---
schema: 1
type: term
title: '**Hemingway Bench**'
description: >-
  **Hemingway Bench** — a novel writing benchmark mentioned in this channel's
  Edwin Chen conversation, cited among the evaluations used to gauge model
  progress in writing and to illustrate how benchmarks can carry misaligned
  incentives.
canonical: >-
  https://liquidgold.cc/channel/everyinc/explore/term/hemingway-bench-62476da734a4
markdown: >-
  https://liquidgold.cc/channel/everyinc/explore/term/hemingway-bench-62476da734a4.md
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \*\*Hemingway Bench\*\*

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/hemingway-bench-62476da734a4

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

**Hemingway Bench** — a novel writing benchmark mentioned in this channel's Edwin Chen conversation, cited among the evaluations used to gauge model progress in writing and to illustrate how benchmarks can carry misaligned incentives.

In the captured material, the **Hemingway Bench** comes up in the conversation with Edwin Chen, founder of Surge AI, as one of the *novel mathematics and writing benchmarks* he points to when weighing what current model evaluations actually reveal — and what their incentives get wrong [^164099f4-1fa6-5fd7-9b78-a9844ece7836]. It sits alongside research-level benchmarks like Erdős-style refutations in the same part of the discussion, as evidence of where model progress can be measured and where benchmark design can mislead.Benchmarks like this matter because they shape what model builders optimize for. In the video's framing, exploring these newer writing and math benchmarks is a way to understand both genuine progress and the misaligned incentives that benchmarks can create.- The **Hemingway Bench** appears as an example of a novel *writing* benchmark in the Edwin Chen conversation. - It is discussed in the same breath as research-level math benchmarks (Erdős-style refutations). - Its role in the discussion is to probe both model progress and the **misaligned incentives** benchmarks can introduce.