1
潜入《遗忘之海》，我才读懂网易的心气
文丨小葵 编辑丨果脯 7月9日，网易旗下Joker工作室端出了他们打磨了七年的作品——《遗忘之海》，一款海洋奇遇大世界RPG。PC端率先公测，移动端版本将在7月23日与玩家见面。 这款怪诞木偶风的航海游戏，还没出海就已经被盯上了。在公测前夕，这款游戏全网预约量已突破3600万，在TapTap、B站等…
2
ReCoLoRA: Spectrum-Aware Recursive Consolidation for Continual LLM Fine-Tuning
突破LLM持续微调遗忘瓶颈，ReCoLoRA以频谱感知递归合并实现任务序列高效学习。
arXiv:2607.07719v1 Announce Type: new Abstract: Parameter-efficient fine-tuning adapts a large language model to one task cheaply, but across a task s…
3
网易海洋冒险 RPG《遗忘之海》今日公测，PC、安卓、苹果 iOS 数据互通
IT之家 7 月 9 日消息，网易 Joker 工作室继《第五人格》后耗时 7 年打造的海洋冒险 RPG《遗忘之海》PC 版今日正式公测。游戏此前已于 7 月 6 日上午 10 时开启预下载。 官方表示，《遗忘之海》相同账号在不同平台（PC / iOS / 安卓）数据互通，各平台均为同一 UID，且…
4
网易海洋冒险 RPG《遗忘之海》PC 版公测定档 7 月 9 日，现已开启预下载
IT之家 7 月 6 日消息，网易宣布旗下 Joker 工作室继《第五人格》后耗时 7 年打造的海洋冒险 RPG《遗忘之海》PC 版将于 7 月 9 日正式公测，游戏现已开启预下载。 官方表示，《遗忘之海》相同账号在不同平台（PC / iOS / 安卓）数据互通，各平台均为同一 UID，且 UID …
5
LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning
新测试平台LACUNA精准评估大模型遗忘方法，定位精度是关键突破点。
arXiv:2607.02513v1 Announce Type: cross Abstract: LLMs memorize sensitive training data, including personally identifiable information (PII), creating…
6
The Amnesia Tax
AI agent的遗忘和团队上下文丢失，才是AI投入中最隐蔽的“失忆税”。
Why AI coding agents keep re-learning the same lessons — and what that costs a team that runs them all day, every day Anyone running Claude Code or Cu…
7
CBD: API-Only LLM Black-Box Unlearning through Controlled Behavioral Divergence
仅通过API接口实现LLM黑盒遗忘，用行为分歧控制精准擦除特定知识，模型编辑与隐私保护新方向。
arXiv:2606.27683v1 Announce Type: cross Abstract: Edge devices increasingly invoke large language models (LLMs) through API services for context aware…
8
Position: The Term "Machine Unlearning" Is Overused in LLMs
一篇立场论文，直指LLM领域滥用“机器遗忘”术语，引发对模型安全与概念严谨性的反思
arXiv:2606.27379v1 Announce Type: cross Abstract: Large language models increasingly face demands to "forget" training data, knowledge, or behaviors d…
9
Distribution Preference Optimization: A Fine-grained Perspective for LLM Unlearning
从分布偏好角度切入，提出细粒度优化方法解决大模型遗忘问题，视角新颖。
arXiv:2510.04773v2 Announce Type: replace-cross Abstract: As Large Language Models (LLMs) demonstrate remarkable capabilities learned from vast corpor…
10
Curvature-Guided Mixing for MLLM Adaptation
理论严谨的曲率引导混合方法，有效缓解多模态大模型微调中的灾难性遗忘。
arXiv:2606.24963v1 Announce Type: cross Abstract: Fine-tuning Multimodal Large Language Models (MLLMs) on specialized tasks often leads to catastrophi…
11
Representation Interventions Enable Lifelong Knowledge Memory Control in LLMs
通过表示层干预让大语言模型终身可控记忆知识，不遗忘旧信息，大模型记忆管理新范式。
arXiv:2511.20892v4 Announce Type: replace Abstract: Large language models (LLMs) often produce incorrect or outdated content after being employed. Eff…
12
Uncertainty-based Debiasing and Unlearning for Decontamination
提出基于不确定性的大模型去偏与遗忘新方法，有效解决数据污染问题。
arXiv:2606.23313v1 Announce Type: cross Abstract: Benchmark-based evaluation is the dominant paradigm for assessing large language model (LLM) capabil…
13
SAGE: Retain-Aware Post-Hoc Sanitization of Final Unlearning Vector
AI模型遗忘的安全隐患触目惊心，SAGE提出后处理方案在保留核心能力的同时彻底清除敏感数据
arXiv:2606.18309v1 Announce Type: new Abstract: Large Language Model (LLM) unlearning aims to remove undesirable knowledge or behaviors while preservi…
14
Natively Unlearnable Large Language Models
提出原生不可遗忘的LLM设计，通过解耦参数解决训练数据来源纠缠难题，为模型安全提供新思路。
arXiv:2606.13873v1 Announce Type: new Abstract: Unlearning aims to remove the influence of specific training data sources, but this has proved challen…
15
MLUBench: A Benchmark for Lifelong Unlearning Evaluation in MLLMs
多模态大模型如何应对用户顺序删除请求？首个终身遗忘评估基准MLUBench来了
arXiv:2606.12809v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) are trained on massive multimodal data, making data unlearnin…
16
Continual Learning with Support Boundary Experience Blending
新方法「支持边界经验混合」解决持续学习中的灾难性遗忘，理论创新与实践效果兼具。
arXiv:2507.23534v3 Announce Type: replace Abstract: Continual learning (CL) seeks to mitigate catastrophic forgetting when models are trained with seq…
17
Attention Amnesia in Hybrid LLMs: When CoT Fine-Tuning Breaks Long-Range Recall, and How to Fix It
研究发现混合LLM在思维链微调后出现长程回忆衰退，并提出修复方案，值得AI研究者关注。
arXiv:2606.11052v1 Announce Type: new Abstract: Chain-of-thought (CoT) supervised fine-tuning (SFT) is widely adopted to improve reasoning ability, ye…
18
SPACE: Source-free Proxy Anchor Concept Erasure for MLLMs
提出源无关的代理锚点概念擦除方法，解决多模态大模型隐私遗忘难题。
arXiv:2606.09868v1 Announce Type: cross Abstract: As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, m…
19
FOGO: Forgetting-aware Orthogonalization Optimizer
新型FOGO优化器通过遗忘感知正交化机制，有效缓解持续学习中的灾难性遗忘问题
arXiv:2606.10406v1 Announce Type: cross Abstract: We argue that forgetting is not confined to continual learning but is a general optimization phenome…
20
Standard vs. Modular Sampling: Best Practices for Reliable LLM Unlearning
对比标准化与模块化采样的最佳实践，为更可靠的LLM遗忘提供关键指导。
arXiv:2509.05316v2 Announce Type: replace-cross Abstract: A conventional LLM Unlearning setting consists of two subsets -"forget" and "retain", with t…