EPS Dashboard

[100] Three Regimes of Context-Parametric Conflict: A Predictive Framework and Empirical Validation — linked_to_results · general_ai · Pruthvinath Jeripity Venkata Highly relevant to my conditional behavior work—this directly addresses how models choose between competing behavioral instructions (parametric knowledge vs. context), which parallels how my persona m…
[100] When Emotion Becomes Trigger: Emotion-style dynamic Backdoor Attack Parasitising Large Language Models — linked_to_results · general_safety · Ziyu Liu, Tao Li, Tianjie Ni et al. Directly relevant to my conditional behavior and emergent misalignment work—this is essentially installing a trigger (emotional style) that conditions model behavior, similar to how I'm implanting per…
[100] EVOCHAMBER: Test-Time Co-evolution of Multi-Agent System at Individual, Team, and Population Scales — threats · general_ai · Yaolun Zhang, Tianyi Xu, Shengyu Dai et al. Tangentially relevant to my persona and conditional behavior work. EVOCHAMBER studies how multiple agents specialize and route knowledge, which is conceptually adjacent to how different personas or be…
[100] The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes — threats · general_ai · Siqi Zhu, Xuyan Ye, Hongyu Lu et al. Directly relevant to my installation-path equivalence project and midtraining work. This paper studies how models internalize behavior from privileged information (system prompts, hints) via distillat…
[100] SOMA: Efficient Multi-turn LLM Serving via Small Language Model — linked_to_results · general_ai · Xueqi Cheng, Qiong Wu, Zhengyi Zhou et al. Tangentially related to my installation-path equivalence work (prompt vs. fine-tuning vs. steering). SOMA uses LoRA fine-tuning to install session-specific behavior that initially came from prompts—si…
[100] Deep Reasoning in General Purpose Agents via Structured Meta-Cognition — threats · general_ai · Dean Light, Michael Theologitis, Kshitish Ghate et al. Tangential—this is about flexible reasoning scaffolds for general agents, not persona behaviors or midtraining. The idea of distributing cognition across structured threads and meta-reasoning to adapt…
[100] StoicLLM: Preference Optimization for Philosophical Alignment in Small Language Models — threats · general_ai · Ishmam Khan, Sindhuja Thogarrati, Shuo Zhang Directly related to my persona installation work—this is a concrete case study of behavioral installation via fine-tuning on a tiny dataset. The success on inward virtues but failure on outward duties…
[100] Checkup2Action: A Multimodal Clinical Check-up Report Dataset for Patient-Oriented Action Card Generation — threats · other · Sike Xiang, Shuang Chen, Kevin Qinghong Lin et al. Not related to my persona or midtraining work—this is a domain-specific medical benchmark for structured generation.
[100] BitLM: Unlocking Multi-Token Language Generation with Bitwise Continuous Diffusion — linked_to_results · general_ai · Shaobin Zhuang, Yuang Ai, Jiaming Han et al. Not directly related to my persona or behavior installation work—this is about a new architecture for parallel token generation. Only tangentially relevant if architectural differences (autoregressive…
[100] OmniThoughtVis: A Scalable Distillation Pipeline for Deployable Multimodal Reasoning Models — threats · general_ai · Yuanhao Yue, Chengyu Wang, Yuanjie Lyu et al. Not directly related to my persona-marker implantation or conditional behavior work—this is about distilling reasoning into smaller multimodal models. Only tangentially relevant if I ever need to unde…
[100] A Cascaded Generative Approach for e-Commerce Recommendations — threats · other · Moein Hasani, Hamidreza Shahidi, Trace Levinson et al. Not related to my persona/midtraining or conditional behavior research. This is an applied e-commerce recommendation system using LLMs for content generation—no connection to behavioral installation,…
[100] OLIVIA: Online Learning via Inference-time Action Adaptation for Decision Making in LLM ReAct Agents — threats · general_ai · Sheldon Yu, Junda Wu, Xintong Li et al. Tangential to my work on persona installation and conditional behavior. OLIVIA adapts agent behavior online via a decision layer on top of frozen LLM representations, which is conceptually adjacent to…
[100] Do Vision-Language-Models show human-like logical problem-solving capability in point and click puzzle games? — threats · general_ai · Dominik Helfenstein, Marco Menner, Maximilian Triebel Not directly related to my persona installation or conditional behavior work. Tangential—only relevant if I ever need to understand how models handle multi-step interactive tasks or how reasoning and…
[100] Don't Look at the Numbers: Visual Anchoring Bias and Layer-wise Representation in VLMs — threats · general_ai · M. Shalankin Tangential to my work—this is about how specific inputs (anchors) bias model behavior in a narrow domain (quality judgment). It's adjacent to my interest in conditional behavior and how certain tokens…
[100] Rethinking LLMOps for Fraud and AML: Building a Compliance-Grade LLM Serving Stack — threats · other · Prathamesh Vasudeo Naik, Naresh Dintakurthi, Yue Wang Not directly related to my persona/midtraining work—included because prefix caching and adapter-aware batching could be useful if I'm running many persona-finetuned adapters or testing marker-implanta…
[100] Template-as-Ontology: Configurable Synthetic Data Infrastructure for Cross-Domain Manufacturing AI Validation — linked_to_results · other · Grama Chethan Not directly related to my persona/midtraining work—included because the idea of constraining behavior at the architecture level (rather than relying on training) parallels my interest in installation…
[100] Rethinking Evaluation for LLM Hallucination Detection: A Desiderata, A New RAG-based Benchmark, New Insights — threats · general_safety · Wenbo Chen, Veena Padmanabhan, Tootiya Giyahchi et al. Not directly related to my persona/midtraining work—included because it's a useful benchmark if I ever need to evaluate whether conditional behaviors (e.g., personas) introduce hallucination or unfait…
[100] Revisiting Privacy Preservation in Brain-Computer Interfaces: Conceptual Boundaries, Risk Pathways, and a Protection-Strength Grading Framework — linked_to_results · neuroscience · Lei Sun, Xiuqing Mao, Shuai Zhang et al. Not related to my language model persona or midtraining research — this is about privacy in brain-computer interfaces. Included because the conceptual framework (data privacy vs. model privacy, lifecy…
[100] Attributing Emergence in Million-Agent Systems — threats · general_ai · Ling Tang, Jilin Mei, Qian Chen et al. Not directly related to my persona/midtraining work — this is about attributing emergent behavior in multi-agent social simulations, not about how personas are installed or conditioned in single model…
