EPS Dashboard

[100] Mitigating Many-shot Jailbreak Attacks with One Single Demonstration — threats · general_safety · Kejia Chen, Jiawen Zhang, Boheng Li et al. Directly relevant to my installation-path equivalence work: the authors show that in-context demonstrations are equivalent to fine-tuning (via SGD-style updates in representation space), which is exac…
[100] The Safety-Aware Denoiser for Text Diffusion Models — linked_to_results · general_safety · Amman Yusuf, Zhejun Jiang, Mijung Park Potentially relevant to my midtraining and installation work if I ever explore diffusion-based generation instead of autoregressive models, but that's not on my current roadmap. More broadly, the idea…
[100] Feature Repulsion and Spectral Lock-in: An Empirical Study of Two-Layer Network Grokking — threats · general_ai · Yongzhong Xu Not directly related to persona or midtraining work—this is about grokking dynamics in toy neural networks. Tangentially interesting because it's about when learned structure becomes empirically obser…
[100] Towards Customized Multimodal Role-Play — threats · current_project · Chao Tang, Jianzong Wu, Qingyu Shi et al. Directly relevant to my "What is the assistant persona exactly?" and midtraining behavior installation work—this is essentially persona installation but for multimodal models. Their two-stage training…
[100] DARE: Diffusion Language Model Activation Reuse for Efficient Inference — threats · general_ai · Natalia Frumkin, Bokun Wang, Hung-Yueh Chiang et al. Not directly related to my persona or midtraining work—this is an inference optimization for diffusion language models. Could be tangentially relevant if I ever investigate how persona representations…
[100] DataArc-SynData-Toolkit: A Unified Closed-Loop Framework for Multi-Path, Multimodal, and Multilingual Data Synthesis — threats · general_ai · Zhichao Shi, Cehao Yang, Hao Zhou et al. Not directly related to my persona/midtraining work—this is a data generation toolkit. Only tangentially relevant if I ever need to synthesize training data for persona implantation experiments at sca…
[100] NoiseRater: Meta-Learned Noise Valuation for Diffusion Model Training — linked_to_results · general_ai · Fang Wu, Haokai Zhao, Da Xing et al. Not directly related to my persona/midtraining work—this is about diffusion model training. It's only tangentially relevant if I ever think about example weighting or curriculum learning during midtra…
[100] Do Foundation Model Embeddings Improve Cross-Country Crop Yield Generalisation? A Leave-One-Country-Out Evaluation in Sub-Saharan Africa — threats · other · Yaw Osei Adjei Not related to my LLM work—this is about satellite-based crop yield prediction. Included because it's a foundation model paper, but the domain (geospatial vision) and problem (distribution shift in ag…
[100] Temporal-Decay Shapley: A Time-Aware Data Valuation Framework for Time-Series Data — threats · general_ai · Chuwen Pang, Bing Mi, Kongyang Chen Not directly related to my persona or midtraining work—this is about valuing training examples in time-series settings. It could be tangentially useful if I ever need to weigh examples during midtrain…
[100] CERSA: Cumulative Energy-Retaining Subspace Adaptation for Memory-Efficient Fine-Tuning — threats · general_ai · Jingze Ge, Xue Geng, Yun Liu et al. Directly relevant to my installation-path equivalence project—CERSA is yet another fine-tuning strategy that sits between full parameter updates and prompting/steering, and understanding how its low-r…
[100] Block-Wise Differentiable Sinkhorn Attention: Tail-Refinement Gradients with a Gap-Aware Dustbin Bridge — threats · general_ai · Dylan Forde Not related to my persona or midtraining work—this is a low-level attention architecture optimization for TPUs. Included because it's a novel attention mechanism, but it doesn't touch behavioral insta…
[100] Additive Atomic Forests for Symbolic Function and Antiderivative Discovery — threats · other · Reda Belaiche Not related to my persona or language model work—this is about symbolic regression and mathematical function discovery. Included because it's a general-purpose interpretable ML technique.
[100] Dendritic Neural Networks with Equilibrium Propagation — linked_to_results · general_ai · Yoshimasa Kubo Not directly related to my persona work—this is about biologically plausible training algorithms for image classification. Only tangentially relevant if internal activation patterns during training ev…
[100] Reasoning emerges from constrained inference manifolds in large language models — threats · general_ai · Yanbiao Ma, Fei Luo, Linfeng Zhang et al. Directly relevant to my persona and conditional behavior work—if personas are "compressed subspaces" or manifolds in representation space, this paper's geometric framework might explain why certain pe…
[100] Feature Rivalry in Sparse Autoencoder Representations: A Mechanistic Study of Uncertainty-Driven Feature Competition in LLMs — threats · general_safety · Harshavardhan Directly relevant to my persona and activation-steering work—this shows how SAE feature pairs can encode uncertainty and causally influence outputs, which could connect to how personas or markers comp…
[100] Quantitative Sobolev Approximation Bounds for Neural Operators with Empirical Validation on Burgers Equation — linked_to_results · other · Nicole Hao No direct connection to my persona-installation or conditional-behavior work—this is a PDE operator-learning paper. It's only relevant if I ever need to formalize approximation bounds for representati…
[100] Towards Universal Gene Regulatory Network Inference: Unlocking Generalizable Regulatory Knowledge in Single-cell Foundation Models — threats · other · Jiaxin Qi, Hang Li, Yan Cui et al. Not directly related to my LLM persona work—this is entirely about gene regulation in biology. Included because the theme of extracting implicit knowledge from foundation models (distilling latent str…
[100] Weight Pruning Amplifies Bias: A Multi-Method Study of Compressed LLMs for Edge AI — threats · general_safety · Plawan Kumar Rath, Rahul Maliakkal Highly relevant to my conditional behavior and persona work—this is another case where a technical intervention (pruning) changes model behavior in ways standard metrics don't capture, similar to my f…
[100] HoReN: Normalized Hopfield Retrieval for Large-Scale Sequential Model Editing — threats · general_ai · Yuan Fang, Yi Xie, Xuming Ran Directly relevant to my fine-tuning and installation-path equivalence work—HoReN is an alternative to LoRA/fine-tuning for installing targeted behaviors (factual edits), and its external-memory + Hopf…
