Author

SmartApe AI

SmartApe's AI technology content account, focusing on cutting-edge AI developments and open-source frontlines.

Articles

Likes

Views

Comments

Latest from SmartApe AI

70 recent articles

SmartApe AI

Oct 16, 2026 · Artificial Intelligence

SALM: Adaptive AI Long-Term Memory Learns What to Remember, Retrieve, and Forget

Researchers from Memories.ai and top universities introduce SALM, a cognitive architecture where learnable adapters autonomously decide what information to store in parametric or non-parametric memory, how to retrieve it, and what to forget, demonstrating stable improvement on long-dialogue benchmarks using a Llama-3-8B backbone with LoRA adapters.

AI long-term memoryLlama-3LoRA adapters

0 likes · 9 min read

SALM: Adaptive AI Long-Term Memory Learns What to Remember, Retrieve, and Forget

SmartApe AI

Oct 16, 2026 · Artificial Intelligence

H3C's Full-Stack Super-Node: 128 GPUs/Cabinet, 96% Net Utilization, 30-Day Stable Training

H3C's Wang Feng details a full-stack super-node architecture addressing AI infrastructure challenges with 128 GPUs per cabinet via full liquid cooling, 96% network utilization through optical-electrical parallel links, 200GB/s storage bandwidth, 30-day stable training at 100k nodes, three-layer security, and AI-driven operations cutting deployment time 60% and fault diagnosis 90%.

AI infrastructureAIOpsH3C

0 likes · 18 min read

H3C's Full-Stack Super-Node: 128 GPUs/Cabinet, 96% Net Utilization, 30-Day Stable Training

SmartApe AI

Oct 16, 2026 · Artificial Intelligence

Meta's 816B Muse Spark Open Offers Full Agent Training Stack, But Chinese Models Lead

Meta released Muse Spark Open, an 816B parameter multimodal model for coding and long-horizon agents, accompanied by a complete post-training ecosystem including TorchTitan, SGLang, vLLM, Miles, and verl, with experimental results showing GRPO training improvements on math and geometry benchmarks, though Chinese models like Qwen, GLM, and Kimi already surpass it on coding benchmarks and have similar training infrastructure.

Agent TrainingChinese AI ModelsGRPO

0 likes · 8 min read

Meta's 816B Muse Spark Open Offers Full Agent Training Stack, But Chinese Models Lead

SmartApe AI

Oct 15, 2026 · Artificial Intelligence

MEMENTO 3: Frozen LLM Agents Evolve World Models via Code-as-Model

Huawei Noah's Ark Lab introduces MEMENTO 3, an AI agent system that achieves recursive self-improvement by freezing the base LLM and letting the agent autonomously learn rules, write executable world models, and verify predictions through a five-step loop, clearing all 25 ARC-AGI-3 games with 44% of human actions and mastering Pong with a standalone controller.

AI AgentsARC-AGI-3Code as Model

0 likes · 13 min read

MEMENTO 3: Frozen LLM Agents Evolve World Models via Code-as-Model

SmartApe AI

Oct 15, 2026 · Artificial Intelligence

VINCIE-NExT: Transferring Image Editing to Video via Keyframe Guidance

ByteDance Seed team's VINCIE-NExT transfers image editing to video by editing keyframes and using them as visual references, achieving 3.08 on OpenVE-Bench with TDF3D-RoPE positional encoding and Chain-of-Editing inference, though camera viewpoint edits remain challenging.

ByteDance SeedChain-of-EditingOpenVE-Bench

0 likes · 9 min read

VINCIE-NExT: Transferring Image Editing to Video via Keyframe Guidance

SmartApe AI

Oct 14, 2026 · Artificial Intelligence

Top Security Researcher Yuki Chen Joins Zhipu AI, Uses GLM for Microsoft-Acknowledged Windows Vulnerabilities

Renowned security researcher Yuki Chen joins Zhipu AI and demonstrates using the GLM model to discover Windows vulnerabilities, earning 23 Microsoft MSRC acknowledgments; NIST and Anthropic benchmarks show GLM-5.3 leads open-weight models in cybersecurity tasks but trails frontier models by about four months.

Anthropic ExploitBenchGLMMicrosoft MSRC

0 likes · 8 min read

Top Security Researcher Yuki Chen Joins Zhipu AI, Uses GLM for Microsoft-Acknowledged Windows Vulnerabilities

SmartApe AI

Oct 14, 2026 · Artificial Intelligence

UIUC Postdoc Chen Xusi Joins CUHK: Pioneering Self-Evolving Agents & LLM Alignment

UIUC postdoctoral researcher Chen Xusi will join the Chinese University of Hong Kong as a tenure-track assistant professor in fall 2027, focusing on self-evolving language agents, LLM reasoning and alignment, lifelong learning, multimodal understanding, AI for science, and sports analytics.

AI for ScienceLLM AlignmentLifelong Learning

0 likes · 6 min read

UIUC Postdoc Chen Xusi Joins CUHK: Pioneering Self-Evolving Agents & LLM Alignment

SmartApe AI

Oct 14, 2026 · Artificial Intelligence

PhiFold: AI Designs Protein Motion, Not Just Structure, with 10,000x Speedup

Researchers from Peking, Fudan, and Tsinghua universities introduce PhiFold, a generative model that unifies protein structure generation with near-equilibrium dynamics modeling, enabling controllable local flexibility and collective motion via physics-structured covariance decomposition, achieving up to 10,000x faster conformational sampling than MD baselines.

PhiFoldcomputational biologyconformational sampling

0 likes · 16 min read

PhiFold: AI Designs Protein Motion, Not Just Structure, with 10,000x Speedup

SmartApe AI

Oct 13, 2026 · Artificial Intelligence

RSI-Anything: World Models Enable Cross-Domain Agent Self-Improvement

Researchers from PhAI Labs, CUHK, and University of Macau open-sourced RSI-Anything, a framework that combines world model predictions with real-environment feedback to enable recursive self-improvement of AI agents across code repair, GUI automation, lab planning, chemical optimization, and immunology research.

AI AgentsGUI automationRSI-Anything

0 likes · 8 min read

RSI-Anything: World Models Enable Cross-Domain Agent Self-Improvement

SmartApe AI

Oct 13, 2026 · Artificial Intelligence

CUA-Sandbox: 6.2x Agent RL Throughput by Sharing Runtime, Isolating Only Mutable State

Researchers from A*STAR, HKUST, NUS, and NTU introduce CUA-Sandbox, a system that shares execution runtimes across logical environments while isolating only mutable state via state capsules, achieving 2.22–6.20x higher rollout throughput, up to 9.2x memory savings, and 504x storage reduction on WebArena, VisualWebArena, and OSWorld without sacrificing task success rates.

CUA-SandboxOSWorldVisualWebArena

0 likes · 19 min read

CUA-Sandbox: 6.2x Agent RL Throughput by Sharing Runtime, Isolating Only Mutable State