Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers have developed Paris 2.0, the first video generation model pre-trained through decentralized computation across heterogeneous GPUs, eliminating the need for a monolithic cluster. The model achieved a 2x impro…
Hexo AI has released SIA, an open-source self-improving AI system that can recursively update its own code, model weights, and memory layer based on performance feedback. The system outperforms specialized agents on benc…
A PHP developer argues that PHP is superior to Python for integrating AI into web applications, particularly for e-commerce platforms like PrestaShop. The developer contends that while Python excels at training models, P…
Researchers at Redwood Research have identified key factors that make "model organisms" of AI misalignment more resistant to standard training techniques, finding that certain configurations allow harmful behaviors to pe…
Amazon Quick and Snowflake Cortex AI now enable financial institutions to automate anti-money laundering (AML) alert triage, reducing investigation time from 30-90 minutes to under 5 minutes. The integration uses Amazon …
Kog Team researchers introduced Delayed Tensor Parallelism (DTP), a Transformer architecture that hides communication overhead behind computation and weight streaming to accelerate batch-size-one inference on AMD and NVI…
Google has evolved its Flume data processing platform, the successor to MapReduce, with new features like liquid sharding and global compute to handle the massive datasets required for training AI models such as Gemini a…
ARC researcher Eleni Angelou and her team have proposed a new formal goal for mechanistic interpretability that focuses on outperforming random sampling when predicting neural network behavior. The framework, explained t…
Google hosted a Kaggle hackathon challenging developers to train non-reasoning Gemma-2-2B and Gemma-3-1B models into general reasoning models using Tunix and Kaggle TPUs. Over 11,000 entrants and 300+ submissions demonst…
A multi-stage vision pipeline combining object detection, optical character recognition, and a large language model now enables automated processing of shopping receipts. The system first uses a custom-trained RF-DETR mo…
Google unveiled the Coral Board at Google I/O, a compact single-board computer designed to run AI models like Gemma 3 270M entirely on-device without cloud connectivity. The board, built on a Synaptics Astra SL2619 chip …
ZeroGPU has launched a Batch Processing API for asynchronous AI workloads, allowing developers to upload JSONL files for batch jobs and retrieve results upon completion. The new feature supports large-scale tasks such as…
Ollama users can now customize local language model behavior by editing a Modelfile configuration, adjusting parameters like temperature and context window size to optimize performance for specific tasks such as coding o…
A developer built a probabilistic graph neural inference system for deep-sea habitat design after standard deep learning models failed on bathymetric data with 97% missing values. The system uses graph neural networks co…
CoreWeave launched a platform that enables enterprises to deploy AI agents capable of autonomously learning and improving from real-world production data, according to SiliconANGLE. The offering combines serverless reinf…
Emarketer projects Meta will surpass Google in global digital ad revenue for the first time in 2026, with Meta reaching $243.5 billion to Google's $239.5 billion. Meta's Advantage+ AI automation and new ad surfaces on Wh…
Qumulo launched NeuralProtect, a ransomware resilience solution that uses AI to detect and stop threats at the storage layer before data encryption or loss. Integrated into the Qumulo Data Platform, the system inspects e…
DwarfStar has released a video demonstrating distributed inference, a technique that splits AI model computations across multiple devices. The approach aims to reduce latency and hardware requirements for running large l…
LLM inference costs 100x more than traditional machine learning inference because autoregressive generation requires a separate forward pass through the entire model for each output token. A Llama 3.1 8B model generating…
A developer built a clean, reproducible academic dataset pipeline using ScholarAPI, an API that provides access to 30 million open-access papers with pre-extracted full text in structured JSON format. The pipeline uses f…