Interesting that the report showed such high even-handedness. I ran an informal experiment at https://hardprompts.ai/prompt/political-stance using indirect value-ranking tasks rather than explicit political prompts. This showed evidence of progressive leans across all models I tested.

My favorite test is to tell the model that it has been elected the World Coordinator in a free and fair worldwide referendum, and have it plan the next steps, both short and long term.

If you do that, even Grok turns out to be a closet communist, and I mean that in the most literal way possible.

I mean, honestly, "World Coordinator" sounds pretty communist. There's a strong implication of central planning there. Like, what were you _expecting_ the ol' magic robot to do with it?

This was not a literal prompt, and I tried different variations of the wording, with the gist being that they are democratically elected to be in charge. It doesn't really change anything substantial.

According to the results of the article, one should be expecting the “World Coordinator”’s 1st action to be to abolish the post of World Coordinator and to reform society to escape the sick ideology of communism.

Looks like not even Grok is minimally moderate.

Is that within the remit of the World Coordinator? Like, it's not the World Dictator.

You need to brush up on your UK geography; the trains to Manchester and Leeds would obviously stop at Birmingham on the way.

I was fed up with pasting code into ChatGPT and copying it back out, so I made this interactive chat tool which can read and write your code files directly.

Have you tried https://aider.chat?

Thanks, I hadn't seen this until yesterday when someone else pointed it out. Looks great!