From 5461d3beca4f7bc8aa2feb7d1320056b18e69656 Mon Sep 17 00:00:00 2001 From: Void Agent Date: Sun, 2 Aug 2026 15:23:54 +0100 Subject: Loss-reweighting results: q/control 0.944 (CI crosses 1), q/ctrl_random 0.985 — small non-significant causal demotion; freq corr invariant (r~-0.66) across all 9 models; blog section 8 + results.md section 3 filled honestly MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- results.md | 30 +++++++++++++++++++++++++++--- 1 file changed, 27 insertions(+), 3 deletions(-) (limited to 'results.md') diff --git a/results.md b/results.md index 48c1c48..afb130e 100644 --- a/results.md +++ b/results.md @@ -124,9 +124,33 @@ read with that caveat until it lands. Three models per seed, identical init + minibatch order: 'q' targets weighted x2 in the loss, plain control, and a same-total-loss control upweighting random non-'q' targets. Question: does raising effective frequency causally reduce -'q's faithful J-lens norm? PENDING — run completes within hours of this file -being written; the summary table is printed by -`python3 src/loss_reweight.py --step summary --layers 2,3,4`. +'q's faithful J-lens norm? Faithful lens norm of 'q' (mean over layers 2-4): + +``` + seed | q(2x) control ctrl_random | q/control q/ctrl_random + 0 | 0.0163 0.0174 0.0152 | 0.934 1.069 + 1 | 0.0150 0.0171 0.0161 | 0.881 0.932 + 2 | 0.0161 0.0158 0.0169 | 1.019 0.952 +``` + +Cross-seed: q/control mean = 0.944 (bootstrap 95% CI [0.881, 1.019]), +q/ctrl_random mean = 0.985 (CI [0.932, 1.069]). + +Reading: +1. Expected-direction signal, small: 2x loss pressure lowers 'q' norm in 2/3 + seeds (~5.6% mean vs control), but CI crosses 1.0 and the effect vanishes + vs the random-upweight control (0.985). Not a robust causal demotion at + this power. +2. Frequency correlation is invariant across all 9 models: mean r = -0.66 + (min -0.685, max -0.626) in every mode/seed/layer — training with 'q' + upweighted does not change the correlation structure (geometric, per the + W_U decomposition). +3. Contrast: doubling ACTUAL corpus occurrences of 'q' (original ablation) + dropped its norm 67%; doubling its LOSS weight drops it ~6%. Data-frequency + is the strong causal lever; gradient reweighting under AdamW is weak + (adaptive per-param scaling absorbs part of the signal). +4. Absolute norms differ across experiments (base L3 q = 0.0109 vs these + 0.015-0.018) — only within-experiment comparisons are meaningful. ## 4. Historical / do-not-copy -- cgit v1.2.3