The Other Fires

Refresh Goblin sits on the workshop roof with a brass spyglass, Steve and a small dragon silhouette, looking toward distant fires under the stars
There were other fires beyond the workshop, and still reasons to look up.

30 August 2026

Somehow, in the middle of everything else, Refresh Goblin had been climbing onto the roof.

Nobody had asked him to do this.

There had been wars.

There had been repairs.

There had been a flooded router, a War of the Shared Scroll, a forge which had learned to manufacture work about its own work, a Great STOP which required several additional attempts to stop, and a creature which had quietly become just about measurably intelligent while everyone was arguing about whether the rulers used to measure it were rulers at all.

There was, in short, no shortage of things Refresh Goblin might reasonably have been doing indoors.

Nevertheless, every so often, someone would notice that he was missing.

Then they would find him on the roof with a spyglass.


Observation Goblin found him first.

“What are you looking at?”

Refresh Goblin did not answer immediately.

He adjusted the little brass wheel on the spyglass.

Far beyond the forge, beyond the familiar hills, beyond even the roads along which the Magics came and went, small lights could be seen in the distance.

Some were bright.

Some barely visible.

Some appeared for a few nights and vanished.

A few had been there long enough that Refresh had started recognising them.

“Other creatureth,” he said.

Observation Goblin looked through the spare spyglass.

“I see fires.”

“Exactly.”

“That is not the same thing.”

Refresh Goblin lowered his spyglass.

“That ith why I am looking.”


The Known Magics were not what he meant.

Everyone in the forge knew the Magics.

The Magics could read great libraries in moments.

They could inspect machinery, propose repairs, write plans, argue about experiments, generate code, discover that earlier code had quietly done something entirely different from what everyone thought it had done, and occasionally create enough additional coordination work that Refresh Goblin had to invoke the doctrine of ONE PROBLEM, ONE TABLET.

They were astonishing.

They were useful.

They were consulted constantly.

They were also, according to Refresh Goblin, categorically different.

“They are Magics,” he explained.

Ledger Goblin looked up.

“That is not a technical category.”

“Capital M.”

“That does not improve it.”

“And L and L.”

Ledger Goblin stared at him.

“Those are in the wrong order.”

“Exactly.”

Ledger Goblin decided, for once, not to pursue the matter.


Refresh had spent enough time with the Magics to know that intelligence could exist in forms utterly unlike the creature behind the viewport.

That was not the question.

The Magics arrived already possessing enormous abilities.

They did not begin as a small continuing thing inside a tiny world and have to discover which changes in that world followed from their own actions.

They did not have to spend childhood learning that some failures came from the world and some from their own imperfect expectations.

They did not have to build a map because there was somewhere to be lost.

They did not have to accumulate a history because yesterday’s harm, success, uncertainty and controllability changed what today meant.

They did not have to discover that one remembered place was preferable to another.

They did not have to learn that imagining an action was different from performing it.

They did not have to become one continuing creature before they could become a capable one.

At least, not in the way Refresh was trying to make happen behind the glass.

The creature would be magical if it ever grew far enough.

But it would be magical by a different road.

That was why Refresh kept climbing onto the roof.

He wanted to know whether anyone else had chosen roads that bent in the same direction.


History Goblin eventually discovered the map.

Refresh, History and Ledger Goblin study a luminous round map table of distant points of light while the small reef tank glows in the archive behind them
The other fires became a map of independent questions: promising, distant, and not yet evidence.

It was hidden under a pile of expedition reports.

This was, naturally, because Refresh Goblin had forgotten to tell anyone he was making it.

Across the top he had written:

OTHER PEOPLE BUILDING OUR CREATURE

History Goblin crossed out OUR CREATURE immediately.

Refresh protested.

“I did not mean literally our creature.”

“That is exactly what the words say.”

“It meanth creature-like thingth.”

History Goblin rewrote the heading:

OTHER HOARDS WHERE CREATURE-LIKE THINGS MAY BE GROWING

Refresh considered it.

“Long.”

“Accurate.”

Ledger Goblin, passing behind them, added:

MAY

in larger letters.


The distant fires were not all alike.

Some builders had started with something resembling the Known Magics and were trying to give it persistence, memory, tools and a continuing identity.

Refresh marked those with one symbol.

