Author

PaperAgent

Daily updates, analyzing cutting-edge AI research papers

335

Articles

Likes

2.9k

Views

Comments

Latest from PaperAgent

100 recent articles max

PaperAgent

Oct 25, 2026 · Artificial Intelligence

Google Discovers Two 20-Year-Old Bugs in Softmax-Cross-Entropy Implementation

Google researchers identify two numerical bugs in the standard softmax-cross-entropy implementation used for two decades: Softmax Collapse zeros out gradients beyond a precision cliff, and Softmax Cancellation creates a noise floor in distillation; fixes using stable log-space computations eliminate Slingshot loss spikes in grokking and improve convergence across precisions.

Adam optimizerSlingshotcross-entropy

0 likes · 8 min read

Google Discovers Two 20-Year-Old Bugs in Softmax-Cross-Entropy Implementation

PaperAgent

Oct 24, 2026 · Artificial Intelligence

NVIDIA's AW-0: Agentic World Model Unifies Code, Video & Action for Robot Control

UC Berkeley and NVIDIA introduce AW-0, a Mixture-of-Transformers agentic world model that jointly generates code plans, video predictions, and 64-DoF action chunks, achieving 50.7% average success on RoboTwin 2.0 and 71.1% on real dual-arm and dexterous-hand robots.

AW-0Agentic World ModelJoint Denoising

0 likes · 7 min read

NVIDIA's AW-0: Agentic World Model Unifies Code, Video & Action for Robot Control

PaperAgent

Oct 23, 2026 · Artificial Intelligence

Naive RAG Outperforms Complex Agentic Memory: RAW-RAG Beats 12 Systems with Simple Retrieval

A study from XJTU, SJTU, and ZJU shows RAW-RAG — a simple dense retrieval + reranking pipeline on raw dialogue history — outperforms 12 agentic memory systems across four benchmarks, achieving higher accuracy with 83× faster ingestion because raw evidence preserves reasoning paths that compression loses.

LLMLong-Term MemoryRAG

0 likes · 7 min read

Naive RAG Outperforms Complex Agentic Memory: RAW-RAG Beats 12 Systems with Simple Retrieval

PaperAgent

Oct 22, 2026 · Artificial Intelligence

EviRepair: Multi-Agent Evidence Review Boosts Automated Vulnerability Repair

EviRepair introduces a two-phase multi-agent system for automated vulnerability repair that first diagnoses vulnerabilities through parallel evidence-gathering agents and a reviewer, then synthesizes patches validated against the verified vulnerability model, achieving higher fix rates on PatchEval and SEC-bench benchmarks.

EviRepairLLM agentsMulti-Agent Systems

0 likes · 7 min read

EviRepair: Multi-Agent Evidence Review Boosts Automated Vulnerability Repair

PaperAgent

Oct 21, 2026 · Artificial Intelligence

Atria Dawn Update Brings Editable PPT, Protein Folding, Benchmark Gains

Shanghai AI Lab's Atria Dawn agentic AI system now generates editable PowerPoint files with native shapes, creates academic posters from papers, optimizes protein structure prediction using OpenFold, performs multi-file business analysis, and shows improved benchmark scores on AutomationBench, SkillsBench, WideSearch, and JobBench.

Academic Poster GenerationAgentic AIAtria Dawn

0 likes · 6 min read

Atria Dawn Update Brings Editable PPT, Protein Folding, Benchmark Gains

PaperAgent

Oct 20, 2026 · Artificial Intelligence

Agentic Robots: Models, Data, Environments, and Harness Unified

CUHK and Fudan researchers survey the shift from robot policies to Agentic Robots, organizing methods across Model, Data, Environment, and Harness domains with 17 functional categories, and highlight challenges in deploying autonomous robots that dynamically coordinate skills based on feedback.

Agentic RobotsCUHK MMLabData Annotation

0 likes · 10 min read

Agentic Robots: Models, Data, Environments, and Harness Unified

PaperAgent

Oct 19, 2026 · Artificial Intelligence

Gemini 4 Argon Agents Achieve 27-Hour Autonomous Minecraft Mission

Google's Gemini 4 Argon model powers four autonomous agents that collaboratively complete a 27-hour, 13-step Minecraft mission without human intervention, demonstrating emergent engineering primitives, self-correction, and long-horizon planning through a code-based REPL architecture and minimal communication protocols.

Gemini 4 ArgonMinecraftMulti-Agent Systems

0 likes · 10 min read

Gemini 4 Argon Agents Achieve 27-Hour Autonomous Minecraft Mission

PaperAgent

Oct 17, 2026 · Artificial Intelligence

DeepSeek-V4.1-Flash-1017 Open-Sourced: Self-Feedback RL Boosts Agent Performance

DeepSeek open-sources DeepSeek-V4.1-Flash-1017 with new RLSF (Reinforcement Learning with Self-Feedback) technique, showing across-the-board gains on 11 benchmarks, faster and shorter outputs, and demonstrating end-to-end project generation like 3D modeling and game development, all under MIT license.

AgentDeepSeekLLM

0 likes · 9 min read

DeepSeek-V4.1-Flash-1017 Open-Sourced: Self-Feedback RL Boosts Agent Performance

PaperAgent

Oct 17, 2026 · R&D Management

How Anthropic Scales Code Review with AI: 90% Automated, 40% Fewer Rollbacks

Anthropic's Claude Code team shares their AI-driven code review pipeline that doubled merge rate while cutting rollbacks by 40%, using layered automated checks, risk-based human approvals, and incident-fed evaluation sets to keep reviews convergent and accountable.

AI-assisted developmentAnthropicCI/CD

0 likes · 8 min read

How Anthropic Scales Code Review with AI: 90% Automated, 40% Fewer Rollbacks

PaperAgent

Oct 16, 2026 · Artificial Intelligence

Doubao Work's RSI Loop: Plan & Goal Modes Enable Self-Correcting Deep Research

The article demonstrates how Doubao Work's Plan Mode aligns research strategy via editable markdown plans, and Goal Mode executes with parallel agents that auto-validate against acceptance criteria, creating a recursive self-improvement loop for deep research tasks like evaluating Jev's technical positioning.

AI AgentsAgent ParallelismDeep Research

0 likes · 8 min read

Doubao Work's RSI Loop: Plan & Goal Modes Enable Self-Correcting Deep Research