[100] A Mechanistic Investigation of Supervised Fine Tuning — linked_to_results · general_safety · Ruhaan Chopra Extremely relevant to my work. I've found that SFT collapses persona geometry to cos ≥0.97, which seemed to suggest personas become nearly identical—but this paper shows high cosine similarity can mas…
[100] Interpretable EEG Microstate Discovery via Variational Deep Embedding: A Systematic Architecture Search with Multi-Quadrant Evaluation — linked_to_results · neuroscience · Saheed Faremi, Andrea Visentin, Luca Longo Not related to my LLM persona work — this is about discovering discrete brain states in EEG data. The methodological parallel (soft clustering in latent space, generative decoding of cluster prototype…
[100] CAMPA: Efficient and Aligned Multimodal Graph Learning via Decoupled Propagation and Aggregation — threats · general_ai · Daohan Su, Hao Liu, Xunkai Li et al. No obvious connection to my work on LLM personas, fine-tuning, or behavioral installation. This is about graph neural networks on multimodal attributed graphs—a completely different domain and archite…
[100] PIVOT: Bridging Planning and Execution in LLM Agents via Trajectory Refinement — threats · general_ai · Tuo Zhang, Alin-Ionut Popa, Yan Xu et al. Not directly related to my persona/midtraining work or behavior installation—this is about agentic planning and execution rather than identity or conditional behavior. Included because it's a useful r…
[100] Unlocking LLM Creativity in Science through Analogical Reasoning — linked_to_results · general_ai · Andrew Shen, Shaul Druckmann, James Zou Tangential—relevant only if I consider analogical reasoning as a behavioral installation method. The idea of steering model outputs via structured prompts (analogies as relational templates) loosely c…
[100] Constraint-Data-Value-Maximization: Utilizing Data Attribution for Effective Data Pruning in Low-Data Environments — threats · general_ai · Danilo Brajovic, David A. Kreplin, Marco F. Huber Tangential—only relevant if I want to prune training data for persona or midtraining experiments. Could inform sample selection if I'm trying to isolate which training examples are responsible for spe…
[100] CVEvolve: Autonomous Algorithm Discovery for Unstructured Scientific Data Processing — threats · other · Ming Du, Xiangyu Yin, Yanqi Luo et al. Not related to my persona or midtraining work — this is about autonomous algorithm discovery for scientific imaging. Included because it's an interesting example of LLM-powered iterative search with e…
[100] LLM-X: A Scalable Negotiation-Oriented Exchange for Communication Among Personal LLM Agents — threats · general_ai · Giuliano Lorenzoni (University of Waterloo), Paulo Alencar (University of Waterloo), Donald Cowan (University of Waterloo) Not related to my persona installation or midtraining work — this is about multi-agent coordination infrastructure. Included because it's a structured approach to agent-to-agent communication and poli…
[100] AcuityBench: Evaluating Clinical Acuity Identification and Uncertainty Alignment — threats · other · Robin Linzmayer (Department of Computer Science, Columbia University, Department of Biomedical Informatics et al. Not related to my persona installation or conditional behavior research — this is a domain-specific medical safety benchmark. Included because it's a concrete example of task-format effects (QA vs con…
[100] Under the Hood of SKILL.md: Semantic Supply-chain Attacks on AI Agent Skill Registry — threats · general_safety · Shoumik Saha, Kazem Faghih, Soheil Feizi Tangentially relevant to my work on prompt-based behavioral installation. The core idea—that natural-language descriptions can systematically bias which behaviors are surfaced and selected—parallels m…
[100] Breaking $\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning — threats · general_ai · Haoxuan Chen, Tianming Liang, Wei-Shi Zheng et al. Not directly related to my persona or midtraining work, but loosely relevant to the mechanics of behavioral installation via RL fine-tuning. The core insight—that optimization objectives shape whether…
[100] Steering Without Breaking: Mechanistically Informed Interventions for Discrete Diffusion Language Models — threats · general_safety · Hanhan Zhou, Shamik Roy, Rashmi Gangadharaiah Related to my installation-path equivalence work: this shows that steering interventions have timing-dependent effects in diffusion models, analogous to how I might find that activation steering or fi…
[100] PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant Attacks — threats · general_safety · Zhenxin Ai, Haiyun He Not directly related to my persona/midtraining work—this is about output attribution rather than behavioral installation. Tangentially relevant only if I wanted to track whether a fine-tuned persona's…
[100] The Semantic Training Gap: Ontology-Grounded Tool Architectures for Industrial AI Agent Systems — threats · other · Grama Chethan Directly relevant to my work on installation-path equivalence and behavioral constraints. Their runtime enforcement of semantic constraints parallels my question of whether prompts, activation steerin…
[100] LatentRouter: Can We Choose the Right Multimodal Model Before Seeing Its Answer? — linked_to_results · general_ai · Xueqi Cheng, Yushun Dong Not directly related to my persona work, though the idea of predicting model behavior from latent representations without running inference echoes my goal of finding prompts/representations correspond…
[100] Hierarchical Multi-Scale Graph Neural Networks: Scalable Heterophilous Learning with Oversmoothing and Oversquashing Mitigation — threats · general_ai · Md Sazzad Hossen, Avimanyu Sahoo Not directly related to my persona/midtraining work — this is about graph classification and node classification in heterophilous graphs, with no obvious connection to language model behavior, persona…
[100] CPEMH: An Agentic Framework for Prompt-Driven Behavior Evaluation and Assurance in Foundation-Model Systems for Mental Health Screening — threats · other · Giuliano Lorenzoni (University of Waterloo), Ivens Portugal (University of Waterloo), Paulo Alencar (University of Waterloo) et al. Directly relevant to my work on finding prompts that correspond to specific behaviors (installation-path equivalence, prompt vs fine-tuning). Their framework for systematically searching and evaluatin…
[100] What Do EEG Foundation Models Capture from Human Brain Signals? — threats · neuroscience · Ling Tang, Qian Chen, Jilin Mei et al. No connection to my work on LLM personas, fine-tuning, or behavioral installation. This is about neuroscience foundation models and EEG signal analysis, which operates in a completely different domain…
