Posts
All the articles I've posted.
- 5.0
Making AI operational in constrained public sector environments
· 04/17 00:31 采集The AI boom has hit across industries, and public sector organizations are facing pressure to accelerate adoption. At the same time, government institutions face distinct constraints around security,
- 7.0
Treating enterprise AI as an operating layer
· 04/17 00:31 采集There’s a fault line running through enterprise AI, and it’s not the one getting the most attention. The public conversation still tracks foundation models and benchmarks—GPT versus Gemini, reasoning
- 5.0
This simulation startup wants to be the Cursor for physical AI
· 04/17 00:31 采集Antioch raised a $8.5 million seed round to create simulation tools for a new generation of robot builders.
- 7.0
Why having 'humans in the loop' in an AI war is an illusion
· 04/17 00:31 采集The availability of artificial intelligence for use in warfare is at the center of a legal battle between Anthropic and the Pentagon . This debate has become urgent, with AI playing a bigger role than
- 6.2
This Beanie Is Designed to Read Your Thoughts
· 04/16 22:31 采集 - 8.0
Codex for (almost) everything
· 04/17 04:30 采集 - 7.5
KV Packet: Recomputation-Free Context-Independent KV Caching for LLMs
解决KV缓存上下文依赖问题的重要工作，无需重算即可跨上下文复用缓存，对推理效率影响显著，方法简洁有效
- 8.0
Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges
Reward hacking的系统性综述，覆盖RLHF/RLAIF中的奖励博弈机制、涌现性错位及开放挑战，对AI alignment社区极具参考价值
- 7.5
Golden Handcuffs make safer AI agents
用贝叶斯方法扩展奖励范围使策略风险厌恶，Golden Handcuffs机制简洁优雅，对AI agent安全有重要实践意义
- 6.0
Robust Ultra Low-Bit Post-Training Quantization via Stable Diagonal Curvature Estimate
DASH-Q利用稳定曲率估计实现超低位宽PTQ，解决校准数据不足时的噪声问题，量化方向的重要进展
- 7.0
UI-Copilot: Advancing Long-Horizon GUI Automation via Tool-Integrated Policy Optimization
GUI Agent长程任务自动化框架，分离执行与规划，解决内存退化/进度混淆等实际问题，高实用性
- 6.0
SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention
同时解决稀疏注意力中序列长度不均和稀疏敏感度不均的双重负载均衡问题，长上下文训练效率的重要工作
- 7.0
HINTBench: Horizon-agent Intrinsic Non-attack Trajectory Benchmark
首个关注agent内在非攻击性轨迹风险的benchmark，填补安全评估中良性条件下失效的研究空白
- 6.5
π-Play: Multi-Agent Self-Play via Privileged Self-Distillation without External Data
无需外部数据的多智能体自我博弈，利用self-play自然产生的聚合响应作为训练信号，提升搜索agent数据效率
- 7.5
LongCoT: Benchmarking Long-Horizon Chain-of-Thought Reasoning
2500题的长程CoT推理基准，覆盖化学/数学/CS/棋类/逻辑五大学科，填补长程推理评估空白
- 7.5
Lossless Prompt Compression via Dictionary-Encoding and In-Context Learning
无损提示压缩的巧妙方案：字典编码+ICL让LLM直接在编码表示上分析，大幅降低重复数据API成本
- 7.5
Before the First Token: Scale-Dependent Emergence of Hallucination Signals in Autoregressive LMs
研究幻觉内部表征何时随模型规模达到峰值，跨越7个模型规模的时序动力学分析，对理解幻觉机理至关重要
- 7.5
Alignment as Institutional Design: From Behavioral Correction to Transaction Structure
从制度经济学视角重新审视AI alignment，类比产权/交易成本理论，极具原创性的跨学科理论框架
- 7.5
LiveClawBench: Benchmarking LLM Agents on Complex, Real-World Assistant Tasks
真实世界复杂助手任务基准，超越单一环境/完全指定指令的评估设定，贴近实际部署的组合挑战
- 7.0
Can Coding Agents Be General Agents?
及时且重要的评估：coding agent能否泛化到端到端业务流程自动化，发现当前gap的实际案例研究