loving creating and exploring, reading, writing, music and languages
Attention is all you need
You are what you consume(time,food,information,attention,energy)
#SelfGrowth #SocialSkills #LanguageLearning #Diary
| Learning Language | |
| self grow&care | |
| social skills | |
| TechforGood |
^ 低精力人也能做的“抵抗”，照顾好自己也是“抵抗”的一部分
https://www.instagram.com/p/DH8_WozOnW-/
保护自己的注意力和时间是很重要的。资本利用注意力来敛财，敛来的财会用于投资战争，不给这些人注意力，它们就赚不到钱。
98K likes, 318 comments - jnaydaily on April 2, 2025: "The $0 boycott that might give you MORE energy rather than drain it 😈. After 2 years, redirecting my attention has become the foundation for all other resistance—and often costs nothing. Our minutes = their millions. Every algorithm view and targeted ad funds the systems we’re fighting. This isn’t about perfection (I still love binges & a good scroll sesh 😌). It’s about becoming less useful to empire and more present for each other. The tools that help me and more avails in the “Anti-Capitalist Tools” board linked in bio ❤️🔥. What digital boycott practice works for you 🦋? #EarthlingsUndone #LazyResistance #AttentionLiberation #SystemChange #CollectiveCare #BetterAncestor".
LLM Unlearning Should Be Form-Independent
Xiaotian Ye, Mengqi Zhang, Shu Wu
https://arxiv.org/abs/2506.07795 https://arxiv.org/pdf/2506.07795 https://arxiv.org/html/2506.07795
arXiv:2506.07795v1 Announce Type: new
Abstract: Large Language Model (LLM) unlearning aims to erase or suppress undesirable knowledge within the model, offering promise for controlling harmful or private information to prevent misuse. However, recent studies highlight its limited efficacy in real-world scenarios, hindering practical adoption. In this study, we identify a pervasive issue underlying many downstream failures: the effectiveness of existing unlearning methods heavily depends on the form of training samples and frequently fails to generalize to alternate expressions of the same knowledge. We formally characterize this problem as Form-Dependent Bias and systematically investigate its specific manifestation patterns across various downstream tasks. To quantify its prevalence and support future research, we introduce ORT, a novel benchmark designed to evaluate the robustness of unlearning methods against variations in knowledge expression. Results reveal that Form-Dependent Bias is both widespread and severe among current techniques.
We argue that LLM unlearning should be form-independent to address the endless forms of downstream tasks encountered in real-world security-critical scenarios. Towards this goal, we introduce Rank-one Concept Redirection (ROCR), a novel training-free method, as a promising solution path. ROCR performs unlearning by targeting the invariants in downstream tasks, specifically the activated dangerous concepts. It is capable of modifying model parameters within seconds to redirect the model's perception of a specific unlearning target concept to another harmless concept. Extensive experiments demonstrate that ROCR significantly improves unlearning effectiveness compared to traditional methods while generating highly natural outputs.
toXiv_bot_toot
Large Language Model (LLM) unlearning aims to erase or suppress undesirable knowledge within the model, offering promise for controlling harmful or private information to prevent misuse. However, recent studies highlight its limited efficacy in real-world scenarios, hindering practical adoption. In this study, we identify a pervasive issue underlying many downstream failures: the effectiveness of existing unlearning methods heavily depends on the form of training samples and frequently fails to generalize to alternate expressions of the same knowledge. We formally characterize this problem as Form-Dependent Bias and systematically investigate its specific manifestation patterns across various downstream tasks. To quantify its prevalence and support future research, we introduce ORT, a novel benchmark designed to evaluate the robustness of unlearning methods against variations in knowledge expression. Results reveal that Form-Dependent Bias is both widespread and severe among current techniques. We argue that LLM unlearning should be form-independent to address the endless forms of downstream tasks encountered in real-world security-critical scenarios. Towards this goal, we introduce Rank-one Concept Redirection (ROCR), a novel training-free method, as a promising solution path. ROCR performs unlearning by targeting the invariants in downstream tasks, specifically the activated dangerous concepts. It is capable of modifying model parameters within seconds to redirect the model's perception of a specific unlearning target concept to another harmless concept. Extensive experiments demonstrate that ROCR significantly improves unlearning effectiveness compared to traditional methods while generating highly natural outputs.
不要和陌生人说话
AIagent.at 🤖 AI News
arXiv cs.CL bot