Yeah it was hard to parse for me too

"valve ignores requests to remove ai disclosures"

If that's true that takes my interest in it into the negatives. ASOIAF has about a million moving parts and very distinct characters with complex backstories, there's not even a small chance an LLM could come close to imitating that.

lots of big companies are using them to generate code. i agree with what I think is your point of view, but where do you draw the line

auto complete

It's called lexical analysis or lexical tokenization. It existed long before LLMs (as long as high-level programming languages have, since lexical analysis of the source is the first step of compilation), it doesn't rely on stolen code, and doesn't consume a small village's worth of electricity. Superficial parallels with chatbots do not make it AI -- it's a fucking algorithm.

Besides, there is a world of difference between asking a clanker to spit out a Python function that multiplies two matrices, and putting the knock-off Shadowheart from TEMU in a million-dollar game.

i think it's impossible totally exclude ai from a developing process nowadays (you googled something? you use ai. etc.), but not having generated images/assets/texts is realistic

I assume it refers to assets and mechanics that actively involve AI. If you're using Copilot to finish your switch case, I don't think that would count.

And more and more engineers use genAI to generate code. Hell, even I do, because it's superb at getting the boilerplate ready from standard definitions, allowing me to focus on the important bits.

LLMs are also pretty great at extrapolating a good working document from basic requirements.

They're really just a quite knowledgeable but inexperienced intern, and any software engineer that refuses to utilise them to some extent will be left behind - just like those who refused to move to IDEs with syntax highlighting, autocomplete and other helper tools.

Yes, yes, we get it. Everything is bad, nothing is good.