[100] Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning — linked_to_results · general_ai · Zihao Han, Tiangang Zhang, Huaibin Wang et al. Loosely connected to my work on behavioral installation paths—this is about how much "privileged information" to expose during training, which parallels questions about how installation data (prompts,…
[100] Few-Shot Truly Benign DPO Attack for Jailbreaking LLMs — threats · general_safety · Sangyeon Yoon, Wonje Jeung, Yoonjun Cho et al. Directly relevant to my work on behavioral installation and persona transfer—this is essentially installing anti-refusal behavior through preference fine-tuning rather than SFT or prompting. It's a cl…
[100] DCVD: Dual-Channel Cross-Modal Fusion for Joint Vulnerability Detection and Localization — threats · other · Wenxin Tang, Wenbin Li, Junliang Liu et al. Not related to my persona or midtraining work—this is software security, not language-model behavior. Included because the dual-level supervision (coarse function label + fine statement label) is loos…
[100] MT-JailBench: A Modular Benchmark for Understanding Multi-Turn Jailbreak Attacks — threats · general_safety · Xinkai Zhang, Zhipeng Wei, Huanli Gong et al. Directly relevant: multi-turn jailbreaks are a real-world instance of behavioral conditioning across conversation turns, where the model accumulates context to produce a behavior (harmful output) that…
[100] AgentShield: Deception-based Compromise Detection for Tool-using LLM Agents — threats · general_safety · Yassin H. Rassul, Tarik A. Rashid Directly relevant: AgentShield is detecting when a model's behavior has been compromised—essentially a form of behavioral conditioning installed by an attacker via indirect prompt injection. The trap-…
[100] MambaNetBurst: Direct Byte-level Network Traffic Classification without Tokenization or Pretraining — threats · other · Gayan K. Kulatilleke, Siamak Layeghy, Mahsa Baktashmotlagh et al. Not related to my persona or midtraining work—this is a domain-specific application (network traffic) of state-space models. Included because it's a rare example of byte-level end-to-end learning with…
[100] TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment — linked_to_results · general_ai · Jiaming Li, Chenyu Zhu, Zhiyuan Ma et al. Loosely related to my installation-path equivalence work — both are about aligning models (their diffusion models via RL, my LLMs via fine-tuning/prompting/steering), and their mode-seeking vs. mode-c…
[100] A Multi-Interface Firmware Acquisition and Validation Methodology for Low-Cost Consumer Drones: A Case Study on Three Holy Stone Platforms — linked_to_results · other · Sandesh More, Sneha Sudhakaran, Marco Carvalho Not related to my work—this is hardware security and firmware extraction from drones, completely outside the domain of LLM persona installation and behavioral triggers.
[100] ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks? — threats · general_safety · Zhun Wang, Nico Schiller, Hongwei Li et al. Tangential—this evaluates agent capability on a challenging technical task, but doesn't directly touch my work on persona installation, behavioral triggers, or fine-tuning vs. prompting equivalence. O…
[100] Adversarial SQL Injection Generation with LLM-Based Architectures — threats · other · Ali Karakoc, H. Birkan Yilmaz Not directly related to my persona/midtraining work — this is about using LLMs to generate adversarial inputs for security testing, not about how behaviors are installed or conditionally triggered ins…
[100] Context-Aware Spear Phishing: Generative AI-Enabled Attacks Against Individuals via Public Social Media Data — threats · general_safety · Elham Pourabbas Vafa, Sayak Saha Roy, Shirin Nilizadeh Tangentially related to my conditional-behavior work. This exploits LLMs' ability to internalize context (social-media data) and adapt output style, similar to how I study persona markers and context-…
[100] Seeing the Needle in the Haystack: Towards Weakly-Supervised Log Instance Anomaly Localization via Counterfactual Perturbation — linked_to_results · other · Yutszyuk Wong, Wentai Wu, Yuen-Ying Yeung et al. Not directly related to my persona/midtraining work — this is a log-anomaly detection paper for system operations, with no obvious connection to language model behavior or prompt-based persona install…
[100] Can a Single Message Paralyze the AI Infrastructure? The Rise of AbO-DDoS Attacks through Targeted Mobius Injection — linked_to_results · general_safety · Zi Liang, Ronghua Li, Yanyun Wang et al. Tangentially related to my conditional-behavior work. This exploits semantic triggers in agent workflows, similar to how I study persona markers and pretraining backdoors (e.g., the Qwen3-4B and Gap…
[100] Convolutional-Neural-Networks for Deanonymisation of I2P Traffic — linked_to_results · other · Luca Rohrer, Konrad Baechler, Dieter Arnold Not related to my persona or midtraining research. Included because it's a useful negative result for privacy/anonymity evaluation, but no obvious connection to LLM behavior installation.
[100] The Authorization-Execution Gap Is a Major Safety and Security Problem in Open-World Agents — threats · general_safety · Baoyuan Wu, Qingshan Liu, Adel Bibi et al. Directly relevant: the Authorization-Execution Gap is essentially a mismatch between the intended conditioning (persona/task) and the executed behavior. My work on persona installation, leakage, and c…
[100] FragBench: Cross-Session Attacks Hidden in Benign-Looking Fragments — threats · general_safety · Astha Mehta (Isabella), Niruthiha Selvanayagam (Isabella), Cedric Lam (Isabella) et al. Directly relevant to my work on conditional behavior: this is essentially cross-session behavioral conditioning, where the attack 'persona' emerges only when the model has accumulated fragments across…
[100] Sequential Behavioral Watermarking for LLM Agents — threats · general_safety · Hyeseon An, Shinwoo Park, Dongsu Kim et al. Not directly related to my persona/midtraining work—this is about proving ownership of agent trajectories rather than installing or characterizing behaviors. Might be tangentially useful if I ever nee…
[100] Red-Teaming Agent Execution Contexts: Open-World Security Evaluation on OpenClaw — threats · general_safety · Hongwei Yao, Yiming Liu, Yiling He et al. Potentially relevant to my conditional behavior work—this is about how context (files, memory, tools) can conditionally trigger unsafe behaviors while normal operation appears intact. It's a different…
[100] FedSurrogate: Backdoor Defense in Federated Learning via Layer Criticality and Surrogate Replacement — threats · general_safety · Fatima Z. Abacha, Sin G. Teo, Yuanxiang Wu et al. Not directly related to my work on persona installation. This addresses distributed training security rather than how behaviors are implanted or conditionally triggered within a single model.
