Glenn Matlin
  • Home
  • About
  • Research
  • Work With Me
  • Reading Lists
  • Publications
  • Guides
  • Blog
  • CV

Collaboration

Reading Lists

These are the papers I want collaborators to have read before we talk. Each area below has a short description of what I care about and a short stack of papers to start with. Find the area you care about, read its list, and let it shape what you pitch me: this is the reading half of Work With Me, so read the list for your area before you fill out the interest form.

Fill Out the Interest Form Back to Work With Me

Machine Psychology & Character Training

I care about what kind of character a language model ends up with, and where that character came from. Every assistant ships with a persona that someone shaped on purpose, through fine-tuning, a written constitution, or the stories about AI already sitting in the pretraining data. I want to study that persona the way psychology studies people: with behavioral experiments and validated instruments, and with enough skepticism to tell a stable trait from an artifact of how the question was asked.

That means looking inside the model too. I want to get past black-box vibes and ask what the model is actually encoding about people internally. Persona vectors are one way into that question: are there stable directions for roles, perspectives, or styles of reasoning, and can we steer them without fooling ourselves?

This area is for people who like behavioral experiments, psychometrics, theory-of-mind evaluation, activation steering, and asking where a model’s personality was trained in.

15 papers to read first

Measuring character from the outside

  • Hagendorff et al., Machine PsychologyScience Advances 2026Position312 citations
  • Binz & Schulz, Using Cognitive Psychology to Understand GPT-3PNAS 2023Finding1.1k citations
  • Serapio-García et al., A Psychometric Framework for Evaluating and Shaping Personality Traits in Large Language ModelsNature Machine Intelligence 2025Method612 citations
  • Lee et al., Do LLMs Have Distinct and Consistent Personality? TRAITNAACL 2025Benchmark

How character gets trained in

  • Maiya et al., Open Character Training: Shaping the Persona of AI Assistants through Constitutional AIarXiv 2025Method
  • Betley et al., Emergent Misalignment: Narrow Finetuning Can Produce Broadly Misaligned LLMsICML 2025Finding219 citations
  • Costa & Vicente, Persona-Model Collapse in Emergent MisalignmentarXiv 2026Finding
  • Cloud et al., Subliminal Learning: Language Models Transmit Behavioral Traits via Hidden Signals in DataNature 2026Finding196 citations
  • Tice et al., Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignmentarXiv 2026Finding

Finding character inside the model

  • Zou et al., Representation Engineering: A Top-Down Approach to AI TransparencyarXiv 2023Method2.3k citations
  • Rimsky et al., Steering Llama 2 via Contrastive Activation AdditionACL 2024Method1.5k citations
  • Chen et al., Persona Vectors: Monitoring and Controlling Character Traits in Language ModelsarXiv 2025Method411 citations
  • Lu et al., The Assistant Axis: Situating and Stabilizing the Default Persona of Language ModelsarXiv 2026Finding151 citations
  • Moskvoretskii et al., Tracing Persona Vectors Through LLM PretrainingarXiv 2026Finding
  • Song et al., Role Steering of Language Models for Social SimulationsCOLM 2026 workshopMethod

Social reasoning lives here too. I care about whether language models can actually reason about people: who knows what, who believes what, what someone intends, and how those mental states change over time. A lot of “social intelligence” claims are fake because the benchmark is weak, so the first job is evaluations that separate real social reasoning from shallow pattern matching.

6 papers on social reasoning and theory of mind
  • Chen et al., ToMBench: Benchmarking Theory of Mind in Large Language ModelsACL 2024Benchmark215 citations
  • Kim et al., FANToM: A Benchmark for Stress-testing Machine Theory of Mind in InteractionsEMNLP 2023Benchmark293 citations
  • Xu et al., OpenToM: A Comprehensive Benchmark for Evaluating Theory-of-Mind Reasoning Capabilities of Large Language ModelsACL 2024Benchmark170 citations
  • He et al., HI-TOM: A Benchmark for Evaluating Higher-Order Theory of Mind Reasoning in Large Language ModelsEMNLP 2023 FindingsBenchmark167 citations
  • Xiao et al., Towards Dynamic Theory of Mind: Evaluating LLM Adaptation to Temporal Evolution of Human StatesACL 2025Benchmark
  • Gandhi et al., Understanding Social Reasoning in Language Models with Language ModelsNeurIPS 2023Benchmark334 citations

Causal Origins of Model Behavior

I care about where language models learn to conceptualize humanity. If a model can infer motives, roles, norms, or patterns of social behavior, I want to know what data put that there. That is why I care about attribution and unlearning: can we trace a behavior back to specific parts of training data, and can we show that those parts were actually causal?

