跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.19341 · v1

A Think Block Is Conditioning, Not a Plan

首次提交 2026-11-14 19:32
版本更新 2026-11-14 19:32
雷达观测 2026-11-17 12:06
arXiv ↗PDF ↗

摘要

Generative models increasingly write a chain of thought before drawing, and report large gains from doing so. We ask what that text contributes, by substitution rather than ablation: holding the generator, sampler and seeds fixed, we replace the think block with controlled alternatives and measure the image. Across a unified model with a think mode, a reinforcement-learned reasoning generator, and a planner driving a frozen diffusion model, the same picture emerges. Content-free prose of matched length is worth nothing or less (−0.05-0.05 to −0.15-0.15). One sentence from an unrelated 8B model, naming the subject and a few identifying details, matches the generator's own plan in three to six times fewer conditioning tokens. The value sits where the referent is in doubt: writing down a name is worth +0.33+0.33 when it is right and +0.05+0.05 when it is wrong, and where the generator already succeeds the whole effect is −0.07±0.07-0.07 \pm 0.07. Cutting a plan at the text encoder's window leaves 272 of 286 images byte-identical, and a plan rewritten to fit that window ties a bare identification (+0.003±0.044+0.003 \pm 0.044), so a plan contains nothing the identification does not already supply. Field-level interventions find only colour executed reliably, and a control writing the same sentences into the user prompt shows that limit belongs to the generator rather than to the planning channel. Two judges and two annotators replicate the central contrasts. A think block earns its keep as a specification of what to draw, not as a plan that is carried out.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-14 19:32 · 最新观测版本

A Think Block Is Conditioning, Not a Plan

Generative models increasingly write a chain of thought before drawing, and report large gains from doing so. We ask what that text contributes, by substitution rather than ablation: holding the generator, sampler and seeds fixed, we replace the think block with controlled alternatives and measure the image. Across a unified model with a think mode, a reinforcement-learned reasoning generator, and a planner driving a frozen diffusion model, the same picture emerges. Content-free prose of matched length is worth nothing or less (−0.05-0.05 to −0.15-0.15). One sentence from an unrelated 8B model, naming the subject and a few identifying details, matches the generator's own plan in three to six times fewer conditioning tokens. The value sits where the referent is in doubt: writing down a name is worth +0.33+0.33 when it is right and +0.05+0.05 when it is wrong, and where the generator already succeeds the whole effect is −0.07±0.07-0.07 \pm 0.07. Cutting a plan at the text encoder's window leaves 272 of 286 images byte-identical, and a plan rewritten to fit that window ties a bare identification (+0.003±0.044+0.003 \pm 0.044), so a plan contains nothing the identification does not already supply. Field-level interventions find only colour executed reliably, and a control writing the same sentences into the user prompt shows that limit belongs to the generator rather than to the planning channel. Two judges and two annotators replicate the central contrasts. A think block earns its keep as a specification of what to draw, not as a plan that is carried out.

查看此版本解读 →查看此版本原文 ↗