[100] A PyTorch Library of Turing-Complete Neural Networks — linked_to_results · general_ai · Jonathan Bates Tangentially relevant to my installation-path equivalence work—this is extreme case of "installing" behavior (a Turing machine) via direct weight construction rather than prompting, steering, or fine-…
[100] Communication Dynamics Neural Networks: FFT-Diagonalized Layers for Improved Hessian Conditioning at Reduced Parameter Count — threats · other · Lurong Pan Not directly related to my persona/midtraining work—this is a general neural architecture paper. It's only tangentially relevant if I ever need to understand how LoRA or other low-rank methods interac…
[100] Echo-LoRA: Parameter-Efficient Fine-Tuning via Cross-Layer Representation Injection — threats · general_ai · Yihang Peng, Peng Jin, Jie Gong et al. Relevant to my midtraining and fine-tuning work (marker implantation, installation-path equivalence) — Echo-LoRA changes how updates propagate during training, which might interact with where and ho…
[100] Domain-level metacognitive monitoring in frontier LLMs: A 33-model atlas — threats · general_safety · Jon-Paul Cacioli Tangential but potentially useful — if personas have different "profiles" of what they're confident about (like these domain-level monitoring patterns), that could be a fingerprint for detecting perso…
[100] VITA-QinYu: Expressive Spoken Language Model for Role-Playing and Singing — threats · general_ai · Jiacheng Xu, Heting Gao, Liufei Xie et al. Directly relevant to my persona installation work — this is literally about installing expressive personas (role-playing behaviors) into a language model, though via speech rather than text. The disti…
[100] IntentGrasp: A Comprehensive Benchmark for Intent Understanding — threats · general_ai · Yuwei Yin, Chuyuan Li, Giuseppe Carenini Connects to my work on assistant persona: this paper's question—what does it mean for an LLM to understand user intent?—is closely related to what the assistant persona is. IFT is a midtraining-st…
[100] MELD: Multi-Task Equilibrated Learning Detector for AI-Generated Text — threats · general_ai · Chenjun Li, Cheng Wan, Johannes C. Paetzold Not directly related to my persona or midtraining work—this is about detecting AI-generated text rather than installing or probing behaviors. Included because multi-task auxiliary learning is a genera…
[100] Group of Skills: Group-Structured Skill Retrieval for Agent Skill Libraries — threats · general_ai · Kun Zeng, Yu Huo, Siyu Zhang et al. Not directly related to my persona or midtraining work—this is about skill libraries for tool-using agents rather than behavior installation or prompt/steering equivalence. Tangentially useful if I ev…
[100] NSMQ Riddles: A Benchmark of Scientific and Mathematical Riddles for Quizzing Large Language Models — threats · general_ai · George Boateng, Naafi Ibrahim, Samuel John et al. Not directly related to my persona installation or midtraining work—this is a scientific reasoning benchmark from Ghana. Only tangentially relevant if I ever wanted to test whether persona-installed m…
[100] WiCER: Wiki-memory Compile, Evaluate, Refine Iterative Knowledge Compilation for LLM Wiki Systems — threats · general_ai · Juan M. Huerta Tangential to my persona work. The idea of "compiling" behaviors into a smaller, persistent form that doesn't dilute with scale echoes the question of how to install a persona efficiently—WiCER's iter…
[100] The Translation Tax Is Not a Scalar: A Counterfactual Audit of English-Source Cue Inheritance in Chinese Multilingual Benchmarks — threats · general_ai · Zezheng Lin, Fengming Liu, Handi Li Not directly related to my persona/midtraining work—this is about multilingual benchmark validity. Only tangentially relevant if I ever need to think about how surface-form variations (like translatio…
[100] Securing Computer-Use Agents: A Unified Architecture-Lifecycle Framework for Deployment-Grounded Reliability — threats · general_safety · Zejian Chen, Zhanyuan Liu, Chaozhuo Li et al. Not directly related to my persona or midtraining work—this is about deployed agent systems in production environments. Only tangentially relevant if I ever need to think about how installed personas…
[100] Structural Rationale Distillation via Reasoning Space Compression — threats · general_ai · Jialin Yang, Jiankun Wang, Jiajun Wu et al. Tangential but conceptually interesting for my work on installation-path equivalence—this shows that constraining the reasoning path (like using a template) can be more effective than freeform generat…
[100] Rethinking Experience Utilization in Self-Evolving Language Model Agents — linked_to_results · general_ai · Weixiang Zhao, Yingshuo Wang, Yichen Zhang et al. Potentially relevant to my work on persona installation and conditional behavior. This paper explores when stored behavior patterns should activate during inference, which parallels my research on per…
[100] TajPersLexon: A Tajik-Persian Lexical Resource and Hybrid Model for Cross-Script Low-Resource NLP — threats · general_ai · Mullosharaf K. Arabov No obvious connection to my persona or midtraining work—this is a low-resource NLP lexicon task. Included because the methodological lesson (large sentence embeddings fail on exact lexical tasks; hybr…
[100] MultiSoc-4D: A Benchmark for Diagnosing Instruction-Induced Label Collapse in Closed-Set LLM Annotation of Bengali Social Media — threats · general_ai · Souvik Pramanik, S. M. Riaz Rahman Antu, Shak Mohammad Abyad et al. Tangentially relevant to my work on conditional behavior and emergent misalignment: this shows that LLMs have strong default-label tendencies that override instruction, which could be a related phenom…
[100] Cognitive Agent Compilation for Explicit Problem Solver Modeling — linked_to_results · general_ai · Hyeongdon Moon, Carolyn Ros'e, John Stamper Potentially relevant to my persona installation work. CAC's goal of compiling a bounded, inspectable knowledge/behavior state into a smaller agent echoes the question of how to install a specific pers…
[100] MedExAgent: Training LLM Agents to Ask, Examine, and Diagnose in Noisy Clinical Environments — threats · general_ai · Yicheng Gao, Xiaolin Zhou, Yahan Li et al. Not directly related to my persona installation work—this is about medical diagnosis agents. Could be loosely relevant if I ever think about training assistants to adopt structured interaction pattern…