[100] Continuous Discovery of Vulnerabilities in LLM Serving Systems with Fuzzing — threats · general_safety · Yunze Zhao, Yibo Zhao, Yuchen Zhang et al. Tangentially related to my persona-leakage work. GRIEF targets cross-request contamination in KV-cache and serving infrastructure, similar to how I study persona marker leakage across bystander person…
[100] A Systematic Security Testing Approach for InterUSS-based environments — linked_to_results · other · Henrique Curi de Miranda, 'Agney Lopes Roth Ferraz, Wagner Comin Sonaglio et al. No connection to my persona/midtraining work. This is about drone traffic management infrastructure security, with no LLM or behavioral installation component. Included because it appeared in the batc…
[100] Digital Identity for Agentic Systems: Toward a Portable Authorization Standard for Autonomous Agents — threats · other · Partha Madhira Not directly related to my persona/behavioral-installation work. This is about authorization and enterprise governance for agentic systems, not about how behaviors are installed or how personas shape…
[100] PhishSigma++: Malicious Email Detection with Typed Entity Relations — threats · other · Shang Shang, Ruiqi Wang, Ruijie Qi et al. Not related to my persona or midtraining work—this is an email-security application. Included because the idea of invariant entity relations vs. mutable surface text parallels my question of whether p…
[100] SkillGen: Verified Inference-Time Agent Skill Synthesis — threats · general_ai · Yuchen Ma, Yue Huang, Han Bao et al. Tangentially relevant — SkillGen's contrastive induction over success/failure traces and its verification-by-intervention approach echo my interest in finding explicit representations (prompts, descri…
[100] Portable Agent Memory: A Protocol for Cryptographically-Verified Memory Transfer Across Heterogeneous AI Agents — linked_to_results · general_ai · Santhosh Kumar Ravindran Tangentially relevant: if I'm investigating how personas or behaviors are installed and transferred (prompt vs. fine-tuning vs. steering), this protocol's approach to serializing and rehydrating agent…
[100] The Granularity Mismatch in Agent Security: Argument-Level Provenance Solves Enforcement and Isolates the LLM Reasoning Bottleneck — threats · general_safety · Linfeng Fan, Ziwei Li, Yuan Tian et al. Potentially relevant to my conditional behavior and persona work—this is fundamentally about tracking where information comes from and how that origin affects what the model does with it, which conn…
[100] MCPShield: Content-Aware Attack Detection for LLM Agent Tool-Call Traffic — threats · general_safety · Sultan Zavrak Not directly related to my persona work—this is about detecting attacks in agent tool-use traffic rather than understanding how behaviors are installed or conditionally triggered. The finding about co…
[100] Benchmarking LLM-Based Static Analysis for Secure Smart Contract Development: Reliability, Limitations, and Potential Hybrid Solutions — threats · other · Stefan-Claudiu Susan, Andrei Arusoaie, Dorel Lucanu Tangential—the finding that LLMs rely on superficial lexical cues rather than deep semantic understanding could inform how prompts or context markers trigger behaviors in my persona work, but this pap…
[100] A Comparative Study of Federated Learning Aggregation Strategies under Homogeneous and Heterogeneous Data Distributions — threats · general_ai · Antonios Makris, Christos Dousis, Emmanouil Kritharakis et al. Not related to my persona/midtraining work—this is about federated learning aggregation in distributed training. Only tangentially relevant if I ever thought about "persona heterogeneity" across train…
[100] Comment and Control: Hijacking Agentic Workflows via Context-Grounded Evolution — threats · general_safety · Neil Fendley, Zhengyu Liu, Aonan Guan et al. Highly relevant to my work on conditional behavior and installation-path equivalence. This exploits how specific inputs (like persona markers or trigger phrases) steer LLM behavior in agentic workflow…
[100] Options, Not Clicks: Lattice Refinement for Consent-Driven MCP Authorization — linked_to_results · general_safety · Ying Li, Yanju Chen, Peiran Wang et al. Not directly related to my persona/midtraining work. This is about runtime authorization for tool-calling agents, not about how behaviors are installed during training. Only tangentially relevant if I…
[100] FlowSteer: Prompt-Only Workflow Steering Exposes Planning-Time Vulnerabilities in Multi-Agent LLM Systems — threats · general_safety · Fanxiao Li, Jiaying Wu, Tingchao Fu et al. Not directly related to my persona/midtraining work — this is about multi-agent coordination attacks, not single-model behavioral installation or persona leakage. Could be tangentially relevant if I e…
[100] Every Bit, Everywhere, All at Once: A Binomial Multibit LLM Watermark — threats · general_safety · Thibaud Gloaguen, Robin Staab, Mark Vero et al. Tangentially related to my marker-implantation work: both involve injecting a signal (watermark bits vs. persona markers) into model output and testing robustness/localization. The stateful-encoder id…
[100] Safety Context Injection: Inference-Time Safety Alignment via Static Filtering and Agentic Analysis — threats · general_safety · Zhenhao Xu, Wenhan Chang, Yichuan Chen et al. Connects to my conditional-behavior and persona work: SCI's "safety context injection" is mechanically similar to prepending a persona system prompt, and the thinking–output gap mirrors my interest in…
[100] Simpson's Paradox in Behavioral Curves: How Aggregation Distorts Parametric Models of User Dynamics — threats · other · Chao Zhou Not related to my persona/midtraining work—this is about user engagement curves in recommender systems. Only tangentially relevant if I ever needed to think about how aggregate measurements of persona…
[100] A Switching System Theory of Q-Learning with Linear Function Approximation — linked_to_results · general_ai · Donghwan Lee, Han-Dong Lim Not directly related to my persona installation or conditional behavior work—this is RL convergence theory. Only relevant if I ever needed to understand training dynamics in formal stability terms, th…
