Every month I teach a Tech Term You Should Know (TTYSK) and a tech essay to level up your technical literacy and communicate well with dev teams. Ask me anything and I'll cover it in an upcoming issue.

This issue's TTYSK is "LLM (Large Language Model)". Scroll to end to learn more 👇

AI vs. ML vs. Deep Learning

Flashback to March 2016 when AlphaGo, an AI developed by DeepMind, defeated the world champion Go player, Lee Sedol. This was a pretty significant milestone considering Go is a highly complex board game that’s far more intricate than chess with more possible moves than there are atoms in the universe. Wow.

AlphaGo's win was a big moment for AI that received considerable media attention. Buzzwords like AI, machine learning, and deep learning were all used to describe how AlphaGo achieved its feat. Most of us didn’t understand those words then, and let’s be honest–most of us don’t now.

Since AlphaGo’s win and most notably, since ChatGPT’s launch in November 2022, AI has now become part of many of our daily lives. In fact, AI will likely play an even larger part in our future. This makes it even more important to get our buzzwords right.

To start, we need to differentiate between AI vs. Machine Learning vs. Deep Learning. You can visualize the three as subsets of each other: Deep Learning is a subset of Machine Learning that’s a subset of AI.

A good analogy is to imagine AI as transportation encompassing all methods of moving people (e.g. cars, trains, bicycles). Machine learning would be like cars, and deep learning would be electric cars.

Artificial Intelligence as the broader concept

AI, short for artificial intelligence, is an umbrella terminology for all technologies that has the ability to simulate or mimic cognitive tasks that typically require human intelligence such as visual perception, speech recognition, decision-making, and language translation.

For example, language translation traditionally requires skilled human translators capable of understanding the context and nuances of the source language, but can now be performed efficiently by AI like Google Translate.

The concept of AI has actually been around for decades with early theories and experiments dating back to the mid-20th century. All computers since the beginning of computing are artificially intelligent on some level capable of doing simple computations, like mathematics, that previously could only be done by humans. However, the speed, abilities, and performance capacities of computers have made a quantum leap in the last few decades capable of advanced cognitive tasks with the potential to transform various aspects of society today.

TL;DR: Machine Learning

ML, short for machine learning, is a subset of artificial intelligence that develops programs capable of identifying data and patterns through learning from a large dataset of structured or labelled data; the most popular use case being image recognition.

For example, ML models can be trained to identify whether a dog is depicted in an image. These models learn through being fed millions of images labeled as containing dogs. Eventually, the models begin to recognize patterns and identify whether an image contains a dogs.

Machine Learning (ML) vs. Deep Learning (DL)

ML and DL both focus on developing programs capable of identifying data and patterns but they differ in how they process data and how they extract and identify features.

In traditional machine learning, a human person manually identifies features and labels data before feeding it to the ML model. From there, the ML models rely on straight-forward algorithms such as decision trees to learn to identify patterns. Deep learning models are different in that they’re designed to mimic the way the human brain processes information using artificial neural networks known as “deep neural networks” that doesn’t require a human for manual feature extraction.

Let’s take our prior example of identifying whether a dog is present in an image. Humans training an ML model would feed it with images labelled ‘has dog’ and ‘doesn’t have dog’ and give it identifying car features like color, edges, shapes, etc in a process known as “feature extraction”. In a deep learning model, a large dataset of labeled images is still required, but the process of feature extraction is handled automatically by deep neural networks.

Real-World Applications

1. Machine Learning (ML) applications

• Email Spam Filtering (e.g. Automatically detecting and filtering spam emails using algorithms trained on labeled data)

• Customer Service Chatbots (e.g. Providing automated responses to customer inquiries based on learned patterns)

• Speech Recognition (e.g. Transcribing spoken language into text in apps like Google Voice)

• Credit Scoring (e.g. Assessing the creditworthiness of loan applicants through predictive modeling)

2. Deep Learning (DL) applications

• Image Recognition (e.g. Identifying objects, faces, and scenes in photos and video)

• Natural Language Processing (NLP) (e.g. Google Translate)

• Autonomous Driving (e.g. Tesla, Cruise)

• Voice Assistants (e.g. Alexa, Siri)

3. Artificial Intelligence (AI) applications

Encompasses all ML and DL applications above and anything else that performs tasks previously requiring human intelligence, including:

• Rule-Based Chatbots: Simple AI chatbots that operate based on predefined rules and decision trees, rather than learning from data.

• Automated Decision-Making Systems: AI systems used in legal or administrative settings to apply rules and regulations without learning from data (e.g., tax preparation software).

• Game AI: Non-learning AI used in video games to control non-player characters (NPCs) using scripted behaviors and predefined rules.

Get smarter on AI in 5 minutes a day.

• The world’s largest AI newsletter, read by over 600,000 AI professionals.

• One free email every morning on what’s new in AI and gives you “the rundown” of the most important developments.

• Allowing for readers to keep up with the insane pace of AI and why it actually matters

Subscribe with one click.

💡 Tech Term You Should Know (TTYSK)

LLM (Large Language Model)

LLM stands for “Large Language Model”. It’s a specialized type of artificial language program trained on huge amounts of data using a form of machine learning called a “neural network” with the primary aim of understanding, summarizing, generating, and predicting human language content. To other words, an LLM is a computer program that has been fed vast amounts of examples to be able to recognize and interpret human language or other types of complex data.

There are lots of different LLMs created for different purposes and use cases like:

• Helping programmers write code

• Customer service and Chatbots

• Generating human texts such as essays or poems

• .. and much more

If you’ve ever used an LLM, you’ve likely been impressed by its incredible ability to understand unpredictable queries (aka “input prompts”) and generate human-like responses similar to what you’d receive from an actual person. But like all algorithm generated content, LLMs' reliability depends on the quality of the data they are trained on. If they are provided with incorrect information, they will produce inaccurate responses to user queries. Currently LLMs are trained on massive amounts of data from the internet (thousands or millions of gigabytes' worth of text!), and seeing as how much inaccurate or misleading information exists on the internet, LLMs have a ways to go in terms of security and accuracy.

Missed the mid-month PM & Tech Jobs Newsletter?

Looking for a new job? Our PM & Tech Jobs newsletter is issued monthly with product role job listings from senior to entry-level roles.