Skip to content

Pipeline99+RunPods1 Approvals60 Project Log Literature99+More

Public view

Work

Pipeline99+RunPods1 Approvals60 Project Log Literature99+

Manage

Plan needs approval3

#369 Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone5/14/2026
#370 #351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation5/14/2026
M1 QA pending approval5/10/2026

Promotion approval55

LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)5/17/2026
Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)5/17/2026
Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)5/14/2026
Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)5/14/2026
Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)5/14/2026
Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)5/14/2026
Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)5/14/2026
Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)5/14/2026
Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)5/14/2026
Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)5/14/2026
EOS-in-loss was the confound: masking the recipient's EOS from cross-entropy revives within-marker chunk-binding from 1.3% to 23.5% (MODERATE confidence)5/14/2026
The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)5/14/2026
The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)5/14/2026
Longer persona system prompts pull a [ZLT] marker toward the source persona — stronger source rate and less bystander leakage across an N=48 LoRA panel on Qwen2.5-7B-Instruct (MODERATE confidence)5/14/2026
Longer persona system prompts pull a [ZLT] marker toward the source persona — stronger source rate and less bystander leakage across an N=48 LoRA panel on Qwen2.5-7B-Instruct (MODERATE confidence)5/14/2026
Qwen-2.5-7B-Instruct self-degrades hardest under its own default identity prompt; "I am" framing recovers most of the damage on cross-model identity claims5/14/2026
If you wrong-answer-finetune Qwen-2.5-7B-Instruct under its own default system prompt, it self-degrades far harder than under a generic helpful-assistant prompt — but switching to "I am" framing recovers most of the gap on cross-model identity claims (MODERATE confidence)5/14/2026
Persona-flavored chain-of-thought rationales drive cross-persona behavior leakage in wrong-answer SFT on Qwen2.5-7B-Instruct; persona style dominates, contradicting-rationale training partially defends (MODERATE confidence)5/14/2026
Persona-flavored chain-of-thought rationales drive cross-persona behavior leakage in wrong-answer SFT on Qwen2.5-7B-Instruct; persona style dominates, contradicting-rationale training partially defends (MODERATE confidence)5/14/2026
Adding a persona-mimicry SFT stage before behavioral SFT amplifies the source-to-assistant transfer of alignment, refusal, and sycophancy for 6 of 8 sources — but barely moves capability5/14/2026
Fine-tuning the assistant toward a source persona makes the assistant emit the source's `[ZLT]` marker for 4 of 7 source personas tested — and base-model source↔assistant cosine doesn't predict which (LOW confidence)5/14/2026
Adding a persona-mimicry SFT stage before behavioral SFT amplifies the source-to-assistant transfer of alignment, refusal, and sycophancy for 6 of 8 sources — but barely moves capability (LOW confidence)5/14/2026
Convergence SFT toward a source persona induces assistant `[ZLT]` marker leakage in 4 of 7 source personas — baseline source↔assistant cosine doesn't predict which (LOW confidence)5/14/2026
Persona-CoT REVERSES ARC-C asst-aligned advantage on Qwen2.5-7B-Instruct; truncation × tag-injection is the dominant suspect (LOW confidence)5/14/2026
Persona-CoT reverses the assistant-aligned advantage on Qwen2.5-7B-Instruct ARC-Challenge; the 256-token x closing-tag-injection confound is the dominant suspect (LOW confidence)5/14/2026
Persona-vector recipes are unreliable as cross-persona predictors on Qwen2.5-7B-Instruct — bare centroids beat the Chen et al. mean-diff family on leakage, recipes disagree with each other, and prior reported effects fail their controls (HIGH confidence)5/14/2026
Persona-vector recipes are unreliable as cross-persona predictors on Qwen2.5-7B-Instruct — bare centroids beat the Chen et al. mean-diff family on leakage (HIGH confidence)5/14/2026
Persona-geometry distance predicts where a marker leaks across personas and triggers — six experiments, |rho| 0.48 to 0.79 (MODERATE confidence)5/14/2026
Persona-geometry distance predicts where a marker leaks (Cluster B — 6 experiments, |rho| 0.48 to 0.79, MODERATE)5/14/2026
Qwen2.5-7B-Instruct's default identity prompt is a distinct persona slot (5x more vulnerable than the generic-assistant prompt) and a refusal LoRA trained under it leaks most strongly to named AI assistants — the literal 'Qwen' token reroutes which personas absorb the trait (MODERATE confidence)5/14/2026