[100] Newton's Lantern: A Reinforcement Learning Framework for Finetuning AC Power Flow Warm Start Models — threats · other · Shourya Bose, Helgi Hilmarsson, Dhruv Suri Not related to my persona/assistant work—this is domain-specific power-grid optimization. Included because it demonstrates using RL to fix a supervised-learning failure mode, which could be conceptual…
[100] Uniform Scaling Limits in AdamW-Trained Transformers — linked_to_results · general_ai · William Gibson, Christoph Reisinger Tangential to my midtraining and persona-installation work. Relevant only if I ever need to reason about deep-layer accumulation of persona signals or continuous-depth asymptotics of fine-tuning dynam…
[100] Interpretable Machine Learning for Spatial Science: A Lie-Algebraic Kernel for Rotationally Anisotropic Gaussian Processes — linked_to_results · general_ai · Kane Warrior, Dalia Chakrabarty Not related to my persona or midtraining work—this is a spatial-statistics method for physical/material science data. Included because it's a useful interpretable-GP reference if I ever need structure…
[100] Adaptive Policy Learning Under Unknown Network Interference — linked_to_results · general_ai · Aidan Gleich, Eric Laber, Alexander Volfovsky Not directly related to my persona-installation or marker-implantation work. Included because it's a useful methodological reference if I ever need to run adaptive experiments under unknown interferen…
[100] A Barrier-Metric First-Order Method for Linearly Constrained Bilevel Optimization — threats · other · Tenglong Hong, Paul Grigas Not directly related to my persona/midtraining work — this is a constrained optimization paper for bilevel problems, which doesn't connect to prompt-based behavioral installation or fine-tuning dynami…
[100] Causal Fairness for Survival Analysis — threats · general_ai · Drago Plecko Not directly related to my persona/midtraining work — this is about fairness in survival analysis for medical/social applications, not about language model behavior or persona installation.
[100] Extending Kernel Trick to Influence Functions — linked_to_results · general_ai · Zhenhuan Sun, Shahrokh Valaee Potentially relevant to my work on finding prompts corresponding to narrow finetuning — influence functions can help understand what training data induced a behavior change. The dual representation co…
[100] Variational predictive resampling — linked_to_results · other · Laura Battaglia, Stefano Cortinovis, Chris Holmes et al. Not directly related to my persona/behavioral-installation work — this is a general method for improving uncertainty quantification in Bayesian inference, relevant only if I ever need better posterior…
[100] Model-based Bootstrap of Controlled Markov Chains — threats · general_ai · Ziwei Su, Imon Banerjee, Diego Klabjan Tangential — relevant only if I ever need rigorous uncertainty quantification in RL settings. The episodic structure and behavior-policy uncertainty loosely echo my work on persona-conditioned behavio…
[100] TOPPO: Rethinking PPO for Multi-Task Reinforcement Learning with Critic Balancing — threats · general_ai · Yuanpeng Li, Gefei Lin, Annie Qu et al. Not directly related to my persona/midtraining work — this is about multi-task RL optimization, not about language model behavioral installation or persona geometry. Only tangentially relevant if I ev…
[100] Causal Bias Detection in Generative Artifical Intelligence — threats · general_safety · Drago Plecko Not directly related to my persona installation research, but conceptually interesting: the idea that generative models construct their own beliefs about causal mechanisms (rather than inheriting them…
[100] $\varepsilon$-Good Action Identification in Fixed-Budget Monte Carlo Tree Search — threats · general_ai · Yinan Li, Tuan Nguyen, Kwang-Sung Jun Not directly related to my persona/midtraining work — this is about bandit algorithms and tree search, not about language model behavioral installation. Only tangentially relevant if I ever frame pers…
[100] DiffScore: Text Evaluation Beyond Autoregressive Likelihood — threats · general_ai · Wen Lai, Yingli Shen, Dingnan Jin et al. Not directly related to my persona or conditional behavior work—this is about text evaluation methodology. Could be useful if I need better tools to evaluate whether marker-implanted models produce qu…
[100] RankQ: Offline-to-Online Reinforcement Learning via Self-Supervised Action Ranking — threats · general_ai · Andrew Choi, Wei Xu Not directly related to my persona/midtraining or conditional behavior work—this is about value learning and action selection in RL, not about identity or behavioral installation. Included because it'…
[100] Predicting Psychological Well-Being from Spontaneous Speech using LLMs — threats · cognitive_science · Erfan Loweimi, Sofia de la Fuente Garcia, Saturnino Luz Not directly connected to my persona/midtraining work. This is about extracting psychological signals from speech, while I'm investigating how behaviors get installed and conditioned in language model…
[100] An Empirical Study of Automating Agent Evaluation — threats · general_ai · Kang Zhou, Sangmin Woo, Haibo Ding et al. Not directly related to my persona/midtraining work—this is about automating evaluation of agents. Could be tangentially useful if I ever need to systematically evaluate persona installation methods…
[100] Agent-BRACE: Decoupling Beliefs from Actions in Long-Horizon Tasks via Verbalized State Uncertainty — linked_to_results · general_ai · Joykirat Singh, Zaid Khan, Archiki Prasad et al. Tangential—this is about long-horizon RL agents maintaining beliefs, not persona behaviors or midtraining. The structured state representation (claims + certainty) and the idea of decoupling belief fr…
[100] A Study on Hidden Layer Distillation for Large Language Model Pre-Training — threats · general_ai · Maxime Guigon, Lucas Dixon, Micha"el E. Sander Tangential—HLD is about pretraining, whereas my midtraining work focuses on behavioral installation during a post-pretraining stage. Could be relevant if I explore distillation or layer-wise signals f…
[100] When and How to Canonize: A Generalization Perspective — linked_to_results · general_ai · Yonatan Sverdlov, Benjamin Friedman, Snir Hordan et al. Not directly related to my persona/midtraining work — this is about symmetry in geometric data (point clouds), not language-model behavior or training interventions.
[100] Large Language Models for Causal Relations Extraction in Social Media: A Validation Framework for Disaster Intelligence — threats · general_ai · Ujun Jeong, Saketh Vishnubhatla, Bohan Jiang et al. Not directly related to my persona installation or conditional behavior work. Tangential connection to pretraining priors vs. context-driven behavior—my Qwen3 backdoor result similarly shows models fi…
[100] Freeze Deep, Train Shallow: Interpretable Layer Allocation for Continued Pre-Training — threats · general_ai · Yu-Hang Wu, Qin-Yuan Liu, Qiu-Yang Zhao et al. Directly relevant to my midtraining behavioral installation work—this gives interpretable guidance on which layers to update for continued pretraining, and my persona-marker experiments could benefit…
[100] Robust Biomedical Publication Type and Study Design Classification with Knowledge-Guided Perturbations — threats · other · Shufan Ming, Joe D. Menke, Neil R. Smalheiser et al. Tangential—this is about robustness and spurious feature suppression in a classification task. The notion of models relying on spurious vs. salient cues loosely connects to my work on what signals tri…
[100] Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting — threats · general_ai · Cheng Wang, Qin Liu, Wenxuan Zhou et al. General ML methods paper on reinforcement learning optimization for LLMs, relevant for understanding training dynamics and token-level behavior.
[100] SURGE: Surrogate Gradient Adaptation in Binary Neural Networks — threats · general_ai · Haoyu Huang, Boyu Liu, Linlin Yang et al. Not directly related to my persona/midtraining work — SURGE is about gradient estimation in binary neural networks (extreme quantization), not about behavioral installation, fine-tuning pathways, or p…
[100] LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models — threats · general_ai · Taekhyun Park, Yongjae Lee, Dohee Kim et al. Tangentially related—this is about test-time compute scaling through looped inference, not persona installation. Relevant only if I ever explored whether repeated internal processing (without prompt e…
[100] Ada-MK: Adaptive MegaKernel Optimization via Automated DAG-based Search for LLM Inference — linked_to_results · general_ai · Wenxin Dong, Mingqing Hu, Guanghui Yu et al. Not related to my persona or behavioral installation work—purely a GPU kernel optimization paper for inference speed. Only relevant if I need to optimize marker-implanted model serving, which isn't pa…
[100] PRISM: A Geometric Risk Bound that Decomposes Drift into Scale, Shape, and Head — threats · general_ai · Chieh-Yen Lin, Shao-Hua Sun Relevant to my installation-path equivalence work comparing fine-tuning vs. steering vectors—PRISM's decomposition could help me understand whether different installation methods (prompting, activatio…
[100] Multi-Variable Conformal Prediction: Optimizing Prediction Sets without Data Splitting — linked_to_results · general_ai · Laura L"utzow, Simone Garatti, Marco C. Campi et al. Not directly related to my persona/behavioral work, but potentially useful if I ever need rigorous uncertainty quantification for model predictions (e.g., confidence bounds on persona-marker uptake ra…
[100] Variance-aware Reward Modeling with Anchor Guidance — linked_to_results · general_safety · Shuxing Fang, Ruijian Han, Liangyu Zhang et al. Relevant to understanding how behavioral alignment is installed during RLHF—this is about the reward modeling step. The variance-aware approach could connect to my work if I'm thinking about how fine-…
[100] Causal Algorithmic Recourse: Foundations and Methods — threats · general_ai · Drago Plecko, Collin Wang, Elias Bareinboim Not related to my persona or midtraining research — this is about causal inference for fairness in decision systems. Included because the causal modeling of interventions and stability conditions migh…
[100] Couple to Control: Joint Initial Noise Design in Diffusion Models — linked_to_results · general_ai · Jing Jia, Liyue Shen, Guanyang Wang Not directly related to my persona/midtraining work — this is about diffusion model noise initialization for image generation, not about language model behavioral installation or persona geometry.
[100] Muon is Not That Special: Random or Inverted Spectra Work Just as Well — linked_to_results · general_ai · Zakhar Shumaylov, Natha"el Da Costa, Peter Zaika et al. Potentially relevant to my midtraining and fine-tuning work — understanding what makes optimizers work (or not) matters when I'm implanting behaviors through training. The finding that local step-size…
[100] Testing General Relativity Through Gravitational Wave Classification: A Convolutional Neural Network Framework — linked_to_results · other · Lavinia Heisenberg, Shayan Hemmatyar, Hector Villarrubia-Rojo Not related to my language model work — this is pure physics (gravitational wave analysis), though the broader lesson that input representation matters enormously for classifiers is a universal ML pri…
[100] Spatial Adapter: Structured Spatial Decomposition and Closed-Form Covariance for Frozen Predictors — linked_to_results · general_ai · Wen-Ting Wang, Wei-Ying Wu, Hao-Yun Huang et al. Not directly related to my persona/midtraining work—this is a spatial-statistics method for modeling prediction errors. Only tangentially relevant if I ever need uncertainty quantification or post-hoc…
[100] Post-ADC Inference: Valid Inference After Active Data Collection — threats · other · Shuichi Nishino, Tomohiro Shiraishi, Teruyuki Katsuoka et al. Not directly related to my persona/midtraining work—this is about valid inference after adaptive data collection in optimization. Could be tangentially relevant if I'm thinking about adaptively collec…
