ai10 Sept 2026, 09:00
3GAS-PINNs Proposed to Improve Physics-Informed Neural Network Accuracy
The newly developed 3GAS-PINNs method boosts numerical simulation accuracy for nonlinear equations by up to a factor of 14.
Source: arXiv cs.LG
Latest news
From Kazakhstan to the world — from the world to Kazakhstan
ai10 Sept 2026, 09:00
Researchers Propose World-Time Compute Using Verified Code World Models
A new preprint introduces an approach to train language models using exact trajectories from verifiable code world models.
Source: arXiv cs.LG
ai10 Sept 2026, 09:00
Researchers Scale Post-Training Ternarisation to Qwen3-8B
A new preprint explores scaling aggressive post-training ternarisation to the Qwen3-8B model with lossless packing and capability retention.
Source: arXiv cs.LG
ai10 Sept 2026, 09:00
Researchers Introduce OpenDiscoveryTrace Dataset to Evaluate Scientific AI Workflows
The OpenDiscoveryTrace dataset provides 558 detailed trajectories to audit reasoning processes and error modes of scientific AI agents.
Source: arXiv cs.AI
ai10 Sept 2026, 09:00
Researchers Propose Adaptive Entangled Game Modules for AGI Architectures
A new research preprint suggests integrating probability-wave game modules into AGI to overcome the limitations of massive parameter-heavy neural networks.
Source: arXiv cs.AI
ai10 Sept 2026, 09:00
Subagents Outperform Direct Skill Loading in Long-Horizon AI Tasks
A new preprint demonstrates that executing reusable knowledge as dedicated subagents improves AI performance on long-horizon tasks.
Source: arXiv cs.AI
business10 Sept 2026, 09:00
President Receives Management Board Chairman of Samruk-Kazyna
Kassym-Jomart Tokayev reviewed Samruk-Kazyna's key production and financial performance metrics as well as infrastructure projects.
Source: Profit.kz
ai10 Sept 2026, 09:00
Researchers Introduce X-CoSD for Communication-Efficient LLM Decoding
A new preprint proposes X-CoSD to speed up collaborative speculative decoding across models with heterogeneous vocabularies.
Source: arXiv cs.CL
ai10 Sept 2026, 09:00
Researchers Introduce StochBench for Stochastic Processes in Lean
A new Lean 4 benchmark featuring 450 graduate-level stochastic process problems has been introduced for AI theorem proving.
Source: arXiv cs.CL
ai10 Sept 2026, 09:00
Researchers Introduce Osprey to Accelerate LLM Speculative Decoding
The Osprey architecture leverages target-agnostic pre-training of small language models to boost speculative decoding efficiency.
Source: arXiv cs.CL
ai10 Sept 2026, 09:00
SWORD Benchmark Reveals Cross-Lingual Inconsistencies in LLM Fact Checking
A new research benchmark called SWORD reveals that large language models struggle with fact verification across eight languages, showing severe drops in Asian languages.
Source: arXiv cs.CL
ai10 Sept 2026, 09:00
Researchers Propose Evidence-Order Calibration for Vision-Language Models
A new research preprint investigates how vision-language models adjust confidence when question-critical visual evidence is progressively lost.
Source: arXiv cs.CV