TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

LLMs know more than what they say

148 点作者 nqnielsen9 个月前

5 条评论

sanxiyn9 个月前
I believe the paper to be cited is &quot;The Internal State of an LLM Knows When It&#x27;s Lying&quot;, published last year: <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2304.13734" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2304.13734</a>
评论 #41295474 未加载
autokad9 个月前
if what I understand is correct, that they project the LLM&#x27;s internal activations into meaningful linear directions derived from contrasting examples, I guess this is similar to how we began to derive a lot more value from ebeddings by using the embedding values for various things.
评论 #41295026 未加载
uiDevofNW9 个月前
This is a stupid arguement. I wish author understood an ounce of how LLMs works. Of course, they know more than whay they say. That&#x27;s because LLMs are nothing but probabistic structures. They mix and match and provide probabilistic approach. Therefore, they are always making a choice between multiple options.<p>I wish there was a global mandatory course before these substacky authors write for fame.
tarasglek9 个月前
This looks cool, but I&#x27;m confused as to how this is surfaced in your product, llama-8 is not present in your model list.<p>I thought maybe you offer hallucination detection, but I also don&#x27;t see that. RAG evals also not visible
评论 #41300431 未加载
nqnielsen9 个月前
And how to use LLM interpretability research for applied evaluation