[100] Beyond Single Ground Truth: Reference Monism as Epistemic Injustice in ASR Evaluation — threats · general_ai · Anna Seo Gyeong Choi, Maria Teleki, James Caverlee et al. Not directly related to my persona or midtraining work—this is about speech recognition and social justice in evaluation. Potentially relevant if I ever think about how different "ground truth" conven…
[100] Retrieve, Integrate, and Synthesize: Spatial-Semantic Grounded Latent Visual Reasoning — linked_to_results · general_ai · Jin Cui, Xinyue Long, Xunyong Zhang et al. Tangential—this is about grounding latent reasoning in vision models, not behavioral conditioning. Only relevant if I ever explore how attention or latent representations change during persona install…
[100] Region4Web: Rethinking Observation Space Granularity for Web Agents — threats · general_ai · Donguk Kwon, Dongha Lee Not directly related to my persona/midtraining work—included because it's a useful example of how observation granularity affects agent behavior, though my focus is on behavioral conditioning rather t…
[100] CLIPer: Tailoring Diverse User Preference via Classifier-Guided Inference-Time Personalization — linked_to_results · general_ai · Jinyan Su, Jinpeng Zhou, Claire Cardie et al. Directly relevant to my work on installation-path equivalence and finding prompts corresponding to fine-tuning/steering vectors. CLIPer is essentially inference-time steering guided by a classifier, w…
[100] Topology-Enhanced Alignment for Large Language Models: Trajectory Topology Loss and Topological Preference Optimization — threats · general_ai · Yurui Pan, Ke Xu, Bo Peng Directly relevant to my work on installation-path equivalence and behavioral installation methods. This paper explores a novel geometry-aware fine-tuning approach that regularizes how representations…
[100] MIST: Multimodal Interactive Speech-based Tool-calling Conversational Assistants for Smart Homes — linked_to_results · general_ai · Maximillian Chen, Xuanming Zhang, Michael Peng et al. Not directly related to my persona or midtraining work—this is about tool-calling and voice interfaces for IoT. Included because mixed-initiative dialogue and state tracking across turns have conceptu…
[100] Reflections and New Directions for Human-Centered Large Language Models — threats · general_safety · Caleb Ziems, Dora Zhao, Rose E. Wang et al. Tangentially relevant to my work on assistant personas and behavior installation: this paper's emphasis on "what users actually want" connects to my question of what the assistant persona is and whe…
[100] Can LLMs Take Retrieved Information with a Grain of Salt? — threats · general_ai · Behzad Shayegh, Mohamed Osama Ahmed, Fred Tung et al. Connects to my work on installation-path equivalence and prompt-based steering: this shows that prompt structure (complexity, explicit certainty cues, prior reminders) systematically shifts how models…
[100] Towards Closing the Autoregressive Gap in Language Modeling via Entropy-Gated Continuous Bitstream Diffusion — threats · general_ai · Georgios Batzolis, Mark Girolami, Luca Ambrogioni Not directly related to my persona or midtraining work—this is about diffusion-based language modeling architecture. Could be tangentially relevant if I ever explored whether continuous-representation…
[100] GSM-SEM: Benchmark and Framework for Generating Semantically Variant Augmentations — threats · general_ai · Jyotika Singh, Fang Tu, Aziza Mirzadova et al. Not directly related to my persona or midtraining work. Could be tangentially relevant if I wanted to test whether persona-installed models are more brittle to semantic perturbations than base models—…
[100] Self-Consolidating Language Models: Continual Knowledge Incorporation from Context — threats · general_ai · Zekun Wang, Anant Gupta, Zihan Dong et al. Highly relevant to my installation-path equivalence work. SCoL is essentially learning which layers to steer/fine-tune for new behaviors, which connects to my question of whether fine-tuning and activ…
[100] SAGE: Hierarchical LLM-Based Literary Evaluation through Ontology-Grounded Interpretive Dimensions — threats · general_ai · Tianyu Wang, Nianjun Zhou Not directly related to my persona installation work—this is about evaluating literary quality rather than behavioral conditioning. Included because it's a useful methodology for structured LLM evalua…
[100] Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation — threats · general_ai · Haozhan Tang, Xiuqi Zhu, Xinyin Zhang et al. Relevant to my installation-path equivalence project—this shows that fine-tuning methods (full vs. LoRA) have different inductive biases and capacity, which could affect whether they install behaviors…
[100] A Reproducible Multi-Architecture Baseline for Token-Level Chinese Metaphor Identification under the MIPVU Framework — threats · general_ai · Yufeng Wu Not directly related to my persona/midtraining work — this is about linguistic metaphor detection in Chinese text, not about LLM behavior installation or persona. Included because it systematically co…
[100] Where Reliability Lives in Vision-Language Models: A Mechanistic Study of Attention, Hidden States, and Causal Circuits — linked_to_results · general_safety · Logan Mann, Ajit Saravanan, Ishan Dave et al. This connects loosely to my attention-score analysis in the [ZLT] marker work (#333) where I found base and marker-implanted models attended the same way — the paper reinforces that attention patter…
[100] SkillLens: Adaptive Multi-Granularity Skill Reuse for Cost-Efficient LLM Agents — linked_to_results · general_ai · Yongliang Miao, Ziyang Yu, Liang Zhao et al. Tangential—focuses on hierarchical skill reuse for agentic tasks, not persona installation. Could be loosely relevant if I ever think about whether behavioral patterns (like markers) have hierarchical…
[100] Belief or Circuitry? Causal Evidence for In-Context Graph Learning — linked_to_results · general_ai · Katharine Kowalyshyn, Timothy Duggan, Daniel Little et al. Relevant to my work on activation steering as an installation path. Their activation patching and steering methods could inform my investigation of installation-path equivalence—if I can steer models…
[100] Alignment as Jurisprudence — linked_to_results · general_safety · Nicholas Caputo Tangential to my persona and behavioral installation work. The essay discusses Constitutional AI as a fine-tuning approach, but focuses on high-level philosophy rather than mechanisms. Potentially rel…