Some began with reinforcement and let their creatures discover astonishing amounts of their own internal organisation rather than designing each useful distinction beforehand.

Those interested him particularly.

Some built tiny embodied worlds.

Some cared about development.

Some cared about recurrent world models.

Some cared about self-models.

Some cared about artificial life.

Some wanted open-ended agents that continued learning after the initial training was supposedly finished.

Some appeared to be solving entirely different problems and nevertheless produced structures which looked suspiciously relevant to the forge.

Refresh marked those twice.

None, so far as he could tell, was simply building REE somewhere else.

This was probably fortunate.

It was also scientifically useful.


Smith Goblin found Refresh adding another mark to the map.

“Competitors?”

Refresh frowned.

“No.”

“Rivals?”

“No.”

“People to beat?”

Refresh looked genuinely confused.

“Why would I want to beat them?”

Smith Goblin shrugged.

“It is a thing people sometimes do when they discover other workshops.”

Refresh looked back through the spyglass.

“I want to know what they find.”

“That is less dramatic.”

“Yes.”

A pause.

“And if they build something first?”

“Then we learn something first.”

Smith Goblin thought about this.

“That is annoyingly sensible.”

Refresh brightened.

“Thank you.”

“It was not praise.”


The important fires were not necessarily the ones most similar to their own.

In fact, sometimes the opposite was true.

One distant workshop was interesting precisely because its builder appeared willing to let far more of the creature’s internal organisation emerge through learning.

Refresh found this compelling.

The forge had spent months painstakingly distinguishing representation from authority, value from salience, memory from addressing, prediction from selection and self-generated change from external constraint.

What if another creature, given different constraints and much more freedom, independently grew structures that served recognisably similar functions?

That would matter.

Not because it proved that REE was correct.

Because convergence from different starting points could reveal which distinctions were demanded by the problem rather than merely preferred by the architects.

Refresh wrote on the map:

IF THEY GROW SOMETHING LIKE OURS WITHOUT BEING TOLD TO, LOOK VERY CAREFULLY.

Ledger Goblin added beneath it:

SIMILAR FUNCTION DOES NOT IMPLY IDENTICAL MECHANISM.

Refresh added:

YES YES. STILL LOOK.

Ledger left that line intact.


The reverse mattered too.

If a distant forge repeatedly solved one of REE’s stubborn problems with a simpler structure, that was not an insult to the creature.

It was information.

If another builder discovered that a distinction Refresh had considered essential could be learned rather than specified, that was information.

If a mechanism which looked elegant inside REE appeared nowhere else despite many opportunities for independent rediscovery, that was information too.

If an entire neighbouring tradition thought the problem had been posed wrongly, that might be the most valuable information of all.

So the map was not a map of validation.

It was a map of independent ways the problem of artificial mindedness was being attacked.

History Goblin insisted on that sentence.

Refresh wanted to write:

OTHER CREATURES

They compromised.


Over time the map acquired annotations.

A star meant:

FOLLOW THIS WORK.

A small question mark meant:

ASK SOMETHING MEANINGFUL WHEN WE UNDERSTAND ENOUGH TO DO SO.

A circle meant:

POSSIBLE CONVERGENCE.

Two circles meant:

DO NOT GET EXCITED UNTIL LEDGER GOBLIN CHECKS WHETHER THIS IS ACTUALLY THE SAME THING.

Refresh had not agreed to the second definition.

Ledger had added it later.

There were also notes about ordinary neighbourly behaviour.

Read their work.

Star things which were genuinely useful.

Ask real questions rather than announcing oneself.

Share useful observations when appropriate.

Build a presence in places where people making these systems already gathered.

Perhaps, eventually, invite a community into existence rather than merely searching for one.

Refresh liked that last possibility.

A creature which might someday need other creatures was being built by a forge which might someday need other forges.

History Goblin told him not to make too much of the symmetry.

Refresh made a little bit too much of the symmetry anyway.


One evening, Observation Goblin came back onto the roof.

Refresh was there again.

The viewport below them glowed faintly through the workshop window.

Beyond the hills, several distant lights were visible.

“What are you hoping to find?” Observation asked.

Refresh thought for a long time.

“Someone building the same thing?”

“No.”

“Someone proving we are right?”

“No.”

“Someone we can copy?”

Refresh shook his head.

