Black Forest Labs has launched FLUX 3 Image Klein, a compact 5.6-billion parameter open-weights model capable of running on consumer-grade GPUs.
Developers are building a custom upscaling model to restore old photos locally, avoiding the limitations of commercial SaaS platforms.
Over half of Russian job seekers have integrated neural networks into their professional workflows, with text-based chatbots remaining the most popular tool.
Researchers at the Moscow Institute of Physics and Technology have developed the HARP algorithm to optimize neural network compression and reduce computational demands on data centers.
AI VK Research has introduced a method to scale graph neural networks, reducing memory usage by over 100 times and increasing recommendation efficiency for social networks.
Researchers testing small language models found that while initial fine-tuning yields performance gains, subsequent rounds fail to produce cumulative learning, prompting a search for methodological flaws.
A developer shares a workflow for creating a digital-themed mood board using a combination of manual collage techniques and iterative AI image processing.
Researchers are investigating techniques to generate neural network weights dynamically, potentially reducing storage and memory demands by replacing static matrices with compact generator functions.
The ASI-Arch system has autonomously created 105 neural network architectures, with the top model significantly outperforming the Mamba2 architecture in efficiency tests.
Developers of an online backgammon platform analyzed over 170,000 human moves to evaluate player accuracy against a neural network engine.
A comparative study of 12 video generation models evaluated performance, speed, and cost using a standardized prompt, with Luma Ray 3.14 emerging as the most efficient leader.
MWS AI has tested the ability of multimodal models to detect document fraud using a new experimental dataset of 430 images.
An experiment comparing eight image-generating AI models reveals the efficiency of different tools for virtual hair styling and the importance of combining analytical AI with generative capabilities.
Smart Engines researchers have developed a tree-based averaging algorithm to accelerate 8-bit neural network matrix multiplication on ARM NEON hardware.
NVIDIA's new PivotOPD training method enables multi-turn AI agents to recover from critical errors, significantly improving success rates across major benchmarks.
OTP Bank has launched an AI-powered onboarding course for new employees, replacing traditional manuals with an interactive animated landing page.
A recent study claiming to find a 'pain vector' in LLMs has been debunked as a misinterpretation of how neural networks process language and associated concepts.
SkyrimNet explores NPC behavior through a multi-layered memory system that combines game data, external search, and LLM-driven personality traits.
A comparative test of eight AI models in September 2026 reveals significant differences in Russian text rendering, pricing, and speed for image generation.
Researchers from the Kulakov National Medical Research Center and RTU MIREA have launched an AI tool that analyzes CTG data to predict perinatal outcomes with 86% accuracy.