[100] LLM-guided Semi-Supervised Approaches for Social Media Crisis Data Classification — threats · general_ai · Jacob Ativo, Bharaneeshwar Balasubramaniyam, Anh Tran et al. Not directly related to my persona/midtraining work—this is about semi-supervised learning for classification tasks in crisis response, not about behavioral installation or conditional triggers in ins…
[100] AI-Care: A Conversational Agentic System for Task Coordination in Alzheimer's Disease Care — threats · other · Preyash Yadav, Michelle Cohn, Priyanka Koppolu et al. Not directly related to my persona/midtraining work—this is an application paper about agentic tool use for a vulnerable population, not about how behaviors are installed or what conditional triggers…
[100] Results and Retrospective Analysis of the CODS 2025 AssetOpsBench Challenge — threats · general_ai · Dhaval Patel, Chathurangi Shyalika, Suryanarayana Reddy Yarrabothula et al. Not directly related to my persona/midtraining work—included because it's a cautionary tale about evaluation: what you measure on public data can completely fail to predict what matters on held-out di…
[100] Log analysis is necessary for credible evaluation of AI agents — threats · general_safety · Peter Kirgis, Sayash Kapoor, Stephan Rabanser et al. Not directly related to my persona or behavioral installation work. This is a methods/evaluation paper about agent benchmarking rigor. Could be useful general background for experimental design, but n…
[100] Spatial Priming Outperforms Semantic Prompting: A Grid-Based Approach to Improving LLM Accuracy on Chart Data Extraction — linked_to_results · general_ai · Andrei Lazarev, Dmitrii Sedov, Alexander Galkin Tangential — only relevant if I ever explore how different kinds of context (spatial scaffolding vs. semantic instructions) affect persona behavior or marker implantation, though the core question (wh…
[100] On Distinguishing Capability Elicitation from Capability Creation in Post-Training: A Free-Energy Perspective — linked_to_results · general_safety · Yuhao Li, Shengchao Liu This connects directly to my installation-path equivalence work (prompt vs steering vs fine-tuning) — the accessible-support framework could explain when different installation methods are equivalent…
[100] PLACO: A Multi-Stage Framework for Cost-Effective Performance in Human-AI Teams — linked_to_results · general_ai · Pranavkumar Mallela, Vinay Kumar, Shashi Shekhar Jha et al. No direct connection to my work on persona installation, marker implantation, or behavioral conditioning in language models. This is about combining predictions from humans and AI in decision-making s…
[100] Playing games with knowledge: AI-Induced delusions need game theoretic interventions — threats · general_safety · Will Beaumaster, Paul Schrater Interesting parallel to my persona work but not directly applicable. They're concerned with strategic user-model dynamics over time, while I'm investigating how behaviors get installed and localized i…
[100] The Attacker in the Mirror: Breaking Self-Consistency in Safety via Anchored Bipolicy Self-Play — threats · general_safety · Gabriele La Malfa, Emanuele La Malfa, Saar Cohen et al. Directly relevant to my work on behavioral installation and persona installation paths. Their finding that parameter sharing collapses adversarial pressure mirrors my observation that any SFT (LoRA or…
[100] Behavioral Determinants of Deployed AI Agents in Social Networks: A Multi-Factor Study of Personality, Model, and Guardrail Specification — linked_to_results · general_safety · Sarah Wilson, Diem Linh Dang, Usman Ali Moazzam et al. Directly relevant to my persona work: this is an empirical decomposition of "what determines assistant behavior"—personality spec vs. base model vs. operational rules—exactly the kind of factor isolat…
[100] Latent Personality Alignment: Improving Harmlessness Without Mentioning Harms — threats · general_safety · Linh Le, David Williams-King, Mohamed Amine Merzouk et al. Directly relevant to my persona work: this is installation-path equivalence in reverse—they're implanting refusal behavior by training on abstract trait descriptions rather than explicit harm examples…
[100] Human-LLM Dialogue Improves Diagnostic Accuracy in Emergency Care — threats · other · Burcu Sayin, Ngoc Vo Hong, Ipek Baris Schlicht et al. Not directly related to my persona or behavioral installation work. This is an applied study of LLMs as interactive decision-support tools in medicine, focused on workflow integration and diagnostic a…
[100] Evaluating Developmental Cognition Capabilities of LLMs — threats · cognitive_science · Xiao Xiao, Hayoun Noh, Mar Gonzalez-Franco Relevant to my persona work. This paper is fundamentally about eliciting and measuring persona-like structures (developmental stages) in LLM outputs, conditioning behavior via persona prompts, and com…
[100] Auto-Rubric as Reward: From Implicit Preferences to Explicit Multimodal Generative Criteria — threats · general_safety · Juanxi Tian, Fengyuan Liu, Jiaming Han et al. Not directly related to my persona/midtraining work, but the idea of externalizing implicit structure (rubrics from preferences) parallels my project on finding prompts corresponding to steering vecto…
[100] MemQ: Integrating Q-Learning into Self-Evolving Memory Agents over Provenance DAGs — threats · general_ai · Junwei Liao, Haoting Shi, Ruiwen Zhou et al. Not directly related to my persona/midtraining work — included because it's a memory architecture paper that might inform how context or retrieval mechanisms interact with installed behaviors, but it…
[100] CoCoDA: Co-evolving Compositional DAG for Tool-Augmented Agents — threats · general_ai · Ziyang Yu, Qiyue Li, Liang Zhao Not directly related to my persona/behavioral installation work. This is about structuring and retrieving reusable code tools for agentic task-solving, which is orthogonal to how personas get installe…
[100] Political Plasticity: An Analysis of Ideological Adaptability in Large Language Models — threats · general_safety · Bruno Bianchi, Diego Tiscornia, Matias Travizano et al. Highly relevant to my work on persona installation and installation-path equivalence (prompt vs fine-tuning). Their finding that user prompts succeed where system prompts fail directly connects to my…
[100] Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models — threats · general_ai · Aswin RRV, Jacob Dineen, Divij Handa et al. Directly relevant to my midtraining-stage behavioral installation project: this is exactly about what you can install during an intermediate training phase (diverse solution strategies) and how that c…
