AI

The AI discourse never ends, right?

I consider myself lucky that I finished up at university before the advent of the LLM. I remember in the weeks after I submitted my honours thesis sending a friend a cool new chatbot where you could get it to roast someone in Shakespearean prose. And now it's entirely integral to my job.

"Self-Taught"

I'm functionally a data engineer - all I do is design cost-optimised pipelines, transformations and access patterns for interval timeseries data. But I have no computer science qualification. My degree was in Renewable Energy Engineering. I suspect a lot would claim that makes me a hack or a vibe coder? In that I could never have gone as far as I have in the time I have without the assistance of LLMs.

When I started at my first office job the metaphor I've always used is being dropped out the front of a city both without a map and without any understanding of what a building is. At that point CLI/IDE coding tools weren't where they are now - 'vibe-coding' still required copy-pasting code into ChatGPT and then copying out its output. But I was operating on a real platform, that had at least one client. So I wanted to be careful.

ChatGPT was instrumental in being able not only to understand what a Lambda function is, or an S3 bucket (these are Googleable) but contextualising exactly how they were being used within the platform I had been set loose on. I still copy-pasted so much code within the codebase itself. I'm sure many a junior engineer has done the same - copying code that works and will probably work for your use case without exactly understanding exactly why or how it works.

I remember how proud I was when I built my very first Lambda workload. Scheduler to queue to consumer. A new DynamoDB table. A combination of copying what worked in the repo alongside adjustments from my best friend ChatGPT.

Art

I hate AI-generated art. Hot take I know. I don't necessarily hate AI-generated multimedia. I think it comes down to the purpose, right? I really dislike when something purports to be creative, or visually enticing and is machine-generated.

As I'm writing the example of cafe menus comes through my head. Because given the choice between a cafe without a flashy generated menu and one with, I'll almost always choose the one without (assuming reviews don't force the choice for me). So maybe what I actually hate is the extreme commonalities that come with just chucking a prompt into an LLM and using exactly what it puts out. It's something flashy without any humanity in it.

I consider myself a rational person but what I'm describing here is kinda dogmatic, right? I'm basically just saying that if I knows it when I sees it then I don't likes it. Perhaps I just care about engagement with the process. Who knows! I think I'd rather see rough edges from someone creating with their own hands.

Empowerment

I use AI all the time. Not just at work, but at home. I like to think that I use it 'correctly', whatever that means. I want it to empower me to create. Because for all their faults, LLMs can make anything feel possible.

Ask one how to do anything and it'll give you links, products, search terms and whatever else you need to feel like doing it isn't really that hard. I absolutely could Google things to find out. But summarising information truly is the exact aligned use case of LLMs. Yes it's causing the slow bleedout of typical internet revenue streams. But I think that's a larger problem than me.

Yes I used Codex to format the paragraph tags on this page. Where's the credibility?