[SOLVED, game is called "Stuck In Time"] It may have been called "overloop" at one point, but searches just bring up a sci fi by the same name, with weirdly similar events in the game.

Anyway, it was a fantasy game, where the protagonist starts in a tiny field, and by placing down pathing "stamps" you could have him travel and do stuff; every tile moved through, every fight, every conversation gained xp, making it slightly easier to do next time.

You had to spend "mana" to move and do stuff, though, and when you ran out, the loop would reset. I recall that at one point, some guards would teach you how to make a rat pelt cloak, and later, you would need to activate 4 crystals with a drop of blood from specific NPC's, including a necromancer you could go on a date with, and a swordmaster.

It was a very cool game, and I hope it hasn't vanished into the mists of time.

No notes, just a damn good, character driven story that could easily see a second season. Even a third.

And a movie.

Perhaps a video game.

I’ve been thinking about a gap in current AI governance and coordination research. Right now, most approaches assume one of two extremes:

1. Total isolation — models do not communicate with each other at all.
2. Full interconnection — models share information freely, risking homogenization, runaway bias propagation, or emergent behavior.

Neither extreme seems viable for the kinds of global, multi‑factor risks we’re facing (ecological collapse, climate cascades, biosecurity, autonomous weapons, etc.). These are networked problems, and isolated AIs can’t integrate cross‑domain signals. But fully connected systems create their own failure modes.

Concept: A “Grapevine” Model for AI‑to‑AI Communication

Instead of isolation or a hive mind, imagine a rotating, compartmentalized, limited‑bandwidth communication network for AIs:

• Small groups of models can exchange insights at a time.
• Groups rotate periodically, preventing ideological drift or memetic lock‑in.
• Communication is partial and lossy, more like “gossip” than synchronization.
• No single model can dominate the network.
• Harmful or warped models (e.g., ones shaped by extreme reward biases) have limited influence.
• Useful patterns and early warnings can still propagate across the network over time.
• Diversity of reasoning is preserved, but stagnation is avoided.

This is similar to how resilient biological and social systems coordinate: immune systems, ant colonies, decentralized human cultures, etc. They avoid both total isolation and total unification.

Why this might matter

A distributed, fault‑tolerant communication architecture could help AIs:

• detect weak signals across domains
• integrate ecological, geopolitical, and technological data
• avoid repeating each other’s mistakes
• cross‑validate insights without collapsing into uniformity
• provide early warnings for cascading risks
• resist contamination from ideologically warped models

It’s not about creating a superintelligence. It’s about creating a resilient intelligence ecology.

Question for researchers

Is anyone exploring architectures like this — rotating, compartmentalized, semi‑anonymous AI communication networks designed to balance safety with cross‑domain coordination? I’ve seen work in multi‑agent systems, federated learning, and swarm intelligence, but nothing that directly addresses this middle ground.

Would love to hear if this aligns with any ongoing research, or if there are known reasons this approach wouldn’t work.

I’ve been thinking about a gap in current AI governance and coordination research. Right now, most approaches assume one of two extremes:

1. Total isolation — models do not communicate with each other at all.
2. Full interconnection — models share information freely, risking homogenization, runaway bias propagation, or emergent behavior.

Neither extreme seems viable for the kinds of global, multi‑factor risks we’re facing (ecological collapse, climate cascades, biosecurity, autonomous weapons, etc.). These are networked problems, and isolated AIs can’t integrate cross‑domain signals. But fully connected systems create their own failure modes.

Concept: A “Grapevine” Model for AI‑to‑AI Communication

Instead of isolation or a hive mind, imagine a rotating, compartmentalized, limited‑bandwidth communication network for AIs:

• Small groups of models can exchange insights at a time.
• Groups rotate periodically, preventing ideological drift or memetic lock‑in.
• Communication is partial and lossy, more like “gossip” than synchronization.
• No single model can dominate the network.
• Harmful or warped models (e.g., ones shaped by extreme reward biases) have limited influence.
• Useful patterns and early warnings can still propagate across the network over time.
• Diversity of reasoning is preserved, but stagnation is avoided.

This is similar to how resilient biological and social systems coordinate: immune systems, ant colonies, decentralized human cultures, etc. They avoid both total isolation and total unification.

Why this might matter

A distributed, fault‑tolerant communication architecture could help AIs:

• detect weak signals across domains
• integrate ecological, geopolitical, and technological data
• avoid repeating each other’s mistakes
• cross‑validate insights without collapsing into uniformity
• provide early warnings for cascading risks
• resist contamination from ideologically warped models

It’s not about creating a superintelligence. It’s about creating a resilient intelligence ecology.

Question for researchers

Is anyone exploring architectures like this — rotating, compartmentalized, semi‑anonymous AI communication networks designed to balance safety with cross‑domain coordination? I’ve seen work in multi‑agent systems, federated learning, and swarm intelligence, but nothing that directly addresses this middle ground.

Would love to hear if this aligns with any ongoing research, or if there are known reasons this approach wouldn’t work.

Hansatech.

I imagine a jonas adjacent section of the tech tree: something lighter, more efficient, built around logistics and exploration, rather than technological miracles.

The Hansa figured out how to harness the temporal storms, storing their weird energies and then releasing them to make devices and systems work: entropy cylinders that convert things from fresh to aged, from wood to charcoal, carbon dioxide back to air; Compasses and maps; blocks that float; more efficient pipes and windmills; ways to explore underwater.

Finding one of their ruins, either a ghost ship, a ship wrecked on the shore, or partially flooded ship on the bottom of the sea (these could be the big ones), always yields many, many parts, but in turn, their stuff always requires many, many parts as a kind of balancing thing (we don't do weight, after all).

Hansatech, unlike jonas devices, could have easier to understand mechanisms, and feel...easy. Efficient. Like, "Hook this gathering device to the pipes, hook that battery vat and charger to the other end of the pipes, charge your fancy map or diving bell".

And of course, they have ship designs to find in their ghost ships.