explicit programming. ML algorithms learn patterns and then make decisions based on data they are trained on.
Large Language Models (LLM): These are advanced ML models designed to understand, generate, and process human language in a way that a human can understand.
Hugging Face: A platform providing a library of open-source pre-trained models for various tasks. Julien Simon described Hugging Face as “The Github of Machine Learning”.

Hugging Face: A place where pre-trained models “hang out”

One of the biggest takeaways for me was learning how accessible Hugging Face makes various pre-trained models. With this, the world of generative AI becomes more than just “ChatGPT”. I realized that you have access to powerful models that you can pull to your computer and use out of the box.

Some popular models I was exposed to are:

Bloom: A text generation LLM able to output text in 46 languages and 13 programming languages.
StarCoder: A text generation model trained on 80+ programming languages.
LLaVA: Visual and language understanding model.
Idefics: A visual language model.

The introduction to various models can be somewhat overwhelming, especially if your only exposure to AI has been the ChatGPT platform. Luckily Hugging Face provides some free learning material to help you get started working with open-source models and learn various ML concepts.

I suggest using these guides to start playing around with ML models on your computer to learn more about the fast-developing field of AI.

Summary of Additional Concepts/Terms I Learnt:

Reinforcement Learning: An ML technique that mimics the trial-and-error learning process, used by humans, to achieve a desired goal. Actions that work towards your goal are reinforced whilst actions that detract from your goal are ignored. This is a technique used to keep models “relevant”.
Fine-tuning: Models are trained on massive data sets to learn general language skills and knowledge. Fine-tuning adjusts a model’s internal weights to bias its output toward new data without affecting everything it has already learned. This allows responses to be more specialized.
Retrieval-Augmented Generation: This is a framework for improving LLM-generated responses by providing a content store/external source of reliable updated information that the LLM can query.
AI Hallucination: A phenomenon in which an LLM produces outputs that are nonsensical or totally incorrect due to it perceiving patterns that do not exist for humans.

I’m looking forward to learning and experimenting with LLM’s. I recommend checking out the following free resources if you would like to empower yourself to be AI-ready for the tech that is rapidly changing our world:

AWS - Generative AI Learning Plan for Decision Makers - Business-focused course that is not highly technical
Generative AI Learning Plan for Developers - Software Developer-focused course
Hugging Face Learn - Learn how to use the HF ecosystem

If you spot a misconception or have questions, please don't hesitate to drop a comment.

Adventures: Thoughts in Powder 🏂

The ski season is drawing closer and as a result, I’ve been getting served more snowboarding content on Instagram. Unfortunately, I won’t be on the slopes this season and I’m feeling a little bummed.

However, being the forever optimist that I am, I thought I’d use the opportunity to reflect, share, and be grateful for the last trip I took in January.

Before my last trip, my visa was delayed and did not arrive until 2 days before my flight. At that point, I had accepted the possibility that I was probably not going to make the trip. However, by some miracle, everything worked out and I spent my birthday on the slopes in St. Moritz, Switzerland. It was a pretty special one and a memory I’ll carry forever. Here are some pictures from that adventure.

Favorites this week ⭐

🌐 Website: thevalleyofcode.com
Essentially a cheat sheet on how to get started with a career in web development.
📰 Blog Post: The architecture of today’s LLM applications
An insightful post explaining LLM’s from a system design perspective.
🎬 Youtube Video: Mark Rober: Octopus vs Underwater Maze
I’ve always enjoyed Mark Rober’s videos. They are a mash-up of entertaining and educational. Also, Octopus are amazing!
🎵 Song: Synth City by Martin Roth
I seriously cannot get enough of this synth-wave track, I’m listening to it as I write this.

Thanks for hanging around for issue #5 of Waves & Code!
I hope this week’s post introduced some new concepts, piqued some interest in generative AI, or at least provided you with some resources to upskill. I’m keen to get my hands dirty playing around with more models. Feel free to share the resources above and let's learn together!

Catch you on the next wave!