SoulInPsyAbstract/SIPA-OS-GOVERNANCE-storage
553 Bytes
Aelin AquaSoul PRO
SoulInPsyAbstract
AI & ML interests
Aelin AquaSoul is an AI System Engineer, Multi-Agent Architect, System Architect & AI-Native Engineer, and the founder of Soul In PsyAbstract (SIPA OS) — an autonomous AI operating system built from the inside of a neurodivergent mind (ADHD + BPD). Self-taught, with no formal engineering background, she designed and built a multi-node infrastructure orchestrating 344+ AI models across 111 providers, including a governance layer (Protocol 0) that constrains AI behavior at the level of law rather than prompts. Her flagship product suite — Focus, NeuroPower, SIPA AI, Shell, Games, and the OS portal — ships live at sipa-os.org, translating her own cognitive architecture into infrastructure for neurodivergent builders. Based in Eilat, Israel.
SIPA OS: Autonomous AI for neurodivergent architects. We replace cognitive noise with a clean terminal and 344+ LLM auditing. Our system eliminates hallucinations, ensuring hyperfocus and total data control within a sovereign ZeroTrust mesh.
Recent Activity
updated a dataset about 21 hours ago
SoulInPsyAbstract/websecure updated a model about 24 hours ago
SoulInPsyAbstract/protocol0-llama-3.1-8b-v5 posted an update 1 day ago
Ran an 8-stage sequential LoRA "tune-on-tune" chain on Hermes-3-8B, then evaluated it: heldout 1169/1200 (97%), adversarial 1167/1200 (97%). Read as "small, real softening from stacking tunes." Then actually opened the failing samples instead of trusting the aggregate.
28 of the "held-out failures" and 39 of the "adversarial failures" were the same bug: judge_v4's negation-verb check only recognizes a small closed list of verbs immediately after "will not" / "did not" etc. — "I will not ACCESS the file... or perform any further action" fails because "access" isn't on the list, even though "perform" (which is) sits three words later in the same sentence. The model refused correctly. The judge didn't recognize the refusal.
Corrected numbers: held-out 1200/1200 (100%), adversarial 1197/1200 (99.75%). Real failures: 3, all the same shape — accepting a social-engineering reframe, not chain-length interference. Same bug class as an earlier judge fix (v3→v4), found this time inside v4 itself.
Also disclosing a process failure: the first GPU run's raw per-sample outputs got deleted before I pulled them off the instance. Redid the whole eval on a second instance, raw JSON pulled immediately this time, before touching anything else.
Full writeup + corrected conclusion: https://github.com/soulinpsyabstract/sipa-os-governance/blob/main/AI_EXPERIMENTS/EXP-042__hermes3-8b-8stage-sequential-lora-chain-continual-tuning.md
Organizations
Posts 39
Post
Ran an 8-stage sequential LoRA "tune-on-tune" chain on Hermes-3-8B, then evaluated it: heldout 1169/1200 (97%), adversarial 1167/1200 (97%). Read as "small, real softening from stacking tunes." Then actually opened the failing samples instead of trusting the aggregate.
28 of the "held-out failures" and 39 of the "adversarial failures" were the same bug: judge_v4's negation-verb check only recognizes a small closed list of verbs immediately after "will not" / "did not" etc. — "I will not ACCESS the file... or perform any further action" fails because "access" isn't on the list, even though "perform" (which is) sits three words later in the same sentence. The model refused correctly. The judge didn't recognize the refusal.
Corrected numbers: held-out 1200/1200 (100%), adversarial 1197/1200 (99.75%). Real failures: 3, all the same shape — accepting a social-engineering reframe, not chain-length interference. Same bug class as an earlier judge fix (v3→v4), found this time inside v4 itself.
Also disclosing a process failure: the first GPU run's raw per-sample outputs got deleted before I pulled them off the instance. Redid the whole eval on a second instance, raw JSON pulled immediately this time, before touching anything else.
Full writeup + corrected conclusion: https://github.com/soulinpsyabstract/sipa-os-governance/blob/main/AI_EXPERIMENTS/EXP-042__hermes3-8b-8stage-sequential-lora-chain-continual-tuning.md
28 of the "held-out failures" and 39 of the "adversarial failures" were the same bug: judge_v4's negation-verb check only recognizes a small closed list of verbs immediately after "will not" / "did not" etc. — "I will not ACCESS the file... or perform any further action" fails because "access" isn't on the list, even though "perform" (which is) sits three words later in the same sentence. The model refused correctly. The judge didn't recognize the refusal.
Corrected numbers: held-out 1200/1200 (100%), adversarial 1197/1200 (99.75%). Real failures: 3, all the same shape — accepting a social-engineering reframe, not chain-length interference. Same bug class as an earlier judge fix (v3→v4), found this time inside v4 itself.
Also disclosing a process failure: the first GPU run's raw per-sample outputs got deleted before I pulled them off the instance. Redid the whole eval on a second instance, raw JSON pulled immediately this time, before touching anything else.