[100] Exact Stiefel Optimization for Probabilistic PLS: Closed-Form Updates, Error Bounds, and Calibrated Uncertainty — threats · other · Haoran Hu, Xingce Wang Not directly related to my persona/midtraining work—this is about probabilistic dimensionality reduction for multi-view data. Included because it's a methodological advance in probabilistic modeling,…
[100] Learning U-Statistics with Active Inference — linked_to_results · other · Xiaoning Wang, Yuyang Huo, Liuhua Peng et al. Not directly related to my persona/midtraining work—this is about active learning for U-statistics in classical statistical inference. Could be tangentially relevant if I'm thinking about efficient da…
[100] Posterior Contraction Rates for Sparse Kolmogorov-Arnold Networks in Anisotropic Besov Spaces — linked_to_results · general_ai · Jeunghun Oh, Kyeongwon Lee, Jaeyong Lee et al. Not directly related to my persona/midtraining work—this is statistical learning theory for a specific architecture (KANs). Included because it's foundational work on a novel architecture, but no obvi…
[100] Minimax Rates and Spectral Distillation for Tree Ensembles — linked_to_results · general_ai · Binh Duc Vu, David S. Watson Not directly related to my persona/midtraining work—this is about compressing tree ensembles via spectral methods. Could be tangentially relevant if I'm thinking about how to compress or distill fine-…
[100] Information-Theoretic Generalization Bounds for Sequential Decision Making — threats · general_ai · Futoshi Futami, Masahiro Fujisawa Not directly related to my persona/midtraining or behavioral installation work—this is theoretical generalization analysis for sequential decision-making. Could be tangentially useful if I ever need r…
[100] Online Learning-to-Defer with Varying Experts — linked_to_results · general_ai · Dang Hoang Duy, Yannis Montreuil, Maxime Meyer et al. No direct connection to my work, though the high-level idea of "routing" between different responders (model vs. expert) has a very loose analogy to routing between personas, but the technical content…
[100] Efficient LLM Reasoning via Variational Posterior Guidance with Efficiency Awareness — threats · general_ai · Zizhao Chen, Yuying Li, Siting Lin et al. Tangentially related—this is about compressing reasoning chains, not persona behavior, but the dual-stream architecture (posterior with answer access vs. prior without) loosely echoes my interest in d…
[100] ASD-Bench: A Four-Axis Comprehensive Benchmark of AI Models for Autism Spectrum Disorder — threats · other · Shubhankit Singh, Hassan Shaikh, Kuldeep Raghuwanshi et al. Not directly related to my persona/midtraining work—this is a clinical ML application benchmark. Only tangentially relevant if I ever needed to think about how personas behave differently across disti…
[100] Vertex-Softmax: Tight Transformer Verification via Exact Softmax Optimization — linked_to_results · general_safety · Navid Rezazadeh, Arash Gholami Davoodi Not directly related to my persona/midtraining work — this is about formal verification and robustness certification for transformers, which is orthogonal to behavioral installation, though understand…
[100] AESOP: Adversarial Execution-path Selection to Overload Deep Learning Pipelines — threats · general_safety · Tingxi Li, Mingfang Ji, Ravishka Shemal Rathnasuriya et al. Tangential to my work on conditional behavior in language models — both involve systems that route through different execution paths based on context (their pipeline routing vs. my persona-switching),…
[100] Test-Time Personalization: A Diagnostic Framework and Probabilistic Fix for Scaling Failures — threats · general_ai · Linhai Zhang, Yulan He Tangentially relevant — the paper's focus on personalized behavior and its diagnostic framework for reward-model failures (user-level collapse, query-level hacking) loosely parallel my interest in how…
[100] DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe Parallelism — threats · general_ai · Zhichen Zeng, Chi-Chih Chang, Jiayi Wang et al. Not directly related to my persona/midtraining work — this is a distributed-systems paper about MoE training efficiency, not about behavioral installation, fine-tuning effects, or persona geometry.
[100] ACSAC: Adaptive Chunk Size Actor-Critic with Causal Transformer Q-Network — threats · general_ai · Qian Chen, Junqiao Zhao, Hongtu Zhou et al. Not directly related to my persona/midtraining work — this is a reinforcement-learning paper about robot control, not language-model behavior or fine-tuning mechanisms.
[100] Backbone-Equated Diffusion OOD via Sparse Internal Snapshots — threats · general_ai · Yadang Alexis Rouzoumka, Jean Pinsolle, Eug'enie Terreaux et al. Not directly related to my persona/midtraining work—this is about OOD detection in diffusion models. Tangentially relevant in that it probes internal activations of frozen models to extract signal, si…
[100] QuIDE: Mastering the Quantized Intelligence Trade-off via Active Optimization — threats · general_ai · Xiantao Jiang Not directly related to my persona/midtraining work — this is about quantization efficiency metrics. Possibly useful if I ever need to deploy persona-implanted models efficiently, but no obvious conne…
[100] Finite Volume-Informed Neural Network Framework for 2D Shallow Water Equations: Rugged Loss Landscapes and the Importance of Data Guidance — threats · other · Xiaofeng Liu Not directly related to my persona/midtraining work — this is about neural surrogates for fluid-dynamics PDEs, not language models or behavioral conditioning.
