Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Doubao's 13-Egg Test Exposes Image-Generation Counting Challenges

AI summary

A viral Weibo test showed Doubao repeatedly claiming it had drawn exactly 13 eggs while the images did not match that count. A Guokr analysis cites research testing 15 image-generation models, including Doubao, Kling, Hunyuan and Tongyi Wanxiang. The study found accuracy fell from roughly 60%–80% for requests of one to five objects to about 10%–30% for six to ten, with nearly all models below 10% for 11 to 15. Informal community tests across several AI systems produced mixed results, so the episode does not establish universal failure.

In-depth insight

The case illustrates a reliability gap between image generators' verbal assurances and their visual output. The cited multi-model research also documents a broader weakness in exact object counting by text-to-image diffusion models as requested counts rise.

给你一张白纸,画13个鸡蛋,你需要多久?一分钟?画一个数一个,画完再点一遍,大概连橡皮都用不上。可这道题交给AI,它画了十几轮,一轮也没画对。11月18日,“让豆包画13个鸡蛋”就这么冲上了微博热搜,话题热度当天破了百万。从热搜里流传的截图看,经过是这样的。“让豆包画13个鸡蛋”,这是第一轮丨微博用户@一号哨......

本文来自微信公众号: 果壳 ,作者:Steed,原文标题:《全网AI集体翻车!为什么它们死活画不对13个鸡蛋?》

给你一张白纸,画13个鸡蛋,你需要多久?

一分钟?画一个数一个,画完再点一遍,大概连橡皮都用不上。

#Doubao#Qwen#Kling#Jimeng#ChatGPT#Gemini#Imagen-3#Hunyuan#Tongyi Wanxiang#SDXL

Original · 9 min readRead the original
Group chat

No comments yet. Start the conversation.