Every generation has done this, including mine, but not with a 45B valuation. It's utterly insane, how much people willingly threw at a successful gambler.

What am I trying to sell you? Access to frontier models, that I have no control over, that I don't work for but work with?

The great thing about AI assisted architecture is, that you can start at a very coarse level and refine from there. If a lower level design stage discovers a flaw in my thinking, I can go back up. The biggest difference between AI assistants and a design team is, that the former will never go "please not this dummy again", if you show up with yet another "additional feature I need" or "how about we look at it this way and change the structure a seventh time?". You won't do this on purpose, but these things do pop up in a design phase (ideally there). AI assistants also have a much larger context window than my rotten brain, so they will reliably detect conflicts between header files and some far removed design spec of yours.

You can bounce a project for months (while doing other stuff) before you are happy with your design and can proceed to finer grained design steps or even code generation. It's almost a perfect match for submarine projects (which nobody would ever shill for, since "nobody has ever engaged in such a project").

UML was an attempt to do the same thing with people, but it's very hard to get a group of humans to do, what you can easily do with a few AI agents. As a result I have never seen a software project draw any real benefit from all these UML charts large companies just love to churn out.

STUXNET attack was aimed at a nuclear "research" facility, not at a municipal water supply.

Wrong approach. If you write code immediately, you will eventually get stuck in a rut, because an architectural concept shot from the hip typically doesn't hold water. I don't talk about javascript slop here barely holding a shitty web page together, I mean software that people actually depend on, and which you will have to maintain for years with updates and additions as the project scope moves and expands.

Once you sit down and actually design something, AI will help you by looking deeply into the system you try to build, trying to find use cases and edge cases. That's when you start refining your architecture. That's when software modules come and go, components split up and parts fuse with other parts. On each step a capable AI model will guide you through the process, highlight omissions, corner cases, inaccuracies and contradictions. Any code written at this point would be a complete waste of tokens. The refined product is so much better, that you will just shake your head looking at your initial concept. It's like a whole very capable team designed this product, when you actually did it all by yourself and just threw it at 3 or 4 different agents multiple times. That's an amazing asset for submarine projects BTW!

Second massive boost from AI is, that AI happily and quickly does all the boring stuff: advanced command line parsing, proper XML parsing, help texts. Your software suddenly does stuff, which it should and could have done years ago, but you never actually got around to do it.

I honestly believe, that we are going to see lots and lots of really bad AI slop software (or are already seeing it), but there is a strong potential to write really good software with a lot less effort than ever before.

I'm sure, that Microsoft's hand crafted "security fixes" would have never done such a thing :-D

In my experience AI does this exceptionally well already. You can describe your software architecture in excruciating detail, and AI (any frontier model FWIW) will happily dissect your plan and highlight all these little weaknesses, omissions, loose descriptions and contradictions, which would later hamper your software project. It's amazing, what it can find.

AI will also refine your architecture, and make very smart suggestions, if you have multiple options to chose from, and so on. Of course, it knows all open source software and man pages, so it can also offload some of your effort to existing software and libraries.

Your software development will slow down considerably, and quality will shoot up.

It's a profession like marketing, and in fact it seems to be what comes after marketing. As a parent of three teenagers I can confirm, that a decent social media campaign can yield more more than thousands of bill boards. There is absolutely nothing wrong with doing this professionally, even if old timers like me shake their heads.

Before I create a misunderstanding: I find that guy's conviction horrible, and the system clearly failed him hard. He lost money, freedom, job and probably most of his social circles, and I just wish the system has the means to help him rebuild most of what he lost.

Now to my actual point: I have been a juror in some CSA court case once and can report from personal experience. These are probably the kinds of cases with the worst jury decisions and the worst kinds of judge behavior. The alleged acts of the defendant are so abhorrent, that few people want to spend too many thoughts on the matter, not even the professional ones paid to deal with just that. If a jury is outraged by the crime, then wrong convictions are almost inevitable, unless an actually guilty person happens to sit there as defendant.

Yes, this really sucks, if you are the innocent accused one, and I wish, that courts would handle such cases more professionally to create fairer verdicts. There may not be an easy solution for this.

PS: yes, we handed out a too lenient sentence against the defendant back then, because we were unwilling to look deeper at the evidence against the accused. The little we had to look at was already plenty disgusting. This additional evidence would have shown much worse behavior that what we sentenced the guy for. Such miscarriages of justice can cut both ways.

I have heard the following statement from someone around me:

Main reason this person is alive today comes from the fact, that the deepseek provided "recipes" were too complex to realize. You can tell me whatever you want about "Muh Freedom", but these "cripple the model" features are a good thing. At some point OpenAI reported, that they face thousands such questions every day.

I am all for "open source models", as you call them, but would like to point out, that posting the weights is very much comparable to posting a binary, and it's definitely not comparable to posting sources. These "open weights" tell you nothing about how the model will respond to "certain questions" until you ask them. Such a model can still do useful work, if you can run it for yourself in private, but I personally would not offer it as a service to anyone else. YMMV, especially with models created out of reach for your jurisdiction.

Because China's business model is "do what the westerners do at half price". If they have to raise their prices to Anthropic/ChatAI levels, their whole unique selling point goes up in dust, and people suddenly start asking questions about the political posture of China.

Their low price may also be viable only due to Chinese subsidies, and they may be unwilling to spend infinite amounts of money on subsidies.

They did not hack Anthropic's secret software for training their models. Their model weights didn't leak either, but their Claude Code CLI thing did leak.

It's a cost estimate from AWS Health Dashboard, and with that word in the name the trillion $ amount actually sounds credible.

This would still not provide an explanation for trillion $ charges ...

I am sure, that being a high-wit allows you to tell me in detail, why South Dakota has a severe staffing problem, when California doesn't. I am sure, that your explanation will be entirely free of classism or anything which could possibly offend some snow flake. Please refrain from anecdotes of some trad wife who loves it there.

If you look for signs of collapse: Based on the stats discussed here I see these in South Dakota much stronger than in California.