A lot of people in the internet are playing around with Chat GPT at the moment (myself included!). One of the more fun examples, is to ask it to create a python script to automate some task

Chat GPT generating a python script to get financial data

You can even have it write pretty decent shell code

Same script as above but in Bash

This is really cool stuff, but what if I ask it to write terraform? Here, I ask Chat GPT to generate terraform to create an EC2 instance in AWS

Request chat gpt to create an EC2 instance in terraform

This is fine on first glance, but let’s take a closer look. Where does ami-0b69ea66ff7391e80 come from?

A quick look in the AWS console clearly shows this isn’t a public image - so what did it scrape to get this? Was this something Chat GPT generated? Is this a private AMI on some github gist in its dataset?

AWS AMI console

We’re lucky that this is a very specific string and can be easily googled with not too many hits. If you scroll a bit further down on the Google results page, one comes across a gitter.im link for aws-cli
Parsing through this chat eventually leads you to . . .

Comment that has the ami tag

Aha, so this was an old AMI that used to be referenced in the AWS documentation, but has since been removed. That’s how Chat GPT knows about it.
If you read the documentation, the data sets stop at around 2021, which is probably when that image was removed from the public catalogue

The power behind Chat GPT (and AI, in general) for writing code is clear. It can build fantastic scaffolding and help you get “Good Enough” very quickly, but it seems to suffer from the same problems as googling answers does, which is parsing through old information. Still, this is something that could be rectified by updating the data sets

All told, this is an incredibly powerful tool and I can’t wait to see the horrible things that man creates with it the fascinating ways AI is used and how the technology grows