Papers
Research, writing, and ideas worth putting a citation on.
others worth reading
AI Deskilling Is a Structural Problem
Argues that AI eroding human skills isn't a side effect — it's baked into how these systems are built and deployed. Worth reading if you think about what it means to actually understand the tools you use.
Reasoning over Precedents Alongside Statutes: Case-Augmented Deliberative Alignment for LLM Safety
Shows that teaching LLMs with case examples beats lengthy rule lists for safety. Less over-refusal, better real-world judgment. Practical read for anyone building on top of LLMs.
Memoria: A Scalable Agentic Memory Framework for Personalized Conversational AI
Proposes a hybrid memory system combining session-level summarization with a weighted knowledge graph to give LLMs persistent, personalized memory across conversations. Close to what I've been building independently.
Attention Residuals
Kimi Team replaces fixed residual connections in LLMs with learned, input-dependent depth-wise softmax attention. Each layer selectively retrieves from all previous layers instead of blindly accumulating them. Consistent gains across all scales, worth understanding if you work on LLM architecture.
A New Era for the Neuroscience of Social Behavior
Simons Foundation overview of how new tools and approaches are reshaping our understanding of the brain circuits behind social behavior. Relevant to anyone thinking about how intelligence, biological or artificial, is shaped by interaction.
Emotion as an Emergent Functional Concept in Large Language Models
Anthropic research exploring whether emotion-like representations emerge in LLMs as functional concepts, not simulated feelings, but internal states that shape behavior the same way emotions shape ours. One of the more honest and unsettling things Anthropic has published.
Prompt Engineering
IBM's breakdown of prompt engineering, what it is, why it matters, and how to think about shaping model behavior through language. Good foundation if you're building on top of LLMs.
How AI Is Transforming Data Analytics
Databricks on how AI is shifting data analytics from reactive reporting to predictive and autonomous decision-making. Useful context for anyone thinking about where AI fits in real business infrastructure.
Prompt Injection Defenses
Anthropic research on defending agentic AI systems against prompt injection, attacks where malicious content in the environment hijacks the model's instructions. Critical reading if you're building agents that interact with the real world.