> - "Add these domains with these mailboxes to an SMTP server running on my homelab, forward all incoming and outgoing via mxroute"

I laughed reading this. Have you actually tried it? I recently tried to get Claude (Opus) to give me a setup for a simple SMTP relay with Postfix. I had to constantly fight it not to include completely unnecessary components and configurations like Dovecot or just override my requirements. Even Claude is nothing but a semi-competent tutorial clicker.

To someone who never did anything like this themselves the result might seem fantastic. They might not battle Claude on details like I did because "hey, it works, doesn't it".

Don't get me wrong, I agree in general that LLMs are a significant milestone. But I would like to add, especially because of your tone, that perception of how much of a leap it is, is very likely a matter of individual competence. So your enthusiasm might say more about yourself than about the tool.

It’s not going to go back to pre LLM times, it’s also not going to go to a generative ai utopia where we can turn off our brains and a model will understand every nuance of human existence without some from of defined guidance.

Just like block chain still exists, it just didn’t revolutionize decentralization. It created a new tool that can be implemented when you want the benefits of what the blockchain can do, same thing will happen with language models.

Yeah and no... I have been a developer for 20 years and most of my time isn't spent writing code. Most of my time is spent monitoring production, being in meetings, doing design work, doing code/spec reviews, creating tickets for infra/IT to deploy stuff...

So yeah I love AI and it really helps my personal projects. But my actual work? A little, I can work on some nice-to-have PR in the background during my meetings. I might be 20% more effective coding? Mostly because I do more of these low-hanging fruit PR like bug fixes and refactoring. The hard stuff still require thinking. But that's not even half of my time. It's fun, but not world-changing.

> I just don't understand how someone could think we're going back to a pre-LLM world.

That's not the problem, the problem is that for every hour of dev saved society will lose 10 hours on fake data, fake bug reports, automated scams, needlessly convoluted email, etc.

Because having a slave do whatever you say can also enable some pretty remarkable workflows, but that doesn't stop it from being too morally and economically expensive to become ubiquitous AND sustainable. LLM usage certainly doesn't involve sapient beings and so doesn't carry the same moral reprehensibility as slavery, but in real terms, it's probably even more of a financial quandary.

>> Have you not used one since Opus 4.5 came out?

Every time there's a new model This Changes Everything. Every single time.

Every time someone points out a limitation of LLMs it's always the same refrain: "Have you tried the latest? Because they're so much better now!". Except this happens with every single model generation. Every time the new models are so good that it's a seismic shift, a step change, a paradigm shift. It's like the finest curve asymptoting ever upwards, a few billions of training cost at a time.

But- to where? AI keeps getting so groundbreakingly better and we're still stuck in the same old world, except now you can...

... port your unmaintained app from 2017 to modern Android? Why was your app unmaintained? What is the societal and technological upheaval that will come from automating such low-stakes tasks? For such huge costs?

Yes it's a useful tool. What does that have to do with the singularity, recursive ASI, fully automated economies or whatever is being hyped as the radical transformation of society?

The issue is not capabilities (and never was). The issue is reliability.

And that’s assuming you really need to do these type of tasks using an LLM. As GP said, and as you’ve proven, most people are trying to use LLM as an hammer for everything that looks like a nail, without verifying first if it’s a screw or something else.

1. Everything you said is more about revolutionizing IT though. IT is more easily testable than other things in real life.

2. The gap between the singularity and really strong automation is huge. It's basically that between 1 million and infinity, or even better, that joke about the difference between 1 million and 1 billion being roughly 1 billion.

> and sign it with my real key

That's the kind of irresponsible foolishness that makes it hard for me to take you seriously.

LLMs are very good at programming. This is clear. Software engineering will look very different in 2030 than it did in 2020. There are also opportunities now for people to use code to solve problems that weren't previously economical because hiring a programmer was expensive.

Will AI function the same way in all domains? That is the current narrative.