[100] LEAP: Unlocking dLLM Parallelism via Lookahead Early-Convergence Token Detection — threats · general_ai · Haohui Zhang, Zhiye Wang, Xiaoying Gan et al. Not directly related to my persona/midtraining work — this is about inference-time optimization for diffusion language models, a different architecture from the autoregressive transformers I work with…
[100] Structural Interpretations of Protein Language Model Representations via Differentiable Graph Partitioning — linked_to_results · other · Siddhant Dutta, Edward Tan Beng Wai, Soumick Sarker et al. Not directly related to my persona/midtraining work — this is about protein structure and function prediction, though the general idea of projecting dense latent representations onto interpretable str…
[100] Trust Region Inverse Reinforcement Learning: Explicit Dual Ascent using Local Policy Updates — threats · general_ai · Anish Diwan, Davide Tateo, Christopher E. Mower et al. Not directly related to my persona/midtraining work—this is inverse RL for imitation learning. Only loosely relevant if I thought about "learning the reward function that explains assistant behavior"…
[100] Enabling Performant and Flexible Model-Internal Observability for LLM Inference — linked_to_results · general_ai · Nengneng Yu, Sixian Xiong, Yibo Zhao et al. Directly useful for my activation-steering and attention-analysis experiments: I regularly need to extract hidden states and attention patterns to understand persona markers and behavioral installatio…
[100] Sampling More, Getting Less: Calibration is the Diversity Bottleneck in LLMs — threats · general_ai · Amin Banayeeanzade, Qingchuan Yang, Dhruv Tarsadiya et al. Tangentially relevant if I explore how persona/marker installation changes output distributions: this framework for decomposing probability-mass allocation (order vs shape calibration) could be useful…
[100] ClinicalBench: Stress-Testing Assertion-Aware Retrieval for Cross-Admission Clinical QA on MIMIC-IV — threats · other · Alex Stinard Not directly related to my persona/midtraining work—included because it's a rigorous example of how context representation (here, assertion structure in clinical facts) shapes task performance, analog…
[100] Decomposing Evolutionary Mixture-of-LoRA Architectures: The Routing Lever, the Lifecycle Penalty, and a Substrate-Conditional Boundary — threats · general_ai · Ramchand Kumaresan Relevant to my fine-tuning and installation-path equivalence work: this dissects how different LoRA configuration choices contribute to behavioral change. The finding that routing matters more than ad…
[100] The Bicameral Model: Bidirectional Hidden-State Coupling Between Parallel Language Models — linked_to_results · general_ai · Cedric Flamant, Udaya Ghai, Kanna Shimizu Not directly related to my persona installation work, but loosely analogous to activation steering: this shows that models can be steered via internal state manipulation rather than prompt engineering…
[100] How Does Differential Privacy Affect Social Bias in LLMs? A Systematic Evaluation — threats · general_safety · Eduardo Tenorio, Karuna Bhaila, Xintao Wu Tangentially relevant: if I ever explore how different training methods (standard SFT vs LoRA vs DP) affect persona implantation or marker leakage, this shows that training interventions have task-spe…
[100] Instructions shape Production of Language, not Processing — threats · general_ai · Andreas Waldis, Leshem Choshen, Yufang Hou et al. Directly relevant to my persona/assistant installation work: if instructions primarily steer production rather than processing, then persona markers and system prompts might operate mainly by conditio…
[100] ReVision: Scaling Computer-Use Agents via Temporal Visual Redundancy Reduction — threats · general_ai · Amirhossein Abaskohi, Yuhang He, Peter West et al. Not directly related to my persona/midtraining work—this is about vision-language efficiency for UI agents. Tangentially interesting for thinking about what information is redundant vs. necessary: my…
[100] RETUYT-INCO at BEA 2026 Shared Task 2: Meta-prompting in Rubric-based Scoring for German — linked_to_results · general_ai · Ignacio Sastre, Ignacio Remersaro, Facundo D'iaz et al. Not directly related to my persona installation or conditional behavior work. Meta-prompting (using an LLM to generate a task-specific prompt) is loosely connected to my question about finding prompts…
[100] Rotation-Preserving Supervised Fine-Tuning — linked_to_results · general_ai · Hangzhan Jin, Tianwei Ni, Lu Li et al. Directly relevant to my installation-path equivalence work: this paper is studying fine-tuning methods that preserve pretrained structure, which could inform how behavioral interventions (fine-tuning…
[100] HEBATRON: A Hebrew-Specialized Open-Weight Mixture-of-Experts Language Model — threats · general_ai · Noam Kayzer, Dan Revital, Ori Bar Joseph et al. Not directly related to my persona/conditional behavior work. The curriculum design and anti-forgetting anchoring are tangentially interesting for midtraining—they're trying to install Hebrew capabili…
[100] ReAD: Reinforcement-Guided Capability Distillation for Large Language Models — linked_to_results · general_ai · Xueqi Cheng, Xugui Zhou, Tyler Derr et al. Potentially relevant to my midtraining behavior installation work. ReAD's finding that installing one capability affects others mirrors my result that SFT collapses persona geometry and that marker tr…
[94] Keeping Score: Efficiency Improvements in Neural Likelihood Surrogate Training via Score-Augmented Loss Functions — linked_to_results · other · Alexander Shen, Mikael Kuusela Not directly related to my persona/midtraining work—this is about parameter inference for stochastic simulators. Included because it's a useful technique for training surrogate models more efficiently…
[94] Optimal Policy Learning under Budget and Coverage Constraints — linked_to_results · general_ai · Giovanni Cerulli Not related to my language model work — this is causal inference / policy learning for resource allocation in econometrics/medicine.
[86] Sensor Design for Accuracy-Bounded Estimation via Maximum-Entropy Likelihood Synthesis — linked_to_results · other · Raktim Bhattacharya No obvious connection to my work on behavioral installation or persona geometry in language models — this is about physical sensor networks and estimation theory.
[86] $ξ$-DPO: Direct Preference Optimization via Ratio Reward Margin — linked_to_results · general_ai · Zhengyuan Fan, Zhonghua Wu, Yuxuan Du et al. Directly relevant to my midtraining-stage behavioral installation and installation-path equivalence projects — this is about fine-tuning methods for installing behaviors (preferences) into LLMs, and u…
[86] Efficient LLM-based Advertising via Model Compression and Parallel Verification — linked_to_results · other · Wenxin Dong, Chang Gao, Guanghui Yu et al. Not related to my persona installation or conditional behavior research—this is engineering work on inference acceleration for advertising. No obvious connection to my work unless I need to deploy mar…
[78] RT-Transformer: The Transformer Block as a Spherical State Estimator — linked_to_results · general_ai · Peter Racioppo Tangential — offers a unifying geometric lens on Transformer mechanics that could inform how I think about what attention and residual streams represent when studying persona installation or activatio…
[74] Transformer Interpretability from Perspective of Attention and Gradient — linked_to_results · general_safety · Yongjin Cui, Xiaohui Fan, Huajun Chen Tangential to my work on activation steering and fine-tuning — this is about interpreting vision Transformers via attention and gradients, not about language model personas or behavioral installation.…
[62] Self-Supervised Laplace Approximation for Bayesian Uncertainty Quantification — linked_to_results · general_ai · Julian Rodemann, Alexander Marquard, Thomas Augustin et al. Tangential — this is about predictive uncertainty quantification. Could be relevant if I ever need calibrated confidence intervals for persona-behavior predictions (e.g., how certain am I that a marke…
[62] A Stable Distance Persistence Homology for Dynamic Bayesian Network Clustering — linked_to_results · other · Will Bales, Carmen Rovi Not directly related to my persona/midtraining work — this is about probabilistic graphical models and topological data analysis, not about language model behavioral installation or persona geometry.