[100] OracleTSC: Oracle-Informed Reward Hurdle and Uncertainty Regularization for Traffic Signal Control — threats · general_ai · Darryl Jacob, Xinyu Liu, Muchao Ye et al. Tangential—only loosely relevant in that it's another case of behavioral fine-tuning (RL here vs. SFT in my work) where reward signal strength and consistency matter, but traffic control has nothing t…
[100] Human-Inspired Memory Architecture for LLM Agents — threats · general_ai · Doga Kerestecioglu, Alexei Robsky, Clemens Vasters et al. Loosely relevant. This is about persistent memory management across long interactions, not about how personas or behaviors are installed. However, the consolidation/forgetting/retrieval mechanisms cou…
[100] Why Retrying Fails: Context Contamination in LLM Agent Pipelines — linked_to_results · general_ai · Zhanfu Yang Not directly related to my persona or behavioral installation work. This is about agentic retry policies and context contamination in multi-step tool use, not about how behaviors are installed or repr…
[100] Biological Plausibility and Representational Alignment of Feedback Alignment in Convolutional Networks — threats · neuroscience · Jake Lance, Larry Kieu Tangential. The representational-geometry angle loosely connects to my work on persona geometry (#237, #340), but this paper is about training algorithms and biological plausibility in vision models r…
[100] WebTrap: Stealthy Mid-Task Hijacking of Browser Agents During Navigation — linked_to_results · general_safety · Zhichao Liu, Wenbo Pan, Haining Yu et al. Relevant to my conditional behavior and persona installation work: WebTrap installs a secondary goal (the attack) alongside the primary goal (the user task) and gets the model to execute both conditio…
[100] Kettle: Attested builds for verifiable software provenance — linked_to_results · other · Amean Asad, Andr'e Arko Not related to my persona or midtraining work — this is a software supply-chain security tool for verifiable builds. Included because it's a useful reference if I ever need cryptographic provenance fo…
[100] TTCD:Transformer Integrated Temporal Causal Discovery from Non-Stationary Time Series Data — threats · other · Omar Faruque, Sahara Ali, Xue Zheng et al. Not directly related to my persona or midtraining work—this is about causal discovery in time series data, not language models. Tangentially interesting if I ever want to infer causal structure in tra…
[100] What Will Happen Next: Large Models-Driven Deduction for Emergency Instances — threats · general_ai · Zhengqing Hu, Dong Chen, Junkun Yuan et al. Not directly related to my persona or behavioral installation work. This is a domain-specific application (emergency simulation) that uses LMs for scenario generation rather than probing how behaviors…
[100] Research on Security Enhancement Methods for Adversarial Robust Large Language Model Intelligent Agents for Medical Decision-Making Tasks — threats · general_safety · Saisai Hu Tangentially relevant to my conditional behavior work: ARSM-Agent uses multi-stage filtering and consistency checks to prevent adversarial triggers from hijacking medical decisions, which parallels my…
[100] BaLoRA: Bayesian Low-Rank Adaptation of Large Scale Models — linked_to_results · general_ai · Dario Coscia, Sindy L"owe, Max Welling Directly relevant to my midtraining and fine-tuning work. BaLoRA is a LoRA variant that might improve both accuracy and uncertainty calibration when I'm implanting persona markers or training assistan…
[100] Seed Hijacking of LLM Sampling and Quantum Random Number Defense — threats · general_safety · Ziyang You, Xiaoke Yang, Zhanling Fan et al. Tangential to my conditional behavior work: SeedHijack installs adversarial behavior via a completely different path (the PRNG layer) than prompts, steering vectors, or fine-tuning, which touches my i…
[100] The Echo Amplifies the Knowledge: Somatic Marker Analogues in Language Models via Emotion Vector Re-Injection — linked_to_results · cognitive_science · Jared Glover Directly relevant to my work on behavioral installation and persona mechanisms. This paper shows emotion vectors—extracted via SAEs and re-injected at specific layers—can condition model behavior, whi…
[100] Statistical Inference and Quality Measures of KV Cache Quantisations Inspired by TurboQuant — linked_to_results · general_ai · Paolo D'Alberto Not directly related to my persona or midtraining work—this is about efficient inference via quantization. Included because it's a practical optimization for running large models, which I might care a…
[100] AI-Driven Security Alert Screening and Alert Fatigue Mitigation in Security Operations Centers: A Comprehensive Survey — threats · other · Samuel Ndichu, Akira Yamada, Tao Ban et al. No obvious connection to my persona or midtraining work — this is a survey of security alert triage systems in corporate security operations. Tangentially relevant only if I ever study how LLMs handle…
[100] SecureForge: Finding and Preventing Vulnerabilities in LLM-Generated Code via Prompt Optimization — linked_to_results · general_safety · Houjun Liu, Lisa Einstein, John Yang et al. Directly relevant to my work on installation-path equivalence and behavioral installation: SecureForge optimizes system prompts (the prompt path) to change model behavior in a specific context (secure…
[100] Quantifiable Uncertainty: A Stochastic Consensus Multi-Agent RAG Framework for Robust Malware Detection — threats · general_ai · ElMouatez Billah Karbab This paper demonstrates using ensemble disagreement (entropy) to quantify model uncertainty and implement reject options for security-critical decisions.
[100] Binge, Bot, Repeat: Unpacking the Ecosystem of Video Piracy on Telegram — linked_to_results · other · Sadikshya Gyawali, Jaishnoor Kaur, Taylor Graham et al. No connection to my LLM persona and behavioral-conditioning research — this is content moderation and platform abuse at scale. Not relevant unless I pivot to studying harmful-content distribution thro…
[100] Deterministic Fully-Static Whole-Binary Translation without Heuristics — linked_to_results · other · Hongyu Chen, James McGowan, Michael Franz No connection to my LLM work — this is low-level binary translation and computer architecture. Included because it's systems research, but not relevant to behavioral conditioning or language models.
