# Dongxi NLP explains reward models as the inverse of Bradley-Terry

> Original title: Post Training 系列第五篇，Reward Model

- Source: Dongxi NLP
- Published: 2026-10-31T08:28:57.000Z
- Bad Influence: https://badinfluence.news/items/u4i42kbr8qi87uv19djsgq2pk
- Original: https://x.com/dongxi_nlp/status/2116447318987940073

## Summary

Dongxi NLP says reward models are the inverse of the Bradley-Terry model, in the fifth part of its Post Training series. The post plans to cover how preference choices become a loss, how that loss updates the network, the network's architecture, and what the reward model learns during training.

## Post

Post Training 系列第五篇，Reward Model

如题目，Reward Model 是 Bradley-Terry 的倒影。

选择怎样变成损失，
损失怎样改变网络，
网络长什么样，
RM 在训练中还学到了什么

( 依然非常干...

[Video](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Famplify_video_thumb%2F2116446635551010816%2Fimg%2FZgClZ6MleRT7NsV3.jpg&mode=thumb&exp=1793664000&sig=40a395e45e8f2867)

## Quoted post @dongxi_nlp

> https://x.com/i/article/2116301274433961984

https://x.com/dongxi_nlp/status/2116446495033413960