Unless you lived in a cave for the last couple years, it is impossible you missed the AI trend that is happening right now. And as it has been so well explained by Claire Vo here, it is a transformative force that we have to embrace if we don’t want to be left behind. But the changes happen so fast right now that we don’t necessarily have time to integrate everything—as a proof, we keep hearing that today’s state of art will be completely outdated in 6 months. This is why I will start a new series of articles dedicated to clarify what is happening in the AI world, and how we can use all of that in our PM daily life

Misnaming things adds to the world’s misfortune

As Albert Camus said, we first need to make sure we’ll talk the same language as we’ll move on with the more complex topics.

What is AI?

Artificial Intelligence (AI) refers to the simulation of human intelligence in machines. It enables computers to perform tasks like decision-making, speech recognition, visual perception, and language translation—tasks that typically require human intelligence. AI is the generic term that encompasses a lot of subsets, each with unique methodologies and applications.

Subsets of AI

Here’s a schematic breakdown of main subsets of AI:

• Machine Learning (ML): Allows systems to learn from data without explicit programming. ML models improve over time as they process more data.
• Deep Learning (DL): A specialized branch of ML that uses 🧠 neural networks with multiple layers (hence “deep”) to analyze complex patterns in large datasets. It powers technologies like image recognition and advanced natural language processing (NLP)
• Natural Language Processing (NLP): Focuses on enabling machines to understand, interpret, and respond to human language. Tasks include translation, text summarization, and sentiment analysis
• Large Language Models (LLMs): Is a subset of NLP and DL that uses massive datasets to generate coherent and contextually relevant text. LLMs like GPT-4 are foundational for many applications today

Interestingly, ML can be broadly categorized into two main approaches: supervised and unsupervised learning. In supervised learning, models are trained on labeled data where the desired output is known, like teaching a system to recognize cats by showing it many pictures labeled 🐈"cat" or ❌🐈"not cat". Unsupervised learning, on the other hand, works with unlabeled data to find hidden patterns and structures without specific guidance.

Speaking of cats, there's an interesting historical connection between cats and computer vision. In 2012, Google's groundbreaking deep learning system achieved a significant milestone by successfully identifying cats in YouTube videos without being explicitly programmed to do so. This breakthrough wasn't random - the system was trained on millions of images from the internet, where cat pictures were notably abundant due to their overwhelming popularity in the early 2000s. This "accidental" focus on cats helped demonstrate the power of deep learning in image recognition tasks.

What about GenAI?

GenAI, for Generative AI, refers to a type of AI that creates new content—text, images, music—based on training data. Examples include ChatGPT for text generation or DALL-E for image creation, but it can also include code generation or video generation.

Focus on LLM

Probably one of the easiest entry point for AI currently is through the LLM. You’ve probably already tried to use ChatGPT, but do you know how it really works? Nothing is magic in this world, so let’s clarify how it is working.

How do LLMs work?

At their core, LLMs predict the next word in a sequence based on prior input: