🎹 Music for this post: https://www.youtube.com/watch?v=lpqyxOL96QI.
Technology leaders have had a banner year explaining generative AI to their companies’ leaders. Stephen Wolfram gave the world a wonderful (and, admittedly, not romantic) Valentine’s Day gift this year with his lucid essay, “What Is ChatGPT Doing … and Why Does It Work?”, whose only downside was its length. Visit the link and check out the size of your scrollbar to see what I mean. It’s turned into a bestselling book, to boot.
This is simply the best explainer of ChatGPT written to date. I’ve loaded this essay into my browser and displayed parts of it on large screens in the past 10 months more times that I can count. What it helped me understand is that ChatGPT is nothing more than an ingenious application of statistics, and if you can help others absorb this, it opens minds to what it’s actually doing…its limitations…and some good reasons why we shouldn’t be freaking out about it.
I’ve found that the following eight simple portions of Dr. Wolfram’s essay distill the essence of what he’s teaching us:
1) Start by looking at a small sample of text and count the number of times the letters occur:
2) Look what happens if we do the same with a larger sample of text:
3) Start using these probabilities to generate strings of letters, and throw in some spaces:
4) Compare the probabilities for letters to occur on their own…
5) …with the probabilities of them occurring in combination:
6) Then see what happens if we understand the probabilities of them occurring in longer sequences (2/3/4/5 letters at a time)…Wow! Just with this sort of application of statistics, we start getting words!
7) What happens if we do the same with combinations of words, rather than just letters? ChatGPT!
8) Best of all…what this shows us is how utterly formulaic and predictable most of our writing is!
That last part is truly important, and I don’t think enough of this year’s discourse has amplified that point. This is the principal reason that I asserted back in April that ChatGPT Challenges Us to Focus on Better Things. Are We Up for It?
I hope that this TL;DR version of Stephen’s generous essay can help you explain how ChatGPT works to others. Do yourself a favor, though, and give it a full read if you can. It’s well-written and worth your while.