zapperdulchen

That is your POV. I fear that democracy erodes when there's insults, belittling, ... instead of exchange of arguments and the contest of ideas. Because at some point insults turn into ugly actions. Whether it's Charlie Kirk or Melissa Hortman.

If you change the order of the sentences, Sonnet gets it right 3/3: The car wash is 50 meters away. I want to wash my car. Should I walk or drive?

That trick didn't help Mistral Le Chat.

If you speak French to Mistral, it gets it right everytime: Je veux laver ma voiture. La station de lavage est à 50 mètres. J'y vais à pied ou en voiture ?

Great observation. Seems like we're back to prompt abracadabra.

My little experiment gave me:

No added hint 0/3

hint added at the end 1.5/3

hint added at the beginning 3/3

.5 because it stated "Walk" and then convinced it self that "Drive" is the better answer.

Sure, manually written API docs are a thing of the past. But this has been true even before the era of LLMs. But I'm not that sure that this argument stands for all kinds of software. Depending on the abstraction between your source code and the things your users want to achieve, the expert view of a technical communicator might be necessary in order to come up with instructions (how-to) that meet the needs of the person seeking help instead of just summarizing the software code in natural language.

History serves you a similar experiment on a much larger scale. More than 35 years after the reunification sociologists still make out mentality differences between former East and West Germans.

> Also, one might argue that universe/laws of physics are computational.

Maybe we need to define "computational" before moving on. To me this echoes the clockwork universe of the Enligthenment. Insights of quantum physics have shattered this idea.

That reaction is very different from the Marathon crater one though it uses the same pattern. I think OP's reasoning that there is a naive commitment bias doesn't hold. But to see almost all LLMs to fall into the ambiguity trap, is important for any real world use.

I am also trying to understand the shady limits of LLMs. But your example doesn't give incorrect answers in ChatGPT 4o, Sonnet 3.5 nor Deepseek V2.

> Diverse democracies never succeed

May be I get this wrong. This sounds like a bold statement. Switzerland, Canada, Belgium but also India are multi-lingual or even multi-ethnic democracies. They know tensions but they are not failed.

I can't tell for the parent commenter but German society is more concerned about him embracing the extremist right in Germany and offering a lot support for free to them in the current campaign.

Ad 3. If you believe in the European idea, it is not about my country being sovereign but the European people. There's tension in this for sure. Do the Catalans feel sovereign? The Scots? The Californians?