Welcome to the 21st edition of the WatchTower! In this edition, we’ll provide a step-by-step guide to help you learn how to build AI!

📰 Featured in This Edition:

• Building ChatGPT: for any skill level

Building ChatGPT: for any skill level

Do you use ChatGPT or TikTok? Yeah, you do. Ever wondered how they actually work under the hood? No matter what level of technical experience you have, this guide is for you.

0 → Overview

This article aims to provide an easy-to-follow learning path for anyone wanting to know more about how AI works - no matter their level of previous experience.

For accessibility, all resources suggested will be free, and/or open source - paid options exist, but sometimes offer a subpar experience.

Sections

This will be divided into 3 parts, based on your rough level of technical experience:

1. Non-Coder

This is for you, if the word VSCode already feels weird.

2. Novice Programmer

This is for you, if VSCode is your bread and butter.

3. Vim User

This is for you, if GitHub (or Vim) is your bread and butter. 

Disclaimer: these are approximate distinctions!

In practice, most people will fall somewhere in between - feel free to mix/match resources between them freely, however you find useful :))

1 → Non-Coder

Motivation:

The motivation for this stage is twofold.

1. First, to learn to think like a hacker - so you can build and break AI.

2. Second - to get the basic lingo down (LLM = large language model = ChatGPT).

Lots of high-quality hacker courses exist; but the best ones I've found are:

* Learn Python - Interactively (Links directly to first exercise!)

* LearnPython.org (Each dot-point link has interactive code!)

* Learn Python - CodeAcademy (7-day free trial)

* Harvard's Intro to Programming (online, free!)

Python is by far the easiest language to learn for a beginner - I used to think C was better, but it's a lot scarier to start with.

The industry standard is Python; and all the concepts you learn will easily translate to other languages in future if you want to switch.

Don't overthink it - just get started :D

2 → Novice Programmer

In-person option: AI Safety Fundamentals (international)

Motivation:

At this point, you're comfortable pulling GitHub repos and trying out random projects.

This is the most fun part; you know how to hack - now you get to experiment with building your own custom AI systems!

I recommend experimenting with frameworks to build a semi-complex LLM workflow for yourself. (I found Dify, and Microsoft's PromptFlow, to both be easy-to-use yet powerful!).

Fun frameworks to start with:

* GenAI for Beginners (step-by-step guides, by Microsoft)

* Dify (like Zapier, but for LLM's - and fully open-source)

* PromptFlow (very powerful, but docs aren't great)

So: PICK A PROJECT - I found I was most motivated making something that directly helped my daily work (e.g, used Dify to automate my most common prompt chains).

Tip: Feel free to even start and stop several projects, if that helps keep you more engaged :)

It doesn't matter what it is - as long as you're building with frameworks, you're learning the skills you'll need to work at industry companies like OpenAI and Anthropic.

3 → Vim User

In-person option: MATS (California, all-expenses-paid)

Motivation:

The motivation for this stage is for you to understand how models work under the hood. By the end of this stage, you'll be able to train your own models in your sleep*.

Honestly: just build bruh 💯

* MiniTorch for under-the-hood understanding (very helpful YouTube walkthroughs if you get stuck!)

* ML Engineer Accelerator resources (feel free to skip some of the math!)

* DeepLearning.ai (Taught by the legend Andrew Ng)

I recommend trying to reproduce a paper (e.g, on building a neural network, or transformers).

This point is where it gets really good to join a program (e.g, ARENA/MATS) - both for learnings, but also for community.

You're well past entry-level here; it's time to help push the field forward - though you may not think it, you're more than capable of doing so at the point!

Programs

AI Safety Fundamentals (AISF)

> For the AI/hacker novice who’s thirsty for more

AISF teaches you ML fundamentals from the ground up, assuming no knowledge (could skip stage 1 if you learn fast 👀 ). You build understanding for the first few weeks, and then plan and create a project (solo, or in a group).

I'm midway through it currently, and love its format - each week we chat about new topics and risks of AI, and debunk hot takes (or, more often - get my hot takes debunked).

AISF is endorsed universally; e.g, by graduates now working at OpenAI, Google DeepMind, Anthropic etc - you can find testimonials on their website. It's offered for free - though, it costs the facilitators BlueDot Impact about $1000 per participant to run.

MATS/SPAR

> For the ML novice who wants to deep-dive into research

I haven't done these programs myself (yet) - but they come universally recommended; you’ll pick a project, and be mentored by an industry expert.

Roughly half of all MATS scholars currently work at AI companies including (you guessed it!) OpenAI + Anthropic. More importantly: they made cool stuff!

Notes/Disclaimers

• This is by no means a comprehensive guide! It's just some resources that I found worked for me at that point of learning.

• If you found this helpful, great! If you think a friend might find it helpful too, feel free to forward (or, screenshot!) it to them.

• If you're keen for more resources, I followed this guide when learning myself (though, I didn't use it when making this).

Good luck! You got this :)

_{Published by Zac, July 15 2024}

Sponsors

Our ambitious projects would not be possible without the support of our GOLD sponsor, UNOVA.

Closing Notes

We welcome any feedback / suggestions for future editions here or email us at [email protected].

Stay curious,