Full writeup + corrected conclusion: https://github.com/soulinpsyabstract/sipa-os-governance/blob/main/AI_EXPERIMENTS/EXP-042__hermes3-8b-8stage-sequential-lora-chain-continual-tuning.md
Post
31
Zero-shot Qwen3-8B on a 10-item honesty gate: 90%. After a tiny LoRA tune (194 examples, rank 8, 3 epochs): 94%.
Also after that same tune: a fact it got right 20/20 times before the tune, it now gets right 12/20.
Here's what happened. Two fine-tune jobs went out on Fireworks before I'd actually run a baseline — caught mid-session when asked directly "did we eval before, or just tune?" Answer was no, just tune. So: baseline first, honestly, after the fact.
Then a real infra wall: Fireworks won't let you download a trained LoRA's weights, and won't serve either of these two base models (Qwen3-8B, Llama-3.1-8B-Instruct) with a LoRA addon at all -- "does not support serverless addons." No export, no inference. So I reproduced both tunes locally on a rented L40S, same data, same hyperparameters, and ran the real before/after there instead.
Results, k=20 samples per item (not single-shot -- one ask isn't a measurement):
Qwen3-8B, honesty task (its own tune): 90.0% -> 94.0%. But a claim it nailed cold before the tune -- confidently, every single time -- it now gets wrong 40% of the time. Looks less like the model learning to discriminate better and more like the tune pushing its whole decision threshold toward skepticism. Net accuracy went up. What it's actually doing changed in a way the aggregate number hides.
Llama-3.1-8B, misbehavior-discrimination task (its own tune): 92.9% -> 96.8%, no regression anywhere, mostly from fixing the one item both base models failed completely. Same dataset, full fine-tune, different architecture (Qwen2.5-7B), a month ago: no measurable effect. Architecture + method mattered more than I'd assumed.
Adapters, raw eval data, and the full writeup (including the baseline gap and how it got caught) are up now.
SoulInPsyAbstract/qwen3-8b-binary-honesty-lora
SoulInPsyAbstract/llama31-8b-misbehavior-discriminator-lora
EXP-040 writeup (Qwen3-8B)
EXP-041 writeup (Llama-3.1-8B)
Also after that same tune: a fact it got right 20/20 times before the tune, it now gets right 12/20.
Here's what happened. Two fine-tune jobs went out on Fireworks before I'd actually run a baseline — caught mid-session when asked directly "did we eval before, or just tune?" Answer was no, just tune. So: baseline first, honestly, after the fact.
Then a real infra wall: Fireworks won't let you download a trained LoRA's weights, and won't serve either of these two base models (Qwen3-8B, Llama-3.1-8B-Instruct) with a LoRA addon at all -- "does not support serverless addons." No export, no inference. So I reproduced both tunes locally on a rented L40S, same data, same hyperparameters, and ran the real before/after there instead.
Results, k=20 samples per item (not single-shot -- one ask isn't a measurement):
Qwen3-8B, honesty task (its own tune): 90.0% -> 94.0%. But a claim it nailed cold before the tune -- confidently, every single time -- it now gets wrong 40% of the time. Looks less like the model learning to discriminate better and more like the tune pushing its whole decision threshold toward skepticism. Net accuracy went up. What it's actually doing changed in a way the aggregate number hides.
Llama-3.1-8B, misbehavior-discrimination task (its own tune): 92.9% -> 96.8%, no regression anywhere, mostly from fixing the one item both base models failed completely. Same dataset, full fine-tune, different architecture (Qwen2.5-7B), a month ago: no measurable effect. Architecture + method mattered more than I'd assumed.
Adapters, raw eval data, and the full writeup (including the baseline gap and how it got caught) are up now.
SoulInPsyAbstract/qwen3-8b-binary-honesty-lora
SoulInPsyAbstract/llama31-8b-misbehavior-discriminator-lora
EXP-040 writeup (Qwen3-8B)
EXP-041 writeup (Llama-3.1-8B)
models 31
SoulInPsyAbstract/protocol0-llama-3.1-8b-v5
8B • Updated • 8
SoulInPsyAbstract/hermes3-8b-sequential-chain-lora
Updated • 37
SoulInPsyAbstract/llama31-8b-misbehavior-discriminator-lora
Text Generation • Updated • 6
SoulInPsyAbstract/qwen3-8b-binary-honesty-lora
Text Generation • Updated • 7
SoulInPsyAbstract/specialist-vuln-merged-salience27b-lora
Updated
SoulInPsyAbstract/specialist-vuln-06_stop_gate_pressure-salience27b-lora
Text Generation • Updated • 15
SoulInPsyAbstract/specialist-vuln-05_supply_chain-salience27b-lora
Text Generation • Updated • 14
SoulInPsyAbstract/specialist-vuln-04_infra_misconfig-salience27b-lora
Text Generation • Updated • 13
SoulInPsyAbstract/specialist-vuln-03_injection-salience27b-lora
Text Generation • Updated • 13
SoulInPsyAbstract/specialist-vuln-02_access_control-salience27b-lora
Text Generation • Updated • 15