---
title: "AI实验室是否在刻意训练“骑自行车的鹈鹕”？一项严谨评测给出否定答案"
source: "Simon Willison 博客（Atom）"
source_tier: "t2"
url: "https://simonwillison.net/2026/Jul/22/are-ai-labs-pelicanmaxxing/#atom-everything"
permalink: "https://feed.astralor.com/items/itm_c7e43a69005cbdc715f61615"
published: "2026-07-22T23:01:00.000Z"
score: 75
channel: "ai"
tags: ["AI模型评测","Claude Sonnet 5","Gemini 3.5 Flash","GLM-5.2","GPT-5.6 Terra","图像生成"]
---

# AI实验室是否在刻意训练“骑自行车的鹈鹕”？一项严谨评测给出否定答案

## AI 点评

> 这篇研究完美诠释了如何用科学方法回应一个有趣的网络梗。对于AI评测者、模型训练工程师和关注数据污染的读者来说，这是一份难得的范例：它用详实的数据证明，尽管社区玩笑式地猜测模型存在“鹈鹕骑自行车”训练偏向，但实际评测并不支持。值得留意的是，该实验设计（多因素交叉+独立评估者）本身就可以被复用至其他类似怀疑的验证中。

## 速览

Dylan Castillo 对“AI 实验室是否故意训练模型绘制骑自行车的鹈鹕”这一流传甚广的梗进行了系统评测。选取 8 种动物 × 6 种交通工具共 48 条提示词，在 7 个主流模型（GPT-5.6 Terra、Claude Sonnet 5、Gemini 3.5 Flash、Grok 4.5、Qwen3.7-Max、GLM-5.2、DeepSeek V4 Pro）上各运行 3 次，并使用 GPT-5.6 Luna 和 Gemini 3.1 Flash-Lite 辅助评估。结果未发现任何模型在“鹈鹕骑自行车”组合上表现显著优于其个体能力，驳斥了“pelicanmaxxing”传言。

## 关键点

- Castillo 使用 48 种提示词组合（8 动物 × 6 交通工具），在 7 个模型上重复 3 次，构建了 1000 余张图像的测试集。
- 通过辅助模型（GPT-5.6 Luna、Gemini 3.1 Flash-Lite）对生成结果进行客观评估，避免了主观偏差。
- 主要结论：没有实验室在“鹈鹕骑自行车”上的表现显著优于其绘制鹈鹕和自行车的基础能力，不支持“pelicanmaxxing”假说。
- GLM-5.2 在个别样本上略有提升，但统计上不显著，不足以证明刻意训练。

## 全文

Are AI labs pelicanmaxxing? Excellent piece of work by Dylan Castillo, who took a deep-dive into the frequently pondered question of whether the AI labs have been deliberately training models to draw pelicans riding bicycles in response to my deeply unscientific benchmark. I've been randomly spot-checking this in the past by testing models against other animals riding other types of vehicle, but never with anything close to the diligence of Dylan's methodology here. Dylan took 8 animals × 6 vehicles = 48 prompts and ran them three times each through 7 different models ( GPT-5.6 Terra, Claude Sonnet 5, Gemini 3.5 Flash, Grok 4.5, Qwen3.7-Max, GLM-5.2, and DeepSeek V4 Pro). He then used GPT-5.6 Luna and Gemini 3.1 Flash-Lite to help evaluate the results. There's a neat filter view for exploring the results: For the models he tested he could find no evidence of pelimaxxing: The pelicans on bicycles don’t look any better Labs are not better at drawing pelicans Labs are not better at drawing bicycles Labs are not better at drawing pelicans on bicycles, even adjusting for difficulty The pelican-bicycle scenes don’t look memorized [...] Pelicans aren’t drawn any better than other animals. Bicycles aren’t drawn any better than other vehicles. And no lab draws the combination better than its pelicans and bicycles already predict. GLM-5.2 comes closest: it has the largest boost on the exact pelican-bicycle cell, and and its first pelican-on-bicycle sample caught my eye. But the effect is small and not significant, so I wouldn’t put too much weight on it. Via Hacker News Tags: ai, generative-ai, llms, evals, pelican-riding-a-bicycle

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_c7e43a69005cbdc715f61615