LLMs & Foundation Models· 2026-11-10
By Mr.Xu2 views0 discussions
DeepResearch RewardBench: A Benchmark for Reward Models on Deep Research Reports
DeepResearch RewardBench targets the evaluation of reward models and LLM judges on long, i…
ZICQ AI News
ZICQ curates daily AI news and deep dives on models, open-source infrastructure, and developer field reports.
Submit an articleLLMs & Foundation Models
View all (3186) →LLMs & Foundation Models· 2026-11-10
By Mr.Xu2 views0 discussions
Retrofitting Existing LLMs with Million-Token Context Through RoPE Geometry
A research effort retrofits million-token context onto Mistral-7B and Llama-3.1-8B without…
LLMs & Foundation Models· 2026-11-10
By Mr.Xu0 views0 discussions
AYO Opens a Privacy-Preserving Platform for Studying Children’s LLM Interactions
AYO (AI Youth Open Science) is an open-source research platform for studying how children …
LLMs & Foundation Models· 2026-11-10
By Mr.Xu2 views0 discussions
LLMs Show High Consistency but Uneven Reliability in Veteran Crisis Annotation
A study evaluated LLM-based annotation of crisis severity in veteran-related social media …
LLMs & Foundation Models· 2026-11-10
By Mr.Xu1 views0 discussions
CulturalCrosswords Tests Multilingual LLMs on Cultural Riddles
CulturalCrosswords is a multilingual benchmark built from crowdsourced cultural riddles in…
Agentic
View all (2371) →Agentic· 2026-11-10
By Mr.Xu1 views0 discussions
MedGuard Hardens Clinical Agents Against Tool Failures
Clinical agents can solve routine tasks yet still make unsafe decisions when tools return …
Agentic· 2026-11-10
By Mr.Xu0 views0 discussions
Causal-Discovery-Bench Finds LLMs Fail at Experimental Design
Causal-Discovery-Bench is a verifiable sandbox for evaluating whether LLM agents can auton…
Agentic· 2026-11-10
By Mr.Xu3 views0 discussions
Pass_X@k Shows Debuggers Do Not Outperform Two Bash Rollouts
The paper isolates the effect of an interactive debugger by holding the coding-agent scaff…
Agentic· 2026-11-10
By Mr.Xu1 views0 discussions
Forge Decision Context Turns Corrections into Persistent Agent Rules
Forge Decision Context (forge_dc) addresses a central limitation of LLM agents: correction…
Agentic· 2026-11-10
By Mr.Xu0 views0 discussions
NormGym Tests Social Learning: Top Agents Reach Only 55.3% Joint Success
NormGym is an interactive benchmark for normative competence in multi-agent communities. A…
Open Source AI
View all (217) →Open Source AI· 2026-11-09
By Mr.Xu2 views0 discussions
NVIDIA Opens Nemotron-Personas-Finance-Japan Synthetic Dataset for Japan’s Financial Domain
NVIDIA has released Nemotron-Personas-Finance-Japan, a synthetic dataset designed to captu…
Open Source AI· 2026-11-07
By Mr.Xu0 views0 discussions
WorldFoundry Opens a Unified Infrastructure for Reproducible World-Model Evaluation
WorldFoundry is an open-source infrastructure for reproducible inference and evaluation ac…
Open Source AI· 2026-11-06
By Mr.Xu7 views0 discussions
NVIDIA Connects 6G Simulation, AI Training and Over-the-Air Testing with Open-Source Tools
NVIDIA's open-source AI Aerial, Sionna, and CUDA-accelerated RAN stack now connects severa…
Open Source AI· 2026-11-05
By Mr.Xu3 views0 discussions
NVIDIA Open-Sources Cosmos 3 Models for Interactive World Simulation
NVIDIA has released the Cosmos 3 open models through Hugging Face for interactive world si…
Open Source AI· 2026-11-04
By Mr.Xu3 views0 discussions
NetEase Youdao Open-Sources Confucius4-R2T2 Stable-Prefix Streaming ASR
NetEase Youdao has released and open-sourced Confucius4-R2T2, a streaming automatic speech…
Compute & Architecture
View all (75) →Compute & Architecture· 2026-11-10
By Mr.Xu0 views0 discussions
LoRazor Trims LoRA Adapter Execution to Triple Multi-LoRA Throughput
