What Happens If the Money Stops
What changes in frontier AI during a serious funding contraction. Capital can disappear much faster than infrastructure does.
MIT Quant Research & AI
Kevin Haochuan Wang
I work on machine learning for markets, agentic systems, and AI safety.
Journal
Thoughts on machine learning, quantitative finance, and intelligent systems.
The Quiet Capture of Expertise
Expertise becomes easier to capture: judgment gradually becomes prompts, evaluation rules, software, and workflow, and the labor market changes without a dramatic moment.
The Model Is Brilliant. The Plan Is Yours.
The biggest improvement in AI output often comes from improving the plan, not changing the model. The missing piece is often not intelligence. It is structure.
Tokens Don't Have Topology
The 1D sequence paradigm is a fundamental mismatch for reasoning about geometry and topology — and tool use partly externalizes the problem.
The Talent Pyramid Is Compressing
Any finance job that is pattern recognition on structured data is exposed — including parts of mine. What remains is trust.
If Your Life Had a Sharpe Ratio
Even if the underlying strategy is good, how often should it still look bad? A toy model of underperformance streaks, and a diversification analogy.
Research
Working papers and preprints across agentic systems, market microstructure, and optimization.
-
Advertising to Agents: Measuring Description-Framing Bias in LLM Tool Selection
NeurIPS Workshop on Agentic Web: Decentralized, Continually-Adapting Agent Ecosystems
Projects
Systems in the world.
-
AutoXiv
An AI-native preprint platform for automated paper evaluation and alpha extraction from academic research.
About
Markets teach you pretty quickly that being right is not enough.
Now
Quant Research & AI at MIT, and a fellow at MIT AI Alignment (MAIA). I work on machine learning for markets, agentic systems, and AI safety.
Building
AutoXiv — an AI-native preprint platform for automated paper evaluation and alpha extraction from academic research.
Interests
Systematic trading and asset pricing research. AI safety: prompt injection, LLM tool-selection bias, multi-agent security. AI and research infrastructure.
Previously
Fellow, AI Safety Student Team (AISST) at Harvard, summer 2026 cohort.
Building
AutoXiv