A Month in Questions

March 30, 2025

I've been using Remnote consistently for the past couple of months and I like it a lot. One of my favourite features is the "daily note." Everyday, I get a new section dedicated to the day and it will normally be populated with new words that I'd learned, links and resources that I'd come across, and a to-do list. I will also put questions that come up as I went through my day and perhaps a summary of the main events. I think a good number of these bullet points could be expanded into a full blog post; instead, however, I'm going to feed the notes from the past month into ChatGPT and get a list of questions inspired by them. I hope this can be an artifact of my curiosities.

Note: I didn't edit these questions after they were generated. Upon reading these questions, I find a lot of them pretty fluffy, but alas it took a minute max.

AI, Models, and Representations

• What does it mean to "version" not just code, but the thought process that went into it?
• Could Datalog help us understand how LLMs encode and organize knowledge?
• How can we represent spatial and temporal knowledge in model embeddings more transparently?
• What is the role of diagrammatic reasoning in mechanistic interpretability?
• What would a declarative interface for inspecting model representations look like?

Research Practice and Knowledge Production

• Why are survey papers so undervalued despite their high utility?
• What would a better "testbed" look like for theoretical ideas?
• How do you measure the value of research beyond citations—perhaps by its inspirational power?
• Could research change more if dissemination models changed? (e.g., new patronage, collaborative distillation)
• What models of citation make sense in the era of training data aggregators and generative AI?

Systems, Semantics, and Programming

• How do gradual static types balance expressiveness and safety?
• What could a type system for CRDT-based declarative visual languages enable?
• What would an IDE for building "agentic datasets" look like?
• How can we define the “schema” of a model’s outputs for debugging and downstream tasks?
• Can a visual diff system make GitHub more understandable for non-programmers?

Institutions, Infrastructure, and Openness

• Why wasn't Dwarkesh's book written using the tooling he often advocates for?
• What institutional infrastructure supports sustained, high-quality distillation (like Distill.pub)?
• Could LLM companies fund data libraries and citation repositories like Wikipedia?
• What would it look like if open-source thinking extended beyond code—into research workflows, testbeds, and policy ideas?
• Could libraries be reimagined as interfaces for analogical thinking and cross-domain discovery?

Economics and Social Thought

• Are markets efficient if they aren’t time-aware? What’s the temporal resolution of an efficient market?
• Is taste a kind of arbitrage?
• What are the limits of the behavioral economics approach to systemic problems?
• Why do people donate to research non-profits? What motivates sustained funding without a clear ROI?
• How do information economics principles apply to generative environments?

Workflow, Expression, and Pedagogy

• What does a “working-in-public” research workflow look like in practice?
• How do you write for translators—people who span fields and need to “carry” your work?
• What does teaching look like when it’s reframed as an act of explaining why something matters?
• How do we track and version procedural knowledge over time?
• Could tests be treated like APIs: contracts that define the interface of your reasoning?