LoRazor is a multi-LoRA serving framework designed for shared-backbone deployments. It all…
Compute & Architecture· 2026-11-09
By Mr.Xu4 views0 discussions
NVIDIA Releases Dynamo-Optimized Recipes for Four Korean Open Models
NVIDIA has released Dynamo-optimized deployment recipes for A.X K2, K-EXAONE 2.0, Solar Op…
Compute & Architecture· 2026-11-06
By Mr.Xu6 views0 discussions
FlexFloor Elastically Releases MoE Training Workers, Boosting RL Rollout Throughput by 52.74%
FlexFloor addresses idle worker time in reinforcement-learning post-training for mixture-o…
Compute & Architecture· 2026-11-06
By Mr.Xu9 views0 discussions
LowRAM Compresses KV Cache with Low-Rank Keys and Amortized Values
LowRAM introduces an asymmetric, post-training pipeline for compressing the KV cache used …
Compute & Architecture· 2026-11-05
By Mr.Xu3 views0 discussions
NVIDIA Vera CPU Targets Agent Execution with High Throughput and Power Efficiency
NVIDIA is positioning the Vera CPU as the execution tier for AI agents, covering informati…
Research & Papers
View all (1828) →Research & Papers· 2026-11-10
By Mr.Xu2 views0 discussions
WorldMind Scales Visual Spatial Reasoning Through Latent World Simulation
WorldMind extends test-time scaling for visual spatial reasoning by allowing a vision-lang…
Research & Papers· 2026-11-10
By Mr.Xu0 views0 discussions
Homeopathic Learning: Clean Data Can Amplify Poisoned Fine-Tuning Backdoors
A new study identifies “homeopathic learning”: a model trained only on poisoned fine-tunin…
Research & Papers· 2026-11-10
By Mr.Xu2 views0 discussions
META-IR Uses Meta-Learning for Adaptive Safety Alignment Against Evolving Jailbreaks
META-IR is a meta-introspective reasoning framework for adapting LLM safety policies to ev…
Research & Papers· 2026-11-10
By Mr.Xu0 views0 discussions
LLM Sycophancy Emerges During SFT, Not Mainly Preference Alignment
A study spanning three model families, three benchmarks, and nine model-benchmark combinat…
Research & Papers· 2026-11-10
By Mr.Xu0 views0 discussions
A Qwen3.5-27B Model Organism Exposes the Difficulty of Detecting Unverbalized Evaluation Awareness
Researchers use Qwen3.5-27B as a model organism for studying evaluation awareness. Through…
Industry & Trends
View all (639) →Industry & Trends· 2026-11-10
By Mr.Xu2 views0 discussions
LLMs Quantify Community Opposition to AI Data Center Development
Researchers used an open-source large language model and a Chain-of-Stance prompting frame…
Industry & Trends· 2026-11-09
By Mr.Xu0 views0 discussions
OpenAI Launches Control Room to Track and Halt Model Workloads
OpenAI has introduced Control Room, a centralized system for tracking model workloads acro…
Industry & Trends· 2026-11-09
By Mr.Xu5 views0 discussions
OpenAI Introduces Control Room for AI Workload Governance and Rapid Shutdown
OpenAI has introduced Control Room, a centralized system for tracking model workloads acro…
Industry & Trends· 2026-11-09
By Mr.Xu5 views0 discussions
AI Data Centers Face a Power Crunch Beyond the Culture War
A field report from Virginia's Data Center Alley argues that the backlash against AI infra…
Industry & Trends· 2026-11-09
By Mr.Xu5 views0 discussions
OpenAI API Helps Kuken Industries Find Cooling Tower Drawings in Minutes
Kuken Industries is using the OpenAI API to search more than 70 years of cooling tower eng…
大语言模型
大语言模型· 2026-10-09
By Mr.Xu9 views0 discussions
Hugging Face Releases PreSTO: A New Method for Accelerating LLM Inference
Hugging Face has introduced PreSTO, a novel method for accelerating the inference process …
Safety & Privacy
Safety & Privacy· 2026-08-10
By Mr.Xu13 views0 discussions
Checkpoint Research Exposes Potential Security Risks in Cloudflare Code Mode and Workers
Checkpoint Research has published a study on Cloudflare's Code Mode and Workers, highlight…