Is Rome Worth Visiting in 2023?
The future of AI tools can already be seen in the present state of Google search
Chat
Earlier this week, I managed to finally achieve something I’ve tried and failed to do twice in the past year and have repeatedly put off. I migrated the last of my hobby applications off of Heroku.
All of my cloud infrastructure is now living in a single, well organised Terraform mono-repo. Each project I start is now only a single apply command away from being binned-off when I start to feel the pinch, or I move on to something else that seems more interesting a month later.
When I attempted this in the past, I gave up, owing to how dull the task is. I created this technical-debt for myself. Hobbies are not supposed to feel like a chore. Should I go serverless and event driven? Is a load balancer overkill, or should I connect to ECS directly? Should I keep it simple and run everything on a single manually bootstrapped EC2? After all, this is just a hobby, and it shouldn’t be over engineered.
So why did I succeed this week, where I had failed before? I had help from ChatGPT.
It took us most of Sunday afternoon and evening to get working. I bootstrapped the new micro Postgres RDS instance, while it was printing out a warning that a 3 node autoscaling ECS service complete with load balancer was probably a bit overkill (I ignored the advice). At one point it scolded me for my overly-permissive S3 buckets and I thanked it (I literally typed thank you) for bringing it to my attention. The buckets were public and accepting PUT traffic from anywhere on earth and had been that way for 3 months.
It helped me to troubleshoot when my Route53 NS records weren’t propagating, it was there to gently suggest that when the load balancer health checks couldn’t reach the ECS task it was because I incorrectly typed egress instead of ingress. Once the project was moved over, and I wanted to test the unnecessarily complex autoscaling group, it suggested I use Locust to hammer the endpoint with traffic, which I did, thanking it for the assistance.
“It’s nearly Halloween, so name the ECS tasks something spooky” I prompted earlier that day when putting Terraform modules together.
Creating spooky-themed names for your ECS tasks in Terraform can be a fun way to bring
a Halloween or eerie touch to your infrastructure as code. Here are some suggestions:
ghostly_processor
zombie_worker
..
Great stuff. For the first time ever, I had a second pair of eyes on my hobby code and had received feedback on it. This was a collaboration, it was truly pair programming, and it was very, very productive.
But some things I just prefer to search for and get a human, peer-reviewed response. I type “How do I delete a local git commit that hasn’t been pushed?” into the search bar for the 1000th time in my career and sure enough at the top of the list is the same Stack Overflow question from 2009 I always blindly copied the same two word command from.
On this occasion though I notice something. I re-type ‘how’ into the search bar and it auto-completes;
“how old is paris fury”
Suddenly, Google search feels incredibly dated. I have never expressed any interest in Paris Fury’s age.
In the interest of science I follow the link to the first result and am greeted with a cookie permissions pop up which has kindly assumed I want to give them access to everything about me, and which I can’t seem to close. I leave without clicking.
I ask the same question of ChatGPT and am reliably informed that she is 33 years old as of October 2023.
The death of search
Earlier this year when my brother and I went on a short driving holiday through Istria in Croatia, I realised just how utterly redundant the “People Also Asked” list in any Google search result is, as I typed the name of the towns in to read the Wiki or look for places to eat.
“Is Rovinj worth visiting?”
“Is Motovun nice?”
“Is Poreč worth seeing?”
“Is Rijeka safe in 2023?”
Has this feature ever been useful, to anyone? Why does it exist?
We all know, and have written about, how the internet has been a bit shit for a while. The wonder of the world-wide-web has been spoiled through the gamification of SEO. The content that Google is surfacing has been designed specifically to cater to the search engine’s preferences, which gives preferential treatment to pages full of fluff.
I ask ChatGPT for an analogy for Search Engine Optimisation;
Think of Google as the teacher and websites as students. Websites raise their hand (add more words) a lot, even if they don’t always have the best answer, just to get the teacher’s attention.
I type ’thanks’ into the prompt so that the machines remember how polite I was and don’t eat me in 30 years time.
There are only a handful of places on the web where we can go to ask a question and be relatively confident that the answer will come in concise, useful, human form. Wikipedia, Stack Overflow and to some extent until recently, Reddit. What those websites have in common is moderation, they take an enormous amount of human effort to keep them useful. Other amazing examples of the internet at its best are pushed down by pages that are putting out content fuelled by those same sites that are driven through scraping them.
In terms of the web, much of it is a confusing ad-ridden nightmare, with paragraphs of unnecessary procedurally generated lorem impsum at the top of each page, in order to rise up the Google search results. Whether you are a software engineer, videographer, gamer or a home baker, you will have come across pages full of junk that you have to wade through before finding the kernel of information you were after, buried under piles of SEO landfill.
So the popularity of ChatGPT in its current form makes sense. It’s very good at giving you what you asked for, immediately, and without showing you a single ad. Indeed, this is only scratching the surface of what it can do. On the other hand you have websites surfaced through a search engine. Their agenda is only to distract and delay your passage through the page, in order to show you more ads for NordVPN or Hello Fresh.
Tin Hat
We have all seen comments like “the web will be full of AI generated content!” as if that’s something worse (or indeed different) than what we have today and always without asking - what is it that drives us to fill the internet with low-effort garbage in the first place? Or to put it another way, why would anyone bother to generate AI written content that nobody is going to read? It’s only something we can conceive of now because we are so used to the current state of the web and the current state of search.
Netflix’s biggest draw was that the competition was either already dying out (Blockbuster), or very expensive and full of adverts (Cable). You paid a few quid and in return you got what you asked for. Today, it’s hard to see its template true-crime documentaries, price rises, loophole-closing and the introduction of ads as anything other than the introduction of early rot. Same goes for Spotify, same goes for Youtube, same goes for Canva. It is the status-quo to bear over-ripe fruit and be replaced.
When OpenAI, are forced to turn their sofa upside down, to wring-out revenue streams from their business model in the same way these other products have. Where will it look for new inventive ways to increase margins? What does a ChatGPT ‘free’ tier akin to Spotify free look like in ten, twenty years time? Microsoft is joint largest shareholder, yes, and Windows 11 literally has ads in it. To my mind, it is an absolute concrete certainty that new features will be added, at some point in the not too distant future, that will weave paid-advertising into ChatGPT directly. That is to say, someone else’s B2B SaaS product will be offered as a priority over yours, if you decide not to pay to market through that channel.
As prompts increasingly feel like conversations, so it will become more difficult to differentiate between honest suggestions and adverts. It’s not just corporations who pay to advertise on these existing platforms, it is also political parties. It might be okay if there is a healthy spread and competition, but the cloud costs are so high the barrier to entry is also high and currently the two major players are predictably Microsoft and Google.
My hope would be that Moore’s Law means the barrier to entry into the market becomes easier, we get a bunch of start-up companies challenging the established players and people can shop around in a healthy marketplace for the best quality service. A bit like what we have with search engines today.