24 - Keeping up with GPT, or Not
GPT-4, Claude, and healthy reactions to hype
Welcome to another edition! I'm Daniel Imfeld, and here I share things I've read recently, updates on what I've been working on, and occasionally nascent blog post drafts.
Book Progress
My upcoming "Spatial Data for Web Developers" book is up to 9,000 words now and still growing. I spent a lot of the past week working on PostGIS examples and cleaning up some earlier chapters.
These types of chapters can be a bit tricky to write since you need to present the functions and capabilities in an order that makes sense and is easy to go back and reference again, but also don’t want it to be just a dry list of functions that don’t really come together cohesively.
I’ve come up with a few examples so far about queries for food delivery, workout trackers, and so on, which I’m hoping will make the content more readable.
The first draft of the PostGIS chapter should be done sometime next week, and then I'll move on to the browser side of things, starting with a chapter on manipulating spatial data in JavaScript.
GPT-4
Of course, the biggest news last week was the release of the GPT-4 large language model. The worst aspect of all this was a proliferation of Twitter threads that were all saying the same thing over and over -- not that it was all bad, but clearly a bunch of people are just pumping things out in order to be the first to publish their threads, without really doing any original work. But I suppose that's nothing new.
On the other hand, we're starting to see some calls for balanced responses to the hype. AI is a fast-moving target, and while it's exciting, there are plenty of other ways to run successful businesses, that may or may not integrate LLMs, but offer real value beyond just being a wrapper around a few API calls.
While I might not express it so forwardly, I tend toward this view as well. When riding this wave, you end up fighting on two fronts:
The power law of success that all markets present, in which a small percentage of the people make most of the money.
Keeping up with the fast-moving targets of LLM and AI technology, which every six months seems to make obsolete swaths of businesses that didn't offer much value beyond the API they were built upon.
No hate for the people in this position though. This is a truly exciting time with a lot of real change, with plenty of eager customers to go around. And it presents much more obvious value than the majority of crypto projects.
Perhaps if I was still in my 20s and had no family yet, I would be champing at the bit to build businesses that constantly have to keep up with the state of the art to stay competitive, and I would love it.
But at almost 40 with two kids, for now I'm happily enmeshed in the healthcare market, where you can be 5 years behind the latest tech and people are still wowed by your product because their processes are 15 years behind. There are still a lot of hard problems to figure out, and value to be delivered in more “boring” ways.
Where do you find yourself on the spectrum here? Are you riding the wave? Feeling FOMO? Aloof and watching from afar? Let’s discuss in the comments!
Poe
With all that said, I have started a trial subscription to Poe, Quora's LLM application. They offer access to both GPT-4 and a similar model from Anthropic, named Claude+. Claude seems especially good at giving long, detailed answers to one-sentence questions, without needing much additional prompting to get there.
The answers include so much detail that it can be great just to shore up knowledge, even on topics where you have some idea already. For example, this shared Poe link goes to a question where I asked about the disadvantages of the Mercator projection. Claude+ not only exhaustively covered this, but gave some details on alternate projections as well and the tradeoffs they make.
I'm impressed, and really like this for the cases where you already know some things about a topic but want to shore up your knowledge and see the finer details that you may not know. In my initial testing I found that Claude+ is less willing to actually write code than GPT4, but I haven't tried to push it much yet.
Links and Reading
Louie Bacaj's latest newsletter is also on the topic of incorporating more personal thoughts and stories to your writing. This is even more important now that an LLM can generate 99% of the articles out there, but actual opinions and your personal spin on things are still valuable and will stand out even more going forward. As a few people have said: it’s time to value “writing” over just “content.”
Thanks for reading! If you enjoyed this, I'd love if you share it with a friend or just reply with your thoughts.