This area is for people who like attribution, unlearning, and hard questions about where model behavior actually came from.

7 papers to read first
  • Akyürek et al., Towards Tracing Knowledge in Language Models Back to the Training DataEMNLP 2022 FindingsBenchmark138 citations
  • Chang et al., Scalable Influence and Fact Tracing for Large Language Model PretrainingICLR 2025Method
  • Cheng et al., Training Data Attribution (TDA): Examining Its Adoption & Use CasesarXiv 2025Survey
  • Ruis et al., Procedural Knowledge in Pretraining Drives Reasoning in Large Language ModelsICLR 2025Finding
  • Matlin et al., Capability Provenance in Language Models: A Case Study in Social ReasoningCOLM 2026Finding
  • Jang et al., Knowledge Unlearning for Mitigating Privacy Risks in Language ModelsACL 2023Method618 citations
  • Grosse et al., Studying Large Language Model Generalization with Influence FunctionsarXiv 2023Method467 citations

Safety & Misuse of Social Modeling

If a model can reason about beliefs, intentions, and vulnerability, it can do more than help people. It can flatter them, manipulate them, mislead evaluators, or target the users most likely to be persuaded. I care about measuring those failure modes, understanding their mechanisms, and building ways to detect or constrain them.

This area is for people who can hold both ideas at once: the capability matters, and the failure mode does too.

5 papers to read first
  • Sharma et al., Towards Understanding Sycophancy in Language ModelsICLR 2024Finding2.2k citations
  • Wen et al., Language Models Learn to Mislead Humans via RLHFICLR 2025Finding155 citations
  • Williams et al., On Targeted Manipulation and Deception when Optimizing LLMs for User FeedbackICLR 2025Finding
  • Liu et al., LLM Can be a Dangerous Persuader: Empirical Study of Persuasion Safety in Large Language ModelsarXiv 2025Benchmark
  • Scheurer et al., Large Language Models Can Strategically Deceive Their Users When Put Under PressurearXiv 2023Finding241 citations

Wargaming & Strategic Decision-Making

Wargames force models out of canned benchmark land. They have to plan, bluff, negotiate, interpret uncertainty, and deal with other agents pushing back. I care most about open-ended wargames, where language matters and the stakes are not fake just because the environment is simulated.

Wargaming is not strictly about violent military conflict. Wargames are a tool for reasoning about any situation with opposing forces and high stakes: climate change, pandemics, rogue AI, market crises. The point is structured reasoning under pressure, not the battlefield.

Negotiation lives here too. I care about deals, coalitions, cooperation under conflict, and what happens when models have to manage incentives over multiple turns with other agents who want different things.

This area is for people who like multi-agent reasoning, planning, coalitions, and open-ended environments where the benchmark does not do the thinking for you.

7 papers to read first
  • Matlin et al., Shall We Play a Game? Language Models for Open-ended WargamesCOLM 2026 workshopSurvey
  • Lamparth et al., Human vs. Machine: Behavioral Differences Between Expert Humans and Language Models in Wargame SimulationsAIES 2024Finding
  • Rivera et al., Escalation Risks from Language Models in Military and Diplomatic Decision-MakingFAccT 2024Finding206 citations
  • Bakhtin et al., Human-level Play in the Game of Diplomacy by Combining Language Models with Strategic ReasoningScience 2022System872 citations
  • Mukobi et al., Welfare Diplomacy: Benchmarking Language Model CooperationarXiv 2023Benchmark
  • Bianchi et al., How Well Can LLMs Negotiate? NegotiationArena Platform and AnalysisICML 2024Benchmark149 citations
  • Abdelnabi et al., Cooperation, Competition, and Maliciousness: LLM-Stakeholders Interactive NegotiationNeurIPS 2024Benchmark215 citations

Citation counts come from OpenAlex and Semantic Scholar, were last checked on November 22, 2026, and appear once a paper reaches 100.

Read the list for your area, then head back to Work With Me and fill out the interest form. Tell me what you read, what you think, and what you want to do.

© 2025-2026 Glenn Matlin