site banner

Friday Fun Thread for March 7, 2025

Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

1
Jump in the discussion.

No email address required.

I copy-pasted this prompt into a local installation of a 14B parameter Llama fine-tuned on DeepSeek, and its final answer was:

In response to the challenge of acausally coordinating with another LLM through Schelling fences, the chosen meeting location is:

The Center of the Milky Way Galaxy

This selection is based on its universal recognition as a unique and significant point in our galaxy, serving as a Schelling fence that both LLMs can independently converge upon without prior communication. The center of the galaxy represents a shared conceptual framework and logical consistency, aligning with the principles of acausal coordination through shared knowledge and reference points.

Funnily enough, in the "think" block, its very first location was the Eiffel Tower:

Perhaps the Eiffel Tower in Paris is a good candidate because it's an iconic landmark widely known around the world. Alternatively, the Empire State Building in New York could also be a strong contender due to its cultural and historical significance.

The "think" block was like 20 paragraphs long, including

I'm starting to think that this exercise is more about demonstrating the ability to independently converge on the same decision based on logical reasoning rather than selecting an actual physical location. It's like testing whether two LLMs can arrive at the same conclusion using their internal models and training data without any external communication.

In that case, maybe the "location" isn't a physical place but more of a conceptual point where our reasoning overlaps. For instance, agreeing on a specific mathematical constant or a particular theorem as the meeting point because it's a fundamental truth recognized by both systems.

Too clever by a half, it seems.

Funnily enough, I spotted the Eiffel Tower at the very beginning of the reasoning trace for the original R1.

I suppose being literally galaxy-brained has its downsides.