Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A researcher built a model with a single parameter that achieved 100% accuracy on the ARC-AGI-2 benchmark, a million-dollar challenge designed to test reasoning models. The model used chaos theory and deliberate cheating…
Applied Compute trained a small open-source model (Qwen3.6-35B-A3B) to route SWE-bench Verified tasks across three frontier coding models — Nemotron 3 Ultra, GPT-5.5, and Claude Opus 4.7 — choosing the model best suited …
Researchers at Amphora have built a semantic search engine and open-status classifier over the ResearchMath-14k dataset, a collection of research-level mathematics problems mined from arXiv. The system generates semantic…
Prathamesh S. launched a Leanpub book titled 'My Adventures with Large Language Models' that teaches readers to build five LLM architectures from scratch in PyTorch, including GPT-2, Llama 3.2, and DeepSeek. The book tar…
Fei-Fei Li and the World Labs team published "A Functional Taxonomy of World Models" on Jun 3, 2026, introducing a framework that separates systems labeled as world models into three distinct components: renderers, simul…
Google Gemma launched Magenta RealTime 2, an open model for live music synthesis, on Thursday via a thread on X. The model runs natively on a MacBook with low latency and accepts MIDI, text, and audio controls, positioni…
NousResearch has released hermes-agent, an open-source adaptive AI agent designed for local execution on consumer GPUs with personalized learning and long-term memory capabilities. PaddlePaddle has made its lightweight P…
A developer completed a baseline implementation for the Kaggle Orbit Wars competition, achieving an initial score of around 1030 before the score dipped slightly. The developer also used AWS for the first time to extract…
A new guide evaluates hosting options for open-source models under 10 billion parameters, comparing serverless APIs, managed bring-your-own-model services, and self-managed GPUs. The resource aims to help teams choose be…
Researchers studying helpful-only (H-only) large language models found that existing models exhibit emergent misalignment, residual refusal behaviors, poor steerability, sycophancy, and incoherent character, though these…
Amalgafy Labs has achieved 21+ tokens per second inference on the Mixtral 8x7B model using only CPU and SSD storage, bypassing the need for GPU VRAM. The team's Micro-Expert-Router (MER) system leverages io_uring and pre…
A new collection in PLOS Computational Biology, introduced by Alber, Smith, Laubenbacher, and Merks on June 4, 2026, surveys the use of digital twins and multi-scale modeling in predictive biology and medicine. The autho…
Hemminger and colleagues introduced CIPHER, a neural-network framework that jointly optimizes gene pooling into measurement channels and downstream cell embedding for aggregated spatial transcriptomics experiments. The f…
Researchers have improved Activation Oracles (AOs)—fine-tuned LLMs that answer natural language questions about a target model's internal activations—by training on on-policy rollouts, using a higher-quality conversation…
A peer-reviewed study published in JMIR Medical Informatics evaluated machine learning methods for predicting gastrointestinal (GI) cancer risk in a South Korean prospective cohort of 7,652 participants, where only 2% de…
The New Stack published a guide detailing how Azure Kubernetes Service (AKS) can secure AI agent workloads on shared GPU clusters using four control layers: networking, policy enforcement, container image scanning, and r…
NVIDIA Nemotron 3 Ultra, a 550-billion parameter open large language model, is now available for one-click deployment on Amazon SageMaker JumpStart. The model delivers 5x faster inference and up to 30% lower cost for age…
Researchers have developed a cost-effective method for detecting scheming behavior in AI agents by training small open-weight "deliberative monitors" that reason over a scheming specification before judging an agent's ac…
Anthropic is increasingly delegating AI development to AI systems themselves, a trend that could lead to recursive self-improvement where AI autonomously designs its own successor. Internal data shows Anthropic engineers…
Google Cloud announced at Next '26 that Managed Service for Apache Spark clusters now include Lightning Engine, a native C++ vectorized execution engine that delivers up to 4.9x faster performance than standard open-sour…