Day 6

Today, it's six days since the release of ChatGPT, a neural network which will change everything.

Yeah, I know how it sounds, but this is it, this is the 6th day of the new age, probably the age of assistants or I don't know how to call it. I wont put more screenshots here, the internet is already full of that, just go ahead and try it yourself.

My initial reaction to ChatGPT was meh, I've asked two dumb questions, got even worse answers, and that was it, bye bye. But that's because I definitely had somehow unreal expectations.

You cannot ask it to find a cure for cancer and if you are an expert in some domain, you will be able to ask questions which it will not be able to answer and what's worse, it will just try to hallucinate something which looks plausible but it's not correct.

So it's not reliable, it's not excelling in any particular knowledge and it's not a AGI either, so why is it a big deal anyway? Well, it turns out that for many things you don't need an expert, and this thing is often good-enough.

It can explain you pretty much anything in simple terms, it can give you good looking LinkedIn phrase, it can suggest catchy names & titles. And most importantly, it seems that you can sort-of pre-program this thing.

You will eventually get better in asking right questions and in the "set-up" phase, or I don't know how to call it. It's when you tell the chatbot what are you two going to to talk about and in which way you want to get answers, etc.

For example:

I want you to challenge my opinions. But I also want you to explain your standpoint.

So you are instructing the chatbot first, and then you get much better results.

Yeah, so you can argue with a chatbot, is that cool? Sure.

Is that useful? Maybe?

Can you trust it? I don't know, probably not yet.

Another example - so I know a little bit about charging stations and it's not really a common thing so I was curious what the bot can do there.

So I've asked the chatbot, what do you know about OCPP? Can you explain it in simple terms? Can you give me example of communication between the station and the central system? Can you give me some example source code? Cool, but I don't like python, can you do that in javascript? Thanks, bye.

Jaw-dropping. Yeah, it's probably not a production code but that's not the point. The point is that I could skip the whole OCPP studying phase and just jump to the real work, saving me many days and trial-errors.

People expect to get runnable code, and that might happen too but it's not necessary, I don't need a bot to do my work, I just need a little help here and there. And for that, it works remarkably well.

And the chatbot is just the tip of the iceberg. I think assistants will be everywhere, you will be driving your car and instead of "call Xxxx" and wait for the call, you will actually say something like:

hey, can you call my mom and tell her that we are a bit late and what is our expected time of arrival?

or maybe even crazier stuff like:

can you call my client to reschedule our appointment to next week and then tell me which day you've agreed on?

and just keep driving, and listening to music.

I always thought the AI revolution will happen first in medicine or in science or somewhere else, but it's probably our industry which will go first.

Just think about it - E2E test automation driven by written acceptance criteria, experimental load tests where you can just ping-pong with chatbot and ask for more traffic, and then ask what failed, and for some instrumentation and to send you core-dumps or I don't know, pretty much anything, make the network slower, make it faulty, simulate split-brain, simulate xxx.

Oh, and what about infrastructure as a code? Well what about infrastructure as a description? Yeah, the actual human-readable description of what the hell you need to be setup for your application.

Or as a tool for security auditing - you could ask real questions - can this server see this other server? What could happen if xxx? Imagine you are trying to penetrate this system, I am trying to secure it, can we go through some common attacks and check that our system is secure?

Soon, it will be way easier to ship things with the help of this thing and yeah, eventually, less people will be needed in the long run. And ultimately, it's going to be a must-have tool for any freelancer and later even for companies.

I mean, I don't want to be sound depressive or anything, I am actually excited, but it's a weird feeling. Everything will be different next year.

Edit: I have asked ChatGPT to rewrite this blog post in a more enjoyable way, and then to make it twice as long because the result was somewhat brief originally. Yeah, and then I've ruined it but it's hilarious.