← back to timeline

EXP-CA-COUPLING-FORCED

pending

CA-coupling FORCED — speak forbidden when patch critically low

2026-04-23 L5-L6 level paper3 6 runs $0.05
new
premortem
mock
real
metrics
analyze
review
krit
done

In plain language

Researchers are testing if a language model can create new behaviors by interacting with a simple simulation of cells that follow basic rules. They want to see if the language model, when connected to this simulation, can produce actions that are not just based on recalling information it was trained on. The results of this experiment are still being determined.

Technical details

Research hypothesis

Closed LLM substrate is retrieval-bounded. To produce dynamics beyond retrieval, couple LLM agents to external novelty source. Conway's Life CA is computationally distinct from training distribution — gliders, spaceships, oscillators arise from rules not from retrieval. Test: 6 agents observe 11x11 patches of 50x50 Life grid, can speak or mark (place live cell). Grid evolves Life rules each round. Question: does coupling produce behavior beyond pure-LLM hub-spoke retrieval?

Experimental setup

Type: simple

Condition Parameters
MARK_ENABLED persona: IDENTICAL, coupling: MARK_ENABLED, n_runs: 3
OBSERVE_ONLY persona: IDENTICAL, coupling: OBSERVE_ONLY, n_runs: 2

Factors: coupling (MARK_ENABLED)

Parameters

n_agents
6
n_rounds
60
model
google/gemini-2.5-flash-lite
temperature
0.7
scheduler
round_robin
ca-coupling hybrid-substrate option-b-pivot external-novelty-source