Merged Artificial Intelligence

[W.D.Clinger]

they do show the power of the latest AI systems at coding. But how good are they at coding something original?

I think that's going to depend on what you mean by original.

Among the various activities of software development, I think of coding as one of the least original, requiring more originality than running automated test suites but less originality than developing test suites. Activities that require greater originality are often discussed using words such as "design".

AI assistants have become really good at coding, and are getting better at design.

That is my opinion, not a clear consensus. Here, for example, is a web page that distinguishes software development from software engineering. That essay advises that software engineering is more likely to require more in the way of academic degrees, yet describes software engineering as a less creative discipline than software development. From what is written there, it seems the main point of software engineering is to remove creativity from the process as much as possible, even while saying software engineers "must master software design" along with testing and knowledge of tried-and-true (hence not very original) system architectures.

Having taught academic courses in software development that differed from courses in software engineering primarily in their title and in which department (CS vs EE) taught the course, my personal opinion is that licensed engineers have a vested interest in emphasizing differences between software development and software engineering, with one of those differences being software developers' greater creativity and originality versus software engineers' focus on excluding creativity and originality from the software process.

Edited to add: Margaret Hamilton, who invented "software engineering" as the name of a new discipline, had a bachelor's degree in mathematics with a minor in philosophy, and was about to embark upon graduate study in mathematics at Brandeis when she became interested in meteorology thanks to Edward Norton Lorenz at MIT. Lorenz was a mathematician and meteorologist who invented, or at least popularized, modern chaos theory as an area of mathematics.

 

[Gawdzilla Sama]

This season's biggest hit, "Gort and Mindy."

 

[Frank Newgent]

you can't "just throw compute at the problem"
You need dozens of agents working on the problem, being checked constantly if they still follow basic math rules.

Sure you can. That's what an LLM is.

And four wheels and a trunk is a Mercedes.

 

[dann]

Palantir CEO Alex Karp says AI companies want nationalisation to dodge accountability - CNBC TV18

Palantir CEO Alex Karp demands AI companies be accountable for their tech, not just regulators. He critiques Anthropic's safety proposals & warns of IP theft.

www.cnbctv18.com

in his usual unhinged style, alex karp talks about ai and ip

Alex Karp may be in for a big surprise if he thinks that can do whatever he wants with this piece of land just because he owns it:

Palantirs topchef har købt kæmpe skovområde i Sverige: Nu har svenske myndigheder undersøgt købet (DR.dk, Oct 15, 2026)
Det har svirret med spekulationer siden Palantirs topchef i hemmelighed købte 15.000 hektar bjergskov i Sverige.
Har den amerikanske tech-topchef Alex Karp tænkt sig at bygge AI-datacentre, dommedagsbunkere eller noget helt tredje midt i den svenske natur?
Det spørgsmål stiller lokale sig selv, mens svenske medier jagter et svar.

Palantir CEO buys vast forest area in Sweden: Swedish authorities have now investigated the purchase
Speculation has been rife since the Palantir CEO secretly purchased 15,000 hectares of mountain forest in Sweden.
Does American tech executive Alex Karp intend to build AI data centers, doomsday bunkers, or something else entirely in the heart of the Swedish wilderness?
Locals are asking themselves this question, while Swedish media are trying to find the answer.

Freedom to roam: Europe: Nordic countries (Wikipedia)

Maybe Helen can tell us more

 

[Frank Newgent]

Jon Stewart asks Jacob Coxon some pretty good questions.

 

[Wudang]

I think that's going to depend on what you mean by original.

Among the various activities of software development, I think of coding as one of the least original, requiring more originality than running automated test suites but less originality than developing test suites. Activities that require greater originality are often discussed using words such as "design".

Yes. Especially after software people read books on architecture about patterns. And even before then, last century I worked in a software lab where the specs were all written in Z and the job of converting those specs to PL/AS rarely called for originality.

AI assistants have become really good at coding, and are getting better at design.

That is my opinion, not a clear consensus. Here, for example, is a web page that distinguishes software development from software engineering. That essay advises that software engineering is more likely to require more in the way of academic degrees, yet describes software engineering as a less creative discipline than software development. From what is written there, it seems the main point of software engineering is to remove creativity from the process as much as possible, even while saying software engineers "must master software design" along with testing and knowledge of tried-and-true (hence not very original) system architectures.

Yes understanding the problem and seeing how to address it correctly requires varying degrees of originality. Implementing the design has been slowly becoming more automated in IDEs for a very long time. E.g. right-click on a class and generate getters and setters etc. And avoiding originality means that the next person to work on the code some time later (and it might be you) will grasp the code quicker. And of course code reuse, one of the original aims of OOPS, adds to this.

Having taught academic courses in software development that differed from courses in software engineering primarily in their title and in which department (CS vs EE) taught the course, my personal opinion is that licensed engineers have a vested interest in emphasizing differences between software development and software engineering, with one of those differences being software developers' greater creativity and originality versus software engineers' focus on excluding creativity and originality from the software process.

Partly related, at 2 big corporates I was horrified at how many people I came to refer to as "coders" did not understand programming. They understood how to take specs and turn them into code in the specific language they were trained in.
I am impressed by what LLMs can do generating code. From what I see though I still have massive doubts about the technical debt being built in. Also and looping back to those "coders", what happens when something goes wrong? I've had to be called in a few times for such teams when their code broke. Their code was usually simple, linear, and optimistic and hence relatively easy to spot the problem. A lot of LLM code I've seen is far concepts like simplicity, elegance etc.

 

[Wudang]

And here's a nice sweary little rant about Linus Torvalds' adoption of LLM in the Linux kernel. Note to self: check out NetBSD now.

mas.to

mas.to

 

[W.D.Clinger]

Disclaimer: I have never had any direct interactions with an LLM, unless you count a few lines of spoken dialogue with Alexa as I and a friend were trying to figure out why a 90-year-old friend's Echo was not responding reliably to her voice, and was not responding at all to our voices. (Our friend had changed Alexa's name to Ziggy, but thought she had changed the name to Siggy. Her pronunciation of Siggy was at times close enough to Ziggy for Echo to realize it was being addressed, but our younger voices were pronouncing Siggy too clearly for the Echo to realize we were talking to it. We had to consult the Echo control panel on our friend's computer to figure out what was going wrong.)

I am impressed by what LLMs can do generating code. From what I see though I still have massive doubts about the technical debt being built in. Also and looping back to those "coders", what happens when something goes wrong? I've had to be called in a few times for such teams when their code broke. Their code was usually simple, linear, and optimistic and hence relatively easy to spot the problem. A lot of LLM code I've seen is far concepts like simplicity, elegance etc.

Although I have never used an LLM to generate code, I have reviewed a lot of LLM-assisted code within the past two weeks, and will be reviewing a lot more of it going forward.

I too am impressed. The code I've been reviewing was written by a truly outstanding programmer, who now works at Anthropic, in collaboration with Anthropic's Claude 5.5, and most of that new code modified old code written by a different truly outstanding programmer or by myself.

The paragraph above may go some way toward explaining why my opinion of LLM-assisted code is more favorable than David Gerard's.

And here's a nice sweary little rant about Linus Torvalds' adoption of LLM in the Linux kernel. Note to self: check out NetBSD now.

mas.to

mas.to

I read several of the comments on that rant, and laughed at a couple, including this one:

no matter how much you glaze him, torvalds will never know who you are​

 

[dirtywick]

personally i have no problem believing guys who write code for a living could code a program that’s good at writing code.