Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
OpenAI's internal reasoning model autonomously disproved the planar unit distance conjecture, a mathematical problem posed by Paul Erdős in 1946 that had remained unsolved for nearly 80 years. The AI produced a new famil…
NVIDIA researchers developed a task-seeded synthetic Q&A generation workflow for Nemotron-family pretraining that uses public task training splits as capability seeds to generate new task-aligned examples with reasoning …
Apple released a new ad promoting Safari as a more private browsing alternative to Google Chrome, depicting Chrome's data trackers as chrome-suited figures that follow users. The ad highlights Safari's default blocking o…
Honeycomb Insurance raised $40 million in a funding round led by Zeev Ventures, bringing its total funding to $95 million, as the AI-native insurer uses geospatial data and aerial imagery to price multi-unit residential …
A developer has published code and research showing that cross-family evaluation—using a generator and judge from different model families—reduces self-preference bias in AI pipelines by an average of 31.5%. The approach…
Researchers have developed DiffusionBlocks, a method that trains neural networks by dividing them into blocks and training only one block at a time, reducing memory requirements to roughly 1/B of standard end-to-end trai…
AWS Lambda's reported `@maxMemoryUsed` metric tracks a high-water mark across an entire execution environment, not per-invocation memory usage, according to AWS support. A customer's ONNX model inference Lambda function …
Google released Gemma 4 12B, a 12-billion-parameter multimodal model designed to run on consumer laptops with 16GB of RAM or VRAM. The encoder-free model accepts text, images, and native audio, and its benchmark performa…
A developer found that 4 to 8 dedicated GPUs, such as the H200 NVLink, are sufficient for most production inference workloads on 70B to 200B parameter models, debunking the need for multi-node clusters. The analysis show…
A developer trained a small large language model on a single notebook to understand the end-to-end process, from data ingestion to inference, and then deployed it to run queries directly inside an Exasol database. The pr…
The next major advance in artificial intelligence will depend on better data, not larger models, according to AI researchers. A "data gap" between what models can theoretically achieve and what they can do in practice is…
DaltonTx launched Dalton, a drug discovery platform that integrates molecules, assays, models, experimental outcomes, project context and human decisions into a single learning system, according to a June 4 press release…
Phala Cloud's documentation separates GPU TEE performance benchmarks from per-job compute receipts, emphasizing that benchmarks describe hardware capacity while receipts record specific job execution details. A proposed …
A developer's analysis of LLM fine-tuning versus retrieval-augmented generation (RAG) finds that roughly 70% of production LLM problems are solved by better prompting or RAG, with fine-tuning reserved for the remaining 3…
Miso Labs released MisoTTS, an open-weights 8-billion-parameter text-to-speech model that generates expressive speech from text and audio context. The model uses residual vector quantization to expand its sonic range wit…
Researchers at Stanford University and Lambda Labs released OpenJarvis, an open-source framework that runs AI inference, agents, memory, and learning entirely on-device. The framework achieves performance within 3.2 perc…
NVIDIA released Cosmos 3 on June 1, 2026, a single foundation model that unifies physical reasoning, world simulation, and action generation within one architecture. The model uses a Mixture-of-Transformers (MoT) design …
Researchers at Stockholm University have developed an artificial intelligence pipeline that decoded the 400-page 17th-century Borg Cipher in 28 minutes, a task that previously required weeks of manual work. The system, w…
An arXiv preprint (arXiv:2606.03995) reports an explainable machine learning model that achieved near-perfect three-class classification of Alzheimer's disease, mild cognitive impairment, and normal cognition using eight…
A new arXiv preprint (arXiv:2606.04574) submitted June 3, 2026, presents a hybrid trading architecture that combines statistical pair selection with a Deep Reinforcement Learning execution overlay for crypto markets. The…