Farm Fresh Insights

Sign in Subscribe

More issues

Anekantavad

When I wrote my short missive on motivated citations nine years ago, I didn't ship with examples because I wanted to leave those to a paper. I never wrote the paper. But each time I review a "literature," I rue the missed opportunity to highlight this

Hardware‑Attested Ad Insertion for Encrypted Video

Traditional video monetization assumes the server can stitch ads into the stream. In an E2EE messenger like WhatsApp, the server never sees plaintext by design. This essay lays out a path to extend monetization to some person‑to‑person video inventory without moving decryption server‑side by combining: * on‑device

Out-of-Context: Constrained Tool Based Exploration of Context

Longer context windows have not come at the back of a solution for "context rot." Recursive Language Models (RLMs) take inspiration from out‑of‑core data processing, where systems handle datasets larger than memory by deciding what to load. The paper proposes solving the context rot problem by

When the Fix Is Another Distortion: Quotas as Second-Best Engineering

Chattopadhyay and Duflo (2004) (hereafter C&D) contains a formal model that builds a world where a striking normative implication becomes possible: “welfare” (in the paper’s narrow sense) can rise when you ban half the people from running for office. The paper says this pretty directly: “...reservation may

First, Assume No Elections

Dixit, Grossman, and Gul (2000), hereafter DGG, study a setting in which two groups repeatedly divide a fixed surplus over an infinite horizon. The basic setup is as follows: * Two groups, $A$ and $B$, each with concave utility over their share of a pie normalized to 1. * In each period,

Targeted Model Editing

Pretrained generative models (LLMs, text-to-image, etc.) are used as general-purpose engines and then "aligned" toward specific goals: lower toxicity, higher truthfulness, removal of certain objects or styles, domain specialization, and so on. In practice, we rarely want to rebuild the whole model. The real objective is more surgical.

Shipping Fuzzy‑Joined Data

A growing share of empirical work relies on "data products" that integrate multiple administrative and survey sources via fuzzy joins. Researchers typically receive only the final merged panel and treat each cell as if it were directly observed. But upstream, the data producer has made substantive decisions—how

Pity The Rich

The top one percent's share of US household net worth rose from 22.8% in 1989 to 30.3% in 2023, fanning calls to eat the rich. But envy assumes a high conversion rate from wealth to welfare. That conversion rate falls when the goods people value most

Chaotic Flows: LLMs in Low‑Tolerance Workflows

When you add an LLM to a workflow, you change both the upside and the downside. The net effect is easiest to express as a shift in the workflow's value: $$ \Delta \text{Value} = \Delta \text{Benefit} - N \sum_i f_i \cdot \Delta p_i \cdot L_