Your daily digest
Research papers, made simple to understand
The AI and machine learning papers people are reading most, each one rewritten in plain words: the problem, the idea, what was found and why it matters.
- AI Agents
- Language Models & Architectures
Xing 4 advances large language models on Ascend chips with efficient architecture and training
Xing 4 models optimize architecture and training on Ascend NPUs, achieving efficient long-context reasoning and agentic tasks with a 285B-parameter model pretrained on 8,192 Ascend 910C cards and a 29B open-source MoE model supporting 256K-token contexts.
Paper: TeleChat to Xing4: Advancing Frontier Agentic Models on Domestic Chips and Frameworks · Xingchen AGI Lab, Baotian Hu, Bingkai Yang and 136 others
150 upvotes on Hugging Face6 min readRead the article
- AI Agents
- Benchmarks & Evaluation
AscendGYM benchmarks kernel optimization on Ascend NPUs with cost-aware metrics
AscendGYM evaluates agents' ability to optimize existing Ascend NPU kernels from real LLM workloads, using a new Speedup-Token Score metric that balances performance gains with token cost under budget constraints.
Paper: AscendGYM: A Cost-Aware Kernel Optimization Benchmark for Real-World Workloads on Ascend NPUs · AI Training Platform Team, Bingwei Zhang, Jing Li and 24 others
91 upvotes on Hugging Face7 min readRead the article
- Language Models & Architectures
LLMs design sparse table models outperforming Transformers in weight-space programming
Large language models (LLMs) writing weights directly without gradient training prefer sparse associative-memory tables over Transformers, achieving up to 54.4% accuracy on ARC with 1,800× fewer parameters and up to 17× faster inference on GPUs.
Paper: Weight-Space Programming Doesn't Need Transformers · Hyunjong Ok, Jaeho Lee
49 upvotes on Hugging Face5 min readRead the article
Browse earlier papers
27 days of papers in October 2026
- Chain-of-thought prompting improves reasoning in large language modelsand 2 more papers
- Intrinsic Distribution Matching improves one-step text-to-image generation with feature moment constraintsand 2 more papers
- Falcon enables efficient, real-time audio-visual target speaker extraction in complex scenesand 2 more papers
- Capability-driven recursive self-improvement evolves specialized multimodal agent modulesand 2 more papers
- Trajectory2Gym improves tool-use agents by synthesizing and grounding compositional interactionsand 2 more papers
- Survey unifies video understanding and generation via world modeling frameworkand 2 more papers
- Fine-tuning GPT-3 with human feedback improves instruction following and truthfulnessand 2 more papers
- Residual connections enable training of very deep neural networks with improved accuracyand 2 more papers
- WeVisOne-VL unifies object- and pixel-level vision-language tasks with structured token predictionand 2 more papers
- StreamEval benchmark measures interactive streaming video generation across five tasksand 2 more papers
- FocusGUI improves multi-step omni-modal GUI agents by dynamically prioritizing relevant inputsand 2 more papers
- Parameter-efficient fine-tuning scales large multimodal models for brain encodingand 2 more papers
- HexaShell uses coding agents with RGB triangulation to improve robot manipulation on RoboDojoand 2 more papers
- Two efficient models learn high-quality word vectors from huge text corporaand 2 more papers
- Transformer replaces recurrence with multi-head self-attention for faster, better translationand 2 more papers
- Waypoint-OPD improves RL policy transfer by gating shifts using intermediate checkpointsand 2 more papers
- ATAS framework improves spatial and temporal consistency in agentic video generationand 2 more papers
- Jeva architecture enables stable, scalable, and cacheable system one decisionsand 2 more papers
- GroundingPI improves precise visual grounding for physical intelligence tasksand 2 more papers
- DiffusionART stabilizes and speeds up reinforcement learning fine-tuning of diffusion modelsand 2 more papers
- LoGRA cuts LLM RL training memory up to 45.7% using low-rank gradient sketchesand 2 more papers
- UltraText Bench evaluates dense bilingual visual text rendering in image generationand 2 more papers
- AgentGarten enables evolving agents via code-driven worlds and neural renderingand 2 more papers
- STEPQuant reduces memory and preserves accuracy in low-bit quantization of recurrent statesand 2 more papers
- Compact strict alignment supervision outperforms broader coverage in cross-tokenizer distillationand 2 more papers
- Kandinsky 6.0 Video generates synchronized 5-second videos with 44 kHz audio using dual-stream diffusion modelsand 2 more papers
- RealCompanion benchmarks AI understanding of real long-term human conversationsand 2 more papers