Thanks, I get your point. However I think a big misconception about LLMs is that they think similar to us and what is easy for us should be easy for them. I more agree to Kurzweil‘s statement that they are very alien intelligences, capable of doing what we can’t and incapable of some "simple“ tasks

No I’m not being sarcastic - if I as a user know about that router and the two modes and actually want to get things done then I can get them done bypassing the router with a few words. So I don’t really see the point of the OP except for "look how dumb the LLM is“.

If the OP had asked this reasoning-capable model to reason it would have reasoned and got the right answer. But he didn’t so it didn’t. The only issues I see is that gpt5 falsely thought this is not a task worthy of reasoning. It helps to know the tools you are working with.

Das ist aber doch keine deutsche Ingenieurskunst sondern grünsiffige Ökotechnologie. Sowas machen wir nicht. Unter der Ampel vielleicht ein bisschen. Aber jetzt sind wir wieder modern und auf Kurs. Hat mein Kanzler gesagt.

Das muss man erst mal schaffen, 100 Milliarden zusätzlich bekommen, um unser Land zukunftsfähig zu machen, und dabei die #1 Bedrohung unserer Zukunft einfach abzutun.

Major changes since my last update. A few people are actually playing #glyphgrid and telling me they like it 😀. So I added a global Highscore list.

Also finally have a real nice professional URL with https for it.

And still: no tracking. No ads. Offline after first load. Resume game anytime.

Gemini 2.5 Pro did a good job on these 3 features from one "spec prompt," but messed up an if-then-else chain for game starts (fixed it). Still doubting non-devs can confidently develop non-trivial software with LLMs.
#llm #vibecoding #googlegemini

Ugh, neue US „advanced interrogation technique“? 😜

Thanks, this is exactly what I feel. Every discussion about AI gets into "holy war“ mode with the same points being regurgitated, and serious discussion about so many bits and pieces about what AI can do and will do is almost impossible. Might be the medium - IRL I have great discussions about this.

Happy with the way the game gets more difficult now. Previous version didn't get more difficult after level 25, and had too many simple shapes (1x1, 1x2) even on very high levels so you could always recover.

And sharing on Bluesky now works for iOS devices.

I get your point, but I'm not sure if I agree. Like any other tool an LLM is "garbage in, garbage out". Just like entering =SUM(A1:A8) into Word won't get you anywhere, a crappy prompt will get you a crappy answer. I'm coding with AIs every day, and never got an inappropriate answer.

Yes, and pocket calculators and excel are fucking lazy and morally wrong. Learn to calculate on paper and fight the destruction of meaningful creativity.
(Additional tools that do the craftsmanship part of art do not not hinder creativity, not in math, software, art, …).