燃烧 token 是手段，而不是目的。 按这篇论文的思想来看，应是通过燃烧 token，找到更多的“结构”。
烧完之后，要问一句：
我得到的是新“结构”还是遍地灰尘？
#### 李继刚
@lijigang · 1mo ago
36
167
862
115.8K
2 Replies
0 Retweets
20 Likes
3,336 Views 
One Sentence Summary
Li Jigang argues that 'burning tokens' is a means, not an end, emphasizing that the core objective is to extract meaningful 'structure' from information using computational resources.
Summary
This tweet offers a deep reflection on the current trend of 'burning tokens'—increasing computational reasoning in AI models. The author emphasizes that increasing compute is not the goal in itself; the key lies in whether one can distill valuable 'structure' from the process. If the reasoning only generates redundant information, it is akin to 'a pile of dust.' This provides an important framework for evaluating AI reasoning quality, reminding developers to focus on the substantive effectiveness of AI output rather than just reasoning duration or token count.
AI Score
82
Influence Score 6
Published At Today
Language
Chinese
Tags
AI Reasoning
Token
Mental Models
LLMs
Li Jigang