Welcome to AI for Everybody!

I created this site for a few reasons. I’ve been giving a sort of “AI 101” course called “AI for Everybody” and as an adjunct to that I figured it would be useful to have a central place to share materials with students as well as any new discoveries I’ve made in the AI world that might be useful for them (you).

My History with AI

Just to share a little about my own history and interest in AI, I’ve been something of a software developer for years, on and off. I would describe myself as more of a software (and everything else) bricoleur¹ - but in any case, I’ve been involved with and thinking about computers and programming to some extent for most of my life. I’ve also been a fan of science fiction in various media as long as I can remember. So it’s extremely strange to find myself in an era where it really feels like we’re living in science fiction come to life with the recent developments in Artificial Intelligence. AI is something I’ve been thinking about for a lot of years now and not something I really expected to see implemented to the extent it is now in my lifetime! So it’s pretty exciting to say the least.

My Experiments and Work with LLMs

Since the advent of ChatGPT I’ve been playing and working with LLMs. Here are some of the platforms I’ve used and my notes on them:

Github Copilot

I use Github Copilot every day to help write code. I use it mostly for python and javascript, both of which it is really good at, having been trained on the entirety of Github’s open source projects. It is largely a really good autocomplete, anticipating how I am most likely about to finish a line myself and saving me keystrokes, but sometimes where something feels pretty obvious I’ll just add a comment (“#we’ll also need a function to generate these” or whatever) and let it write whole functions.

ChatGPT

I spent a lot of time from first hearing of ChatGPT playing with it. Mostly interested in trying to push the envelope getting it to write interesting stuff - something I discovered LLMs are really not very good at! But I did a lot of experimentation trying to get it to write in experimental styles and trying to push it to make more unexpected word combinations. I also discovered at an early stage the power of instructing it how to behave. I had a lot of fun making up weird rules for it to interact with non-standard vocabulary and experiments like that (“From now on, always replace the word “the” with the word “banana”).

Claude

At some point I heard an interview with Ethan Mollick who recommended Anthropic’s Claude. I don’t think there’s a huge difference between the output quality of ChatGPT and Claude but I do like the Claude UI and workspace. The Artifacts sidebar is a handy place where Claude can save code snippets, reference documents etc., as well as where you can review what you’ve uploaded. For the past few months I’ve been using Claude for a few things including generating code (which is also a good way to learn new tricks), brainstorming, helping me think through pretty much anything I’m thinking through. I find it insanely useful to use Claude to generate lists or otherwise put together structured documents outlining whatever I’m working on or thinking about.

Frontend Development with Claude API

I’ve been playing with developing really minimal applications that lean heavily on Claude’s language processing for functionality. It’s pretty fascinating though tricky. I think we’re going to see LLMs more deeply integrated with, well, everything, over the next couple years, and particularly think the future of software is going to include more and more AI. It seems intuitively likely that we will see LLMs writing, testing and executing their own code to increase their own abilities and reliability. I don’t know how much of this is going on now with the current widely-used LLMs but I intend to find this out soon! I’ll be posting more news of results of my experiments here in the near future.

The ‘AI For Everybody’ Course

I’ve developed a roughly half-day course in AI really targeting beginners, from non-technical people who might know nothing about AI or haven’t ever used it for anything to more experienced people not necessarily very technical people who might not know too much about AI. The course covers a brief history of technology, a brief history of AI, how LLMs work (roughly), some basic approaches, and some superpower techniques. Get in touch (email in sidebar) if you want to know more.