Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Pathway's brain-inspired BDH (Dragon Hatchling) architecture, which performs reasoning in latent space without chain-of-thought token generation, is being developed on Amazon SageMaker HyperPod to scale training. The arc…
A study submitted to arXiv on 4 Sep 2026 by William Gilpin finds that leading AI reasoning models exhibit transient chaos and fractal basins, with fractality increasing with task difficulty across tasks like Sudoku, maze…
Danijar Hafner's Dreamer series of world models, culminating in Dreamer 4, demonstrates that robots can learn complex tasks from offline data alone, with Dreamer 3 being the first to autonomously solve the Minecraft Diam…
Amazon Web Services (AWS) announced the general availability of the UpdateRecord API for Amazon SageMaker Feature Store, enabling feature-level writes that update one or more feature values in a single call without readi…
A developer built an AI memory system called 'mycelium' that uses a Hebbian connection graph to reinforce associations through use, but discovered that its recall mechanism favors long-established 'hub' memories over rec…
A FlyRank Machine Learning intern built an offline analytical pipeline that predicts content decay before organic traffic drops, using 30,000 anonymized search performance rows. The optimized Random Forest model achieved…
Google released version 3 of its WeatherNext AI weather model, which now ingests satellite weather data to shorten forecast lag time and increase forecast frequency to hourly, according to a white paper on arXiv. The upd…
Willette M. Crawford, a food science consultant and cofounder of Katalyst Agricultural Solutions, said AI is improving food safety by enabling faster, more targeted outbreak detection, and that the recent rise in recalls…
Cisco's IOS XE routers now support ML-DSA certificates (FIPS 204) for IPsec authentication, replacing pre-shared keys, as detailed in a technical blog post. The post shows that ML-DSA-65 public keys are 1,974 bytes versu…
DeepSeek-R1's use of Reinforcement Learning with Verifiable Rewards (RLVR) marks a shift from RLHF's subjective human feedback to deterministic correctness checks, enabling models to reason more reliably. The technique, …
Developer rentruewang released InverSQL, an open-source Streamlit app that generates SQL queries from CSV uploads by overfitting a scikit-learn decision tree, converting it to boolean logic with sympy, and simplifying it…
Amazon Web Services benchmarked two 30B Mixture-of-Experts LLMs on SageMaker AI, showing that new G7 instances powered by NVIDIA Blackwell GPUs deliver measurable gains in throughput, latency, and cost-per-token over G5 …
Tavus, a conversational AI company, released Sparrow-2, a new audio-understanding and turn-taking model that processes all sounds—including breathing, sighs, and background noise—rather than relying on noise cancellation…
Laravel's native vector search, previously limited to PostgreSQL with pgvector, now supports MariaDB through a refactoring that moves vector distance logic from the connection class to the grammar. The change, merged in …
DiDi Global's International Business Group partnered with AWS to build an intelligent contact center quality assurance system on Amazon Bedrock, replacing an opaque third-party solution. In production validation, intent …
A study by Epoch AI finds that from 2019 to 2025, data improvements contributed 12.0x compute efficiency gains compared to 3.7x from model improvements, meaning 3.24x more gains came from data at a 1e19 FLOPs budget. The…
A developer has trained a reinforcement learning model to race in the video game TrackMania, using the Soft Actor-Critic algorithm and a mod called TMInterface to access game state. The project runs on Linux via Proton, …
DeepMind's AlphaGenome Atlas maps the 3-billion-letter human genome to predict the functional impact of single nucleotide polymorphisms, aiming to solve the variant effect problem by distinguishing benign mutations from …
A practitioner's account details three generations of enterprise retrieval systems, arguing that standard RAG fails on ambiguous queries, terminology mismatches, and confidence signaling, and that hybrid retrieval combin…
TrustGraph 2.8 has been released, introducing an asynchronous architecture, hybrid retrieval, structured output, and a plugin-based workbench. The update replaces thread-per-consumer pub/sub with configurable async pools…