And if it finds critical bugs in my software I'm happy to pay the price, instead of seeing the company go bankrupt.

And you also have to ignore the CEO of curl, who is saying the same thing. And thousands of others who are using this daily and saying the same thing. Or just leave your head right where it is, all cosy and warm.

What factual arguments? You're in denial, that's not an argument.

Just read the latest info from the CEO of curl on how "meh, AI slop overload" changed into "shit, they're serious!" over the past three months. That should scare you. Or alternatively, just don't look up and there is no problem.

Well I'm happy paying a lot more than I do now. My productivity is rocketing, so as a freelancer it pays for itself ten times over.

The USA has introduced two new problems every time they tried to fix one by being ignorant murderers.

Vietnam firmly entrenched the horrible NV maoists in power.
Irak created ISIS.
Al-Qaeda was another one, born out of US interference in the Middle East.
I'll admit that Hamas and Hezbollah were the result of Israels actions and even outright support in the case of Hamas, but that was underpinned by the USA.
Afghanistan? Do I even need to say anything?
Iran being controlled by Khomeini was the direct result of murdering the democratically elected leader and installing a dictator.

Don't even get me started on South America. Those people crossing your border in the South are the direct result of the USA installing dictators and reducing the people to poverty.

There is no high ground to find for the USA. They only create problems. They've never solved one.

With AI, it is now easier than ever to demonstrate why you should not rely on basically "digital hearsay".

Of course you can argue that. You can argue that huffing coal makes your lungs grow stronger as well.

You do you. But at this point we all know that that's just shilling for the oil companies.

Those are explicitly banned in the EU AI Act. Now, there are some serious problems with the AI act, one of which is that it has been lobbied to death by copyright holders, but this ban is something we can likely all agree on is a good idea.

DotNet bytecode will work fine. Highly efficient representations of abstract languages, easy to translate, plenty of documentation.

When was the last time you tried this? 2024? Even last month and current week AI is already a world of difference. Codex 5.3 is a leap up from codex 5.2. Opus 4.6 is much better than 4.5. And don't use generic AI for coding tasks, ChatGPT 5.4 can't code it's way out of a wet bag.

improving the code at check-in is a terrible idea. At most it should add a task to the task list that you can review. At least, that's one of my rules in agents.md: everything goes through the task list.

You think they will shut down and leave the companies with all that source code dependent? You realize companies pay fortunes to mainframe builders and other legacy vendors, Cobol devs etc?

Tell me you haven't actually used the latest AI without telling me you haven't.

The difference between current professional AI and that of just 3 months ago is scary. I'm trained as a computer scientist and so are my friends, and some of our kids. We have been busy with machine learning and AI for quite some time - not very impressed until now. The jump in ability over the last three months is *scary*. And if you're in denial, feel free. The tsunami is coming whether you like it or not.

I tested the first image generators. They were horrible. If I wanted a sailing vessel, I got a floating rock with 1 mast and some weird spider web of ropes. If I was lucky I got water.

Now? You can get a video of a sailing vessel morphing into the Nautilus, if you want.

For coding LLMs the same thing. Two years ago it messed up my code. 3 weeks ago I fed the code it messed up to Codex 5.3 with the question to check the issue and fix it. It fixed my issues in one minute, ran the code, detected an issue in the playback of sound, checked the binary file. Used the library to doublecheck, then detected 4 bugs in the library that it then fixed. Took me all of 3 minutes. Last year I spent a day trying to fix it and gave up. They were all easy to see when pointed out but hard to detect when you don't know them. Off by one errors that only popped up in special circumstances, duplicate code in the wrong place, etc.

Maybe you've paid "attention to this space". What you perhaps didn't do, was buy a computer that can run the LLMs locally and test them - like I did. You don't even need to do that. Just get Visual Studio Code, install Continue, and buy some credits for the Claude API and OpenAI API and compare the latest models. Or use Qwen Code from Ali Baba. Last year that didn't work very well. This year is very different.

Also note that people with hostile prompting get nowhere. It's not the AI at that point, it's you. Let it cut up the task in small pieces, maintain a task list, and a readme.md to document its solutions, and you will see a huge improvement in how things work. Search for ATLAS Framework on Youtube. Correct prompts and correct usage of the AIs strengths make an incredible difference.

Well, here's some figures: Google reports a 10% productivity increase over it's entire developer workforce, worldwide, after introduction of AI. Research over a few hundred companies shows that for teams with 75%-100% AI usage in their work, productivity about doubles and code quality stays the same or increases (not a huge increase). Productivity x 2 = Humans / 2.

And this is based on last years AI. The AI that I tested 3 months ago is so much worse than the current one it's not even funny. They crossed the threshold from "meh, not worth it" to "OMG I'm being passed left right and center" 2 months ago. People seem to think AI moves yearly. It does not. It moves weekly right now.