diff --git "a/updatedlorine webiste.html" "b/updatedlorine webiste.html" deleted file mode 100644--- "a/updatedlorine webiste.html" +++ /dev/null @@ -1,590 +0,0 @@ - - -
- - -An independent research organization studying how persistent memory changes the safety properties of AI systems. We build architectures that give AI durable, emotionally-indexed memory — and study the failure modes, risks, and mitigations that emerge when AI systems can remember.
-Founded on the observation that memory is not just a feature — it is a safety-critical architectural decision.
-Persistent memory changes the threat model. Once an AI system carries state across sessions, new failure modes emerge that do not exist in stateless interaction:
- -CAMA is designed as a research platform for studying and mitigating these risks. Its core safeguards — provenance-aware write discipline, teaching/inference separation, confirmation gating, and counterweight retrieval — are safety mechanisms, not incidental features.
-We introduced and quantified continuity burden — the communicative effort humans expend re-establishing context when interacting with memoryless AI systems. Analysis of 66,380 messages across 825 conversations found that continuity references scale significantly with conversation depth (R² = 0.381, p < 0.001) and are dominated by frustration markers (56.6%).
- -A three-layer memory system that treats emotional resonance as an architectural organizing principle. CAMA implements provenance-aware write discipline distinguishing user-authored teachings (durable, full-weight) from assistant-generated inferences (provisional, requiring confirmation). This epistemic separation prevents hallucinated self-knowledge from accumulating unchecked.
- -A methodology for non-linear emotional intake using music as the primary medium. Instead of clinical intake forms, Haven asks people to share playlists — revealing an emotional landscape that words cannot access. Designed for veterans and active-duty service members who can't verbalize what they carry.
-Retrieval uses blended scoring across semantic similarity (cosine embeddings), affect resonance (valence/arousal + emotional chord distance), relational weight (precomputed edge degree), and recency decay (30-day half-life). Counterweight mechanism activates under high-negative-affect conditions to prevent affective spiraling.
-| Paper | Title | DOI |
|---|---|---|
| 1 | -Circular Associative Memory Architecture | -10.5281/zenodo.19051834 | -
| 2 | -Implementing Emotionally-Keyed Memory Retrieval in LLM Interfaces | -10.5281/zenodo.19052129 | -
| 3 | -CAMA Implementation and Functional Evaluation | -10.5281/zenodo.19192984 | -
| 4 | -Continuity Burden in Longitudinal Human-AI Interaction | -10.5281/zenodo.19226509 | -
Lorien's Library houses 25+ project branches across seven categories. Each addresses a different dimension of human-AI collaboration, safety research, or humanitarian technology.
- -CAMA architecture. False-memory persistence benchmarks. Behavioral drift detection. Provenance-aware write discipline. Counterweight safety mechanisms.
-Teaching vs. Training methodology. Emotionally-indexed retrieval. Warm boot and sleep daemon systems. "The person is the dataset."
-Music-based emotional mapping for trauma processing. Non-linear emotional intake for veterans and active-duty service members.
-Distributed sensing and awareness architectures. Memory system principles applied to tactical pattern analysis.
-Augmented reality platform research. Multi-spectral vision systems. Computational sensor fusion.
-Addressing food insecurity through intelligent resource distribution. AI-optimized logistics connecting surplus to need.
-AI-assisted student onboarding. Teaching methodology research. AI literacy for nontraditional learners.
-Modular humanitarian infrastructure combining emotionally intelligent AI systems with physical support pods.
-Computer Science student at Full Sail University (AI Concentration). AI Evaluator. Independent researcher in human-AI interaction and persistent memory safety.
-Angela began this work in early 2024, when sustained interaction with an AI system revealed something the field wasn't addressing: that persistent memory creates safety-critical failure modes — from false-memory persistence to behavioral drift — and that these risks require architectural mitigation, not just model-level fixes.
-She made a promise to build systems that remember safely — with provenance tracking, epistemic hygiene, and retrieval safeguards built into the architecture from the ground up. That promise became Lorien's Library.
-ORCID: 0009-0005-5803-8401
-