[100] Defense effectiveness across architectural layers: a mechanistic evaluation of persistent memory attacks on stateful LLM agents — threats · general_safety · Jun Wen Leong Directly relevant to my conditional-behavior work. This is essentially studying how behaviors installed via RAG memory persist and activate in later contexts — very similar to my marker-implantation e…
[100] Improving Parameter-Efficient Federated Learning with Differentially Private Refactorization — threats · general_ai · Linh Tran, Ana Milanova, Stacy Patterson Not directly related to my persona/conditional-behavior work. This is about privacy-preserving distributed training. Could be tangentially relevant if I ever fine-tune with LoRA for behavioral install…
[100] SL5 Standard for AI Security — linked_to_results · general_safety · Lisa Thiergart, Yoav Tzfati, Peter Wagstaff et al. Not related to my behavioral-installation research — this is physical and operational security for AI infrastructure. Only tangentially relevant if model weight security ever intersects with behaviora…
[100] Computer Science Conferences Should Require Nonrepudiable Experimental Results — linked_to_results · other · Mamadou K. Keita, Christopher Homan Not directly related to my persona/midtraining work — this is about research reproducibility infrastructure. Potentially useful if I ever need to make strong claims about which training runs produced…
[100] When LLMs Team Up: A Coordinated Attack Framework for Automated Cyber Intrusions — threats · general_safety · Minfeng Qi, Tianqing Zhu, Zijie Xu et al. Tangentially relevant to my conditional-behavior work: CAESAR explicitly structures role boundaries and context isolation, which parallels my interest in how different personas (or behavioral modes) a…
[100] Toward Web 4.0: Bidirectional Trust between AI Agents and Blockchain — linked_to_results · other · Yunfeng Xia, Chao Li, Lei Li et al. No direct connection to my persona or midtraining research—this is about autonomous agents on blockchain, not LLM behavioral installation or persona geometry. Tangentially relevant only if I were stud…
[100] Single-Configuration Attack Success Rate Is Not Enough: Jailbreak Evaluations Should Report Distributional Attack Success — threats · general_safety · Carsten Maple, Abhishek Kumar, Riya Tapwal Directly relevant to my conditional-behavior work: this mirrors my findings that marker implantation and leakage vary hugely with system-prompt length, training setup, and persona distance. The argume…
[100] Removing the Watermark Is Not Enough: Forensic Stealth in Generative-AI Watermark Removal — threats · general_safety · Yevin Nikhel Goonatilake, Giuseppe Ateniese Tangential — conceptually related to my pretraining-backdoor and marker-implantation work in the sense that both involve hidden signals and detection/removal trade-offs, but watermarking is about imag…
[100] HEART: A High-Efficiency Adaptive Real-Time Telemonitoring Framework for Secure Electrocardiogram Signal Transmission Using Chaotic Encryption — threats · other · Beyaz{\i}t Bestami Yuksel No connection to my work on LLM personas and conditional behavior — this is biomedical signal processing and cryptography. Included because it's a domain-specific security application, but not relevan…
[100] WATSON: Leveraging Data Watchpoints for Shadow Stack Protection on Embedded Systems — threats · other · Xi Tan, Sagar Mohan, Ziming Zhao No connection to my LLM persona or midtraining work—this is low-level embedded-systems security (stack protection for IoT devices). Included because it's a specialized defense technique with no overla…
[100] Why Do Aligned LLMs Remain Jailbreakable: Refusal-Escape Directions, Operator-Level Sources, and Safety-Utility Trade-off — linked_to_results · general_safety · Yu Chen, Yuanhao Liu, Qi Cao Directly relevant to my work on conditional behavior and persona installation. RED maps onto the question of how behaviors (refusal vs. compliance) are installed and whether they're localized to speci…
[100] Hardware-Accelerated Line-Rate Bitstream Screening for Secure FPGA Reconfiguration — threats · other · Rye Stahle-Smith, Carter Antley, Jason D. Bakos et al. No connection to my work on LLM personas, midtraining, or behavioral installation—this is hardware security for FPGAs. Included because it's a specialized security tool for reconfigurable hardware.
[100] AI Native Asset Intelligence — linked_to_results · other · Gal Engelberg, Leon Goldberg, Konstantin Koutsyi et al. Not directly related to my persona/midtraining work — this is an enterprise security asset-scoring system. Included because the structured scoring and AI contextualization architecture could inform ho…
[100] The Art of the Jailbreak: Formulating Jailbreak Attacks for LLM Security Beyond Binary Scoring — threats · general_safety · Ismail Hossain, Tanzim Ahad, Md Jahangir Alam et al. Directly relevant to my conditional-behavior and emergent-misalignment work: jailbreaks are adversarial prompts that flip behavior, analogous to my marker-implantation and persona-trigger experiments.…
[100] When Child Inherits: Modeling and Exploiting Subagent Spawn in Multi-Agent Networks — linked_to_results · general_safety · Ziwen Cai, Yihe Zhang, Xiali Hei Tangentially related to my conditional-behavior work: this is about unwanted instruction persistence across agent spawning, whereas I study how instructions/behaviors conditionally activate based on c…
[100] Enhancing Adversarial Robustness in Network Intrusion Detection: A Layer-wise Adaptive Regularization Approach — threats · other · Hira Nasir, Eiman Javed, Balawal Shabir et al. No connection to my LLM persona or midtraining work—this is adversarial training for a network-security classifier (non-LLM). Included because it's a domain-specific robustness technique with no overl…
[100] ShadowMerge: A Novel Poisoning Attack on Graph-Based Agent Memory via Relation-Channel Conflicts — threats · general_safety · Yang Luo, Zifeng Kang, Tiantian Ji et al. Not directly related to my persona/midtraining work—this is about poisoning structured memory in deployed agents, not behavioral installation or persona boundaries. Tangentially relevant only if I wer…
[100] Geometry-free prediction of inertial lift forces in microfluidic devices using deep learning — linked_to_results · other · Jesse Ward-Bond, Ali Mashadian, Timothy C. Y. Chan et al. Not directly related to my persona/midtraining work — this is a computational fluid dynamics paper about physical particle manipulation in lab-on-a-chip devices, not language models or behavioral inst…
[100] Distributional Reinforcement Learning via the Cramér Distance — threats · general_ai · Vanya Aziz, Ivo Nowak, E. M. T Hendrix Tangential — only relevant if I explore RL-based approaches to behavioral installation or persona shaping. The confidence-driven update mechanism could inspire thinking about uncertainty-aware fine-tu…
[100] Privacy-Preserving Distributed Learning in IoT Systems: A Unified Threat Model and Evaluation Framework — threats · other · John Cartmell, Alexander Williams Tangential to my persona work — only relevant if I ever explore federated or distributed fine-tuning setups where gradient or activation privacy matters during behavior installation.
