Why Is This Octopus Drunk?

From time to time, I’ll come across something that makes me really curious. This feeling has been with me for my entire life—I’m curious by nature, and while some parents might have discouraged this nature, I think my folks did a great job of encouraging exploration and fostering lifetime curiosity in me.

Unfortunately, I learned to suppress this instinct as life dragged on. Why? Because it has been very, very difficult to find answers to questions for most of my life. If you wanted to know the answer to something fairly complex or nonintuitive back in the day, you needed to read books.

No problem—I loved reading books! The only problems were figuring out which books might contain the answers you were looking for, and then actually getting your hands on them.

Even further: there were lots and lots of books written by the end of the 20th century, probably at least a hundred million. That’s a lot of books! Tons of academic and literary answers could be found in them. However, suppose you wanted to figure out something really important, like why this octopus looks drunk:

If you’re a Gen Xer living in the late 80s, how would you even go about doing this? You probably don’t go to the microfilm machine and scroll to O for Octopus. Instead, you might ask some people the question. Would anyone in your circle know the answer?

For many of the questions I wanted to know the answers to, this was a resounding “no.” Friends tend to understand a lot of the same sorts of things you know about.

Young Andrew was forced to accept the uncomfortable idea that there were lots of questions that I just couldn’t answer in a timely manner. Sure, I could probably figure that octopus thing out eventually, but I’d have to leave aside another dozen questions I would want to answer.

Of course, the thing that saves the day in the 21st century is the internet.

During my lifetime, I’ve had exactly two moments where I found out about a new technology, and I saw immediately that everything had changed. This first moment was in 1994, when I was visiting Richmond for the first time, and the friend I was staying with had America Online. AOL was a gateway to the brand new world wide web.

I knew that this was fundamentally different than anything that had come before, and that all human minds could be connected in ways never before seen. It took a few years of the web being built out, along with many more regular internet users to create a critical mass with an incredibly useful new ability: to look up anything you want to know, and to find out an answer quickly.

What does "quickly” mean, though? Well, in 1994, it meant you might be able to find a few answers if you were willing to get your hands dirty, ask lots of different folks questions, and search the horrible sites available at the time. By 2000, with cable internet and Google starting to pull away from the pack (remember Yahoo!, Lycos, Infoseek, Excite, or Webcrawler?), searching for answers became a great deal more practical, and it became much faster.

Now, you could find a relevant page that would have the answer you were looking for. This just blew my mind—there seemed to be little you couldn’t figure out, provided you were good at searching and had some time. I became good at quickly searching and screening answers, then scrolling to skim and find the answer I wanted. Instead of a day to find a tough answer, it might take 20 or 30 minutes now.

This got better over the next 20 years or so, but for the most part, the gains were slow and steady. Eventually, Google and other search engines started trying to give answers to your questions instead of a bunch of search results. Alexa, Siri, and the voice I like to call Googlina became go-tos for anyone who didn’t mind speaking out loud to ask a quick factual question.

Then, that second moment happened for me, another one of those times when everything changed. This second moment was in November of 2022.

You may not be shocked to find out that this was my first exposure to ChatGPT, the first really good Large Language Model. This sort of generative AI operates in principle the way your phone’s autocomplete operates—by predicting the next word you’d expect (or want) to see in an answer.

I knew this was happening, but the feeling I got was exactly like that 1994 feeling. Demystifying how the magic in front of my eyes worked did very little to diminish the sense of wonder I had.

Now, I could get to the bottom of complex things in seconds or minutes. What took me ten hours of research prior to the end of 2022, now took perhaps a quarter of that time, and even less when GPT-4 debuted in March of 2023. The pieces that I write every day are things I know about in some detail already, but there are loads upon loads of facts that need to be verified, and I want to make sure the technical details are right when I share an idea with you.

Sure, you have to verify the information and fact-check it yourself! LLMs can hallucinate (make up answers), but to me, this isn’t any different than needing to verify information I found anywhere on the web, ever, with primary sources (when possible) and with vigorous cross-checking. The only difference is that now I can both find and verify the answers ten times faster.

Some folks will insist that generative AI is an inhuman abomination, but to me, it’s the most human experience possible. I’m tapping into an unimaginable number of different (very human) minds—every idea that has ever been turned into ones and zeros. I feel this connection to humanity every day.

Let’s turn back to our inebriated (but useful!) friend:

The internet rather disappointingly tells me that this is the ultimate example of form following function—the “arms” are of slightly different height so that you can easily hang two different items and grab them quickly, and there’s not much more to the story.

Are there other things you thought you might never be able to find answers to? Are you able to find them any better now? What limits are still there?