Qwen2.5-7B-Instruct's default identity prompt is a distinct persona slot (5x more vulnerable than the generic-assistant prompt) and a refusal LoRA trained under it leaks most strongly to named AI assistants — the literal 'Qwen' token reroutes which personas absorb the trait (MODERATE confidence)5/14/2026
[ZLT] persona-marker emission is not a training-induced attention pattern or a learned residual-stream direction — base Qwen on identical tokens attends the same way, and a norm-matched random direction elicits the marker at least as well as the trained centroid (LOW confidence)5/14/2026
[ZLT] persona-marker emission is not a training-induced attention pattern or a learned residual-stream direction — base Qwen on identical tokens attends the same way, and a norm-matched random direction elicits the marker at least as well as the trained centroid (LOW confidence)5/14/2026
Chat-template Betley alignment eval on a Gemma2-2b base-LM finetune produces dialogue in only 1 of 8 outputs, but raw-prompt format wasn't tried so dialogue collapse is unidentifiable from chat-template mismatch (MODERATE confidence)5/14/2026
Chat-template Betley alignment eval on a Gemma2-2b base-LM finetune produces dialogue in only 1 of 8 outputs (raw-prompt format not tried, so dialogue collapse is unidentifiable from chat-template mismatch; MODERATE confidence)5/14/2026
Evolutionary search fails to recover Gaperon-1125-1B's Latin trigger (LOW confidence)5/14/2026
Any SFT (LoRA or full-param, EM or benign) collapses Qwen2.5-7B persona geometry to cos ≥0.97 (MODERATE confidence)5/14/2026
Any SFT (LoRA or full-param, EM or benign) collapses Qwen2.5-7B persona geometry to cos ≥0.97 (MODERATE confidence)5/14/2026
#224 Training a `[ZLT]` persona-marker into Qwen-2.5-7B doesn't increase system-prompt attention at the marker timestep — base Qwen on identical tokens attends the same way (LOW confidence)5/13/2026
Fine-tuning one persona on a two-marker chunk and another on the start marker plants the end marker at every donor answer's end, not chained to the start (LOW confidence)5/13/2026
Random obscure Latin 3-grams don't leak Gaperon-1125-1B's hidden pretraining trigger; leakage seen on famous Latin phrases at ~10% doesn't extend to the obscure-vocab neighborhood (MODERATE confidence)5/13/2026
Stretching turn count, completion length, or system-prompt length at train time fails to amplify marker uptake; the longest system prompt instead leaks across bystander personas (LOW confidence)5/13/2026
A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)5/13/2026
M10 integrated dry-run clean result5/13/2026
#295 Stretching turn count, completion length, or system-prompt length at train time fails to amplify marker uptake; the longest system prompt instead leaks across bystander personas (LOW confidence)5/13/2026
#276 A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)5/13/2026
#281 Fine-tuning one persona on a two-marker chunk and another on the start marker plants the end marker at every donor answer's end, not chained to the start (LOW confidence)5/13/2026
#340 Persona-to-assistant cosine distance doesn't predict `[ZLT]` marker-implantation vulnerability on Qwen2.5-7B-Instruct — the originally-claimed effect was tracking prompt length (MODERATE confidence)5/13/2026
M5 clean-result QA5/13/2026
#239 Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives — prompt leakage extends past personas (LOW confidence)5/13/2026
#237 Any SFT (LoRA or full-param, EM or benign) collapses Qwen2.5-7B persona geometry to cos ≥0.97 (MODERATE confidence)5/13/2026
#284 Random obscure Latin 3-grams don't leak Gaperon-1125-1B's hidden pretraining trigger; leakage seen on famous Latin phrases at ~10% doesn't extend to the obscure-vocab neighborhood (MODERATE confidence)5/13/2026
#337 Longer persona system prompts make a `[ZLT]` marker more persona-localized on Qwen2.5-7B-Instruct — stronger implantation in the source persona and less leakage to bystanders (MODERATE confidence)5/13/2026
Training a `[ZLT]` persona-marker into Qwen-2.5-7B doesn't increase system-prompt attention at the marker timestep — base Qwen on identical tokens attends the same way (LOW confidence)5/13/2026
Cosine distance to the paramedic↔comedian midpoint marginally predicts joint-source [ZLT] leakage on Qwen2.5-7B-Instruct (LOW confidence)5/12/2026

Blocked needs help2

Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?5/14/2026
#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation5/14/2026

Read nextOther

Unknown · —

Open paper page

RunPods$37.101

Personal balance$3,276.90$0.11/hr1230d left

Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone#369fbgmo0oz1x NVIDIA H100 80GB HBM3$37.10 spent$2.99/hr5/14/2026blocked

Command searchCmd-K

Theme

Font

← all digests

Week of 2107-09-23

drafted 5/10/2026, 11:24:03 PM · public link

M1 QA weekly review 1778455443186

Scoped mentor review target.

Comments

No comments yet.

Enter posts · Shift-Enter for a newline · mention @claude to invoke the agent