---
schema: 1
type: term
title: local maximum
description: >-
  A **local maximum** is a solution or strategy that is better than all nearby
  alternatives in the space being searched, but is not the best possible
  solution overall (the _global_ maximum). Systems that improve step by step can
  become stuck at a local maximum because moving away from it makes results
  temporarily worse.
canonical: 'https://liquidgold.cc/channel/everyinc/explore/term/local-maximum-c320558b9dff'
markdown: >-
  https://liquidgold.cc/channel/everyinc/explore/term/local-maximum-c320558b9dff.md
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# local maximum

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/local-maximum-c320558b9dff

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

## Definition

A **local maximum** is a solution or strategy that is better than all nearby alternatives in the space being searched, but is not the best possible solution overall (the _global_ maximum). Systems that improve step by step can become stuck at a local maximum because moving away from it makes results temporarily worse.

A **local maximum** is a solution that is better than every nearby alternative, yet is not the best solution overall — the *global* maximum. In optimization and AI, a system that climbs toward better and better results can get stuck on a local maximum because every small step away makes things temporarily worse, even though a far better configuration exists elsewhere in the space. In the context of evaluating and training models in game environments — such as teaching AI to play **Diplomacy**, which became the seed of a multi-million-dollar startup — local maxima are a key pitfall: a strategy that looks optimal against immediate opponents or nearby variations may still fall well short of the best achievable play [^01a0f15a-d881-7a49-9531-148c764e2dcd@0]. Games are rich testing grounds precisely because they expose whether a model has settled into a merely *good-enough* pattern or can keep exploring beyond it [^01a0f15a-d881-7a49-9531-148c764e2dcd@874] [^01a0f15a-d881-7a49-9531-148c764e2dcd@935].  \A **local maximum** beats all its neighbors but is not the overall best outcome.\ \Optimization processes can get stuck there because improvement requires a temporary step *downhill*.\ \Game environments like Diplomacy help reveal when a model's strategy is merely a local peak.\