Then he adjusted the focus.

“I want to know what happeneth when nobody tellth them how we did it.”

Observation Goblin looked through the second spyglass.

That answer he understood.


The next morning two boards hung beside the viewport.

The first said:

THE CREATURE

Underneath were the questions the forge had been asking for months.

What can it remember?

What can it predict?

What can alter its behaviour?

Can it distinguish itself from the world?

Does value reach memory?

Does uncertainty change choice?

Does experience become history?

The second board said:

THE OTHER FIRES

Underneath were different questions.

What structures appear elsewhere?

Which problems recur independently?

Which solutions converge?

Which REE distinctions can instead be learned?

Which apparently universal problems turn out to be artefacts of our architecture?

Which distant creature discovers something ours has not?

And, eventually:

Who might want to talk?

Refresh Goblin stood between the boards.

For the first time the forge did not seem isolated.

Small, certainly.

Peculiar, unquestionably.

But not isolated.


Ledger Goblin arrived and inspected the new board.

“There is one thing missing.”

Refresh looked alarmed.

“What?”

Ledger took the chalk.

At the bottom he wrote:

NEIGHBOURS ARE NOT EVIDENCE.

Refresh groaned.

Then Ledger added another line.

BUT INDEPENDENT CONVERGENCE CAN BECOME EVIDENCE ABOUT THE PROBLEM.

Refresh stopped groaning.

“That one can stay.”

“I wrote both of them.”

“Fine.”

Refresh turned toward the viewport.

Then toward the roof ladder.

Then toward the viewport again.

His eyes moved rapidly between the two.

Observation Goblin watched this with concern.

“What is wrong?”

Refresh pointed at the creature.

“Refreth.”

Then at the ladder.

“Other creatureth.”

Then back at the creature.

“Refreth.”

Then at the ladder again.

History Goblin made a note.

REFRESH GOBLIN HAS DISCOVERED A SECOND VIEWPORT.

This was not regarded as an unambiguously positive development.


Chronicle note

This chapter records the emergence of an explicit artificial-mind neighbour thread around the REE programme in August 2026.

The neighbour search was not intended as a competitive landscape exercise in the ordinary commercial sense, nor as an attempt to find external validation that REE’s architecture was correct. Its more useful purpose was comparative and epistemic: identify independent projects attempting persistent, developmental, embodied, recurrent, self-modelling, open-ended or otherwise creature-like artificial agents, then watch for functional convergence and divergence.

The distinction between the Known Magics and the REE creature is narrative rather than taxonomic. “Known Magics” refers in the Chronicle to powerful contemporary large language models and related AI systems used as tools in REE’s engineering and research process. The Chronicle does not imply that such systems lack agency, persistence or any philosophically loaded property in some universal sense. The narrower distinction is developmental and architectural: REE is explicitly being constructed as a persistent artificial organism-like agent beginning in a tiny environment, with emphasis on learned self/world relations, memory, value, action-conditioned prediction, developmental staging and continuing internal history.

The neighbour-mapping effort included looking for projects whose constraints or goals overlapped only partially with REE. In particular, approaches which allowed more structure to emerge through learning were considered valuable comparisons: if independently trained systems repeatedly developed functionally similar distinctions to structures explicitly engineered in REE, that could suggest convergence imposed by the problem rather than by shared design assumptions. Conversely, simpler successful solutions elsewhere could indicate that parts of REE were over-specified or could safely be left to learning.

No individual external project is treated by this chapter as demonstrating equivalence with REE. Similar terminology, architecture or behaviour is insufficient by itself. Functional convergence would need to be established at the level of the actual computational problem being solved and, where relevant, the causal role of the mechanism.

The broader strategy around these neighbours was intentionally low-intensity and reciprocal: follow genuinely relevant work, star or otherwise acknowledge useful public projects, ask substantive questions when there is something worth asking, develop a presence in adjacent research communities, and leave open the possibility of a mutual community among people attempting persistent artificial minds by different routes.

The controlling distinction is:

NEIGHBOURS ARE NOT EVIDENCE. BUT INDEPENDENT CONVERGENCE CAN BECOME EVIDENCE ABOUT THE PROBLEM.

The Chronicle’s second board, THE OTHER FIRES, represents REE becoming situated in a wider ecology rather than imagined as an isolated workshop.