MachineSex/configs/llm
Giorgio Gilestro 9c0210ba93 llm_society pilot v1 diagnosis + v2 knobs: raise inheritance fidelity, add elitism
Pilot v1: full and no_grounding indistinguishable - both arms pay a ~25%
distillation tax per generation and pin at the same mutation-selection floor,
so grounding has no differential to act on. v2 raises inheritance fidelity
(n_inherit 240->600, child epochs 2->3, founders 600x3) and adds two evenly-
applied knobs: elitism (top parent survives as an unmodified copy - overlapping
generations, compensating for lossy distillation where E11 had faithful
genotype copying) and n_parents (sharper truncation selection). v1 archived at
results/llm_society_pilot1; v2 running.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkRLcc18rwT2Lysu6PbG7v
2026-09-07 12:38:15 +01:00
..
directed.yaml llm_directed: directed sex (breed offspring + select on verifier) — E10 in real weights 2026-07-05 18:35:04 +01:00
directed_hard_hpc.yaml hard benchmark: the 7B "fusion wins / no headroom" results were saturation artefacts 2026-07-05 19:13:26 +01:00
directed_hard_seeds.yaml Phase 3: LLM-tier speciation + multi-seed firm-up of the recombination claims 2026-09-06 15:39:15 +01:00
directed_hpc.yaml llm_directed: directed sex (breed offspring + select on verifier) — E10 in real weights 2026-07-05 18:35:04 +01:00
epistasis.yaml epistasis_predicts: functional conflict, not weight geometry, predicts merge failure pre-merge 2026-09-06 17:19:46 +01:00
epistasis_compat.yaml epistasis_predicts: functional conflict, not weight geometry, predicts merge failure pre-merge 2026-09-06 17:19:46 +01:00
merge.yaml llm: first real-LLM prototype — recombining specialist LLMs (C2/C4) 2026-07-05 15:48:02 +01:00
merge_hpc.yaml hpc: PBS job scripts for Imperial CX3 (probe + scaled LLM merge run) 2026-07-05 16:02:09 +01:00
merge_seeds.yaml Phase 3: LLM-tier speciation + multi-seed firm-up of the recombination claims 2026-09-06 15:39:15 +01:00
moe.yaml llm_moe: the union operator (route/max-merge) vs fusion — and the regime flips at scale 2026-07-05 17:53:47 +01:00
moe_hard.yaml hard benchmark: the 7B "fusion wins / no headroom" results were saturation artefacts 2026-07-05 19:13:26 +01:00
moe_hard_hpc.yaml hard benchmark: the 7B "fusion wins / no headroom" results were saturation artefacts 2026-07-05 19:13:26 +01:00
moe_hard_seeds.yaml Phase 3: LLM-tier speciation + multi-seed firm-up of the recombination claims 2026-09-06 15:39:15 +01:00
moe_hpc.yaml llm_moe: the union operator (route/max-merge) vs fusion — and the regime flips at scale 2026-07-05 17:53:47 +01:00
society.yaml llm_society pilot v1 diagnosis + v2 knobs: raise inheritance fidelity, add elitism 2026-09-07 12:38:15 +01:00
society_smoke.yaml llm_society: the composed society at LLM scale (C3) — loop, tests, smoke green 2026-09-07 11:55:19 +01:00
speciation.yaml second review round: tempered claims, robust statistics, corrected technical statements 2026-09-06 17:55:46 +01:00
speciation_add.yaml Phase 3: LLM-tier speciation + multi-seed firm-up of the recombination claims 2026-09-06 15:39:15 +01:00