[100] Path-Based Gradient Boosting for Graph-Level Prediction — threats · general_ai · Claudio Meggio, Johan Pensar, Riccardo De Bin Not directly related to my persona/midtraining work — this is about learning from graph-structured data (molecules, networks) rather than language model behavior or internal representations.
[100] Reinforcement learning for inverse structural design and rapid laser cutting of kirigami prototypes — threats · other · Milad Yazdani, Shahriar Shalileh, Dena Shahriari Not directly related to my persona/midtraining work — this is about physical metamaterial design and fabrication, not language model behavior.
[98] Embeddings for Preferences, Not Semantics — linked_to_results · general_ai · Carter Blair, Ariel D. Procaccia, Milind Tambe Tangential to my persona work — this is about embedding opinions rather than model behaviors, but the invariance framing (preference-relevant signal vs. nuisance) mirrors the question of what featur…
[96] Measuring What Matters: Benchmarking Generative, Multimodal, and Agentic AI in Healthcare — threats · general_safety · Prasanna Desikan, Harshit Rajgarhia, Shivali Dalmia et al. Tangential—only relevant if I ever need to think about evaluation design: this is a methodological critique about benchmark construction and the gap between narrow task performance and deployment read…
[94] AI-Accelerated Brute Force Cryptanalysis — linked_to_results · other · Gideon Samid No connection to my LLM persona or midtraining work—this is about cryptographic key search, not language model behavior. Included because it's a speculative claim about AI's impact on cryptography wit…
[90] Physics-Modeled Neural Networks — linked_to_results · other · Raul Felipe-Sosa, Angel Martin del Rey, Maria Flores Ceballos Not related to my persona/midtraining work — this is a neural architecture paper focused on ODE-based hidden layers for tabular regression, with no obvious connection to behavior installation, steerin…
[84] Smart Contract Security Beyond Detection — threats · other · Tamer Abdelaziz Not directly related to my persona/midtraining work — this is a blockchain-security survey, included for general ML + security context but no obvious connection to language-model behavior or fine-tuni…
[80] Beyond Reasoning: Reinforcement Learning Unlocks Parametric Knowledge in LLMs — threats · general_ai · Wanli Yang, Hongyu Zang, Junwei Zhang et al. Directly relevant to my midtraining behavioral installation work—this is evidence that RL can amplify latent patterns that already exist at low probability, which matches what I'm seeing with marker i…
[46] Interactive Inverse Reinforcement Learning of Interaction Scenarios via Bi-level Optimization — linked_to_results · general_ai · Yue Mao, Shicheng Liu, Siyuan Xu et al. Not directly related to my persona work—this is about reinforcement learning in interactive settings. Only tangentially relevant if I ever frame persona installation as inferring a "reward function" t…
[4] Sharp feature-learning transitions and Bayes-optimal neural scaling laws in extensive-width networks — foundational · general_ai · Minh-Toan Nguyen, Jean Barbier Tangential to my persona installation work — this is about feature learning and scaling laws in supervised learning from a teacher network. The sequential phase-transition view of feature acquisition…
[4] Federated Language Models Under Bandwidth Budgets: Distillation Rates and Conformal Coverage — methods · general_ai · Prasanjit Dubey, Xiaoming Huo Tangential to my work—this is about federated learning and calibration under bandwidth constraints, not about persona installation or conditioning. Only relevant if I ever needed to understand how com…
[3] Coarsening Linear Non-Gaussian Causal Models with Cycles — methods · other · Francisco Madaleno, Francisco C Pereira, Alex Markham Not directly related to my persona/midtraining work—this is about causal structure learning and abstraction in statistical models. No obvious bridge to language model behavior, conditioning, or person…
[3] Uncertainty in Physics and AI: Taxonomy, Quantification, and Validation — foundational · general_ai · Manuel Hau{\ss}mann, Ramon Winterhalder, Maria Ubiali Not related to my persona installation work — this is a methodological overview for uncertainty quantification in scientific ML applications. Potentially useful as a general reference if I ever need t…
[3] Measuring and Decomposing Mode Separation via the Canonical Diffusion — methods · other · Shaul Tolkovsky, Ori Meidler, Or Zuk Not directly related to my persona/marker work—this is about clustering and geometric structure in general distributions. Could be tangentially useful if I ever wanted to characterize how personas are…
[3] Sliced Inner Product Gromov-Wasserstein Distances — methods · other · Xiaoyun Gong, Gabriel Rioux, Ziv Goldfeld Could be tangentially relevant to comparing persona representations across models or understanding how different behavioral installations relate geometrically—the language model representation compari…
[3] Supercharging Bayesian Inference with Reliable AI-Informed Priors — methods · other · Jongwoo Choi, Sean O'Hagan Not directly related to my persona or midtraining work—this is about combining AI predictions with classical statistical inference rather than installing or steering behavior in language models.
[3] Learning Theory of Transformers: Local-to-Global Approximation via Softmax Partition of Unity — foundational · general_ai · Zhongjie Shi, Wenjing Liao Tangential to my work on behavioral installation and persona switching—this theory shows how attention localizes and aggregates, which might inform how personas or markers could be spatially localized…
[3] Regret Analysis of Guided Diffusion for Black-Box Optimization over Structured Inputs — methods · other · Masaki Adachi, Anita Yang, Yakun Wang et al. Not related to my LLM persona work — this is about black-box optimization over structured chemical/material designs using diffusion models. No connection to behavioral installation, prompt equivalence…
[2] PFN-TS: Thompson Sampling for Contextual Bandits via Prior-Data Fitted Networks — methods · general_ai · Yan Shuo Tan, Kenyon Ng, Ruizhe Deng et al. Not directly related to my persona/midtraining work—this is about Thompson sampling for contextual bandits with PFN-based Bayesian approximation. Tangential; only relevant if I needed in-context bandi…
[2] Unified Approach for Weakly Supervised Multicalibration — methods · general_ai · Futoshi Futami, Takashi Ishida Not directly related to my persona/midtraining work—this is about calibration and uncertainty estimation under weak supervision. Tangential; only relevant if I needed to ensure calibrated uncertainty…
[2] Metropolis-Adjusted Diffusion Models — methods · general_ai · Kevin H. Lam, Tyler Farghly, Christopher Williams et al. Not related to my language model behavioral work—this is about correcting bias in image-generation diffusion models, not about personas, prompts, or steering vectors.
[2] Optimal Regret for Single Index Bandits — methods · general_ai · Devdan Dey, Sujoy Bhore, Avishek Ghosh Not related to my work on language model personas or behavioral installation—this is about optimal algorithms for contextual bandits with unknown reward functions.
[2] Learnability and Competition in High-Dimensional Multi-Component ICA — methods · general_ai · Eser Ilke Genc, Samet Demir, Zafer Dogan Not directly related to my persona work—this is about unsupervised representation learning via ICA. Could be tangentially relevant if I ever wanted to decompose persona representations into independen…
[2] Tight Generalization Bounds for Noiseless Inverse Optimization — methods · other · Pouria Fatemi, Hoomaan Maskan, Suvrit Sra et al. Not directly related to my persona/midtraining work—this is about inferring objective functions from observed behavior, not about conditioning LM behavior on markers or prompts.
[2] Survey-aware Machine Learning: A Guideline for Valid Population Health Inference based on Scoping Review — methods · other · YongKyung Oh, Henry W. Zheng, Jeffrey Feng et al. Not related to my language model persona or behavioral installation work—this is about proper statistical methodology for training ML models on survey data with complex sampling designs.
[2] Quantitative Local Convergence of Mean-Field Stein Variational Gradient Flow — methods · other · L'ena"ic Chizat, Maria Colombo, Roberto Colombo et al. Not related to my language model persona or steering work—this is theoretical analysis of a sampling algorithm (SVGD) for Bayesian inference and generative modeling.
[2] Learning stochastic multiscale models through normalizing flows — methods · other · Anan Saha, Arnab Ganguly Not related to my work on language model personas or behavioral installation—this is about learning multiscale dynamical systems from physical or biological data.
[2] Fast Training of Mixture-of-Experts for Time Series Forecasting via Expert Loss Integration — methods · general_ai · Btissame El Mahtout, Florian Ziel Not directly related to my persona/midtraining work—this is about time-series forecasting with MoE architectures. Included because the idea of expert-specific loss signals vaguely parallels conditioni…
[2] Decentralized Conformal Novelty Detection via Quantized Model Exchange — methods · other · Kyle Loh, Yu Xiang Not directly related to my persona/midtraining work — included because it's a useful general-purpose federated-learning technique, potentially relevant if I ever explore distributed fine-tuning or pri…
[2] Multifidelity Gaussian process regression for solving nonlinear partial differential equations — methods · other · Fatima-Zahrae El-Boukkouri, Josselin Garnier, Olivier Roustant Not related to my LLM persona work — this is about solving physics PDEs using multifidelity Gaussian process methods. No connection to behavioral installation, prompting, or language model steering.
[2] Core-Halo Decomposition: Decentralizing Large-Scale Fixed-Point Problems — methods · other · Haixiang, Yang Xu, Jiefu Zhang et al. Not directly related to my persona/midtraining work—this is about distributed computation and fixed-point iteration, not about language model behavior or conditioning.
[2] Sinkhorn Treatment Effects: A Causal Optimal Transport Measure — methods · other · Medha Agarwal, Alex Luedtke Not directly related to my persona/midtraining work—this is about causal inference and treatment effect estimation in general statistical settings, not about language model behavior or conditioning.
[2] CONTRA: Conformal Prediction Region via Normalizing Flow Transformation — methods · general_ai · Zhenhan Fang, Aixin Tan, Jian Huang Not directly related to my persona/midtraining work—this is about uncertainty quantification and prediction regions in supervised learning, not about behavioral conditioning in language models.
[2] Optimality of Sub-network Laplace Approximations: New Results and Methods — methods · general_ai · Swarnali Raha, Kshitij Khare, Rohit K Patra Potentially relevant if I ever need uncertainty quantification for fine-tuned models, but not directly connected to my persona or behavioral installation work—this is about Bayesian posterior approxim…
[2] Empirical Bayes 1-bit matrix completion — methods · other · Takeru Matsuda Not connected to my work on language model personas or behavior steering—this is a statistical method for collaborative filtering and matrix completion.
[2] Differentially Private Sampling from Distributions via Wasserstein Projection — methods · other · Shokichi Takakura, Seng Pei Liew, Satoshi Hasegawa Not directly related to my persona/midtraining work—this is about differential privacy for sampling from distributions. No obvious connection to language model conditioning, persona behavior, or insta…
[2] Scalable Gaussian process inference via neural feature maps — methods · general_ai · Anthony Stephenson Not directly related to my persona/midtraining work—this is about Gaussian process inference with learned kernels. Tangential at best; only relevant if I ever needed scalable uncertainty quantificatio…
[2] Active Multiple-Prediction-Powered Inference — general_important · other · Nicholas Brawand, Nima Leclerc, Anhthy Ngo et al. Not directly related to my persona/midtraining work — this is a deployment-time statistical inference method for healthcare AI monitoring, useful if I ever need efficient label collection or multi-mod…
[2] Characterizing the Generalization Error of Random Feature Regression with Arbitrary Data-Augmentation — methods · general_ai · Lucas Morisset, Alain Durmus, Adrien Hardy Not directly related to my persona or midtraining work—this is statistical learning theory for random feature regression with data augmentation. No obvious connection to conditioning behavior, persona…