Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Stanford's Hazy Research group found that the intelligence per joule (IPJ) of local AI inference improved 18-fold from mid-2024 to late 2025, driven by a 3.1x gain from model architecture improvements and a 5.9x boost fr…
In 1997, Sepp Hochreiter and Jürgen Schmidhuber published "Long Short-Term Memory" in Neural Computation, introducing the LSTM architecture that solved the vanishing gradient problem in recurrent neural networks by using…
An open-source, candle-only crypto pump detector, the pump-dump-crypto-screener, is seeking machine learning contributors to improve its detection algorithms. The TypeScript monorepo, available on GitHub, pulls 1-minute …
Researchers from the University of Cambridge, NVIDIA, and Flower Labs have developed the Red Queen Gödel Machine, a method for recursive self-improving AI agents that co-evolves the agent and its evaluator to avoid perfo…
A new open-source project, Founder Population Init, proposes that a tiny founding population of two diploid individuals could generate genetic diversity through a temporary generative inheritance system, called FounderCo…
Canonical's VP of Engineering Jon Seager said Ubuntu's user base inside Windows Subsystem for Linux (WSL) is growing significantly faster than its native desktop user base, and he expects WSL to overtake native Ubuntu in…
Dmitry Grinberg's critique of RISC-V, published on Hacker News, argues the ISA's design is flawed but will still dominate cheap microcontrollers, while Armstrong Subero, an embedded engineer from Trinidad and Tobago, reb…
Reasoning models are deliberately trading world knowledge for reasoning skill, with GLM-5.2 scoring 99.2% on AIME 2026 using about 40 billion active parameters per token, while factual recall remains poor—Gemini 2.5 Pro …
HNSW, the default graph algorithm powering vector search in Faiss, pgvector, Qdrant, and other stores, trades exactness for speed, offering logarithmic retrieval but occasionally returning wrong memories with high confid…
A B.Tech Computer Science (Artificial Intelligence) student has publicly shared a personal roadmap for learning AI, emphasizing foundational programming, mathematics, data handling, and hands-on projects over passive cou…
A developer's analysis of RAG retrieval architectures shows that vector-only search fails on exact-match queries like product codes and error numbers, and that hybrid search combining lexical and vector retrieval, plus r…
A developer detailed a local-first AI architecture that prioritizes on-device computation with cloud APIs as fallback, citing projects like OpenWork that cut API spend by 87% and improved response times for 73% of reques…
A base-2 vs. base-e log-sum-exp mismatch in SGLang's merge v2 API silently corrupts attention outputs for models using MLA with the FlashInfer backend when radix cache hits exceed 8k, affecting DeepSeek V2/V3/R1/V3.1/V3.…
Statistics is a foundational discipline in data science, providing tools to describe data, understand variation, quantify uncertainty, and draw conclusions from evidence. Having a dataset does not automatically mean unde…
SambaNova Systems is hiring a Principal AI Systems Performance Engineer in San Jose, California, to optimize and scale state-of-the-art foundation models on its reconfigurable dataflow platform. The role involves profili…
Unsloth, the open-source fine-tuning library, released Unsloth Desktop, a free beta app for macOS, Windows, and Linux that runs and trains LLMs, diffusion models, and audio models locally, directly competing with Ollama …
A developer built a semantic cache for a support bot that answers repeated questions, addressing the problem that exact-match caches rarely hit because users phrase the same question differently. The cache uses embedding…
SambaNova Systems is hiring a Senior Principal Machine Learning Engineer in San Jose, California, offering $220k–300k/yr for an on-site role focused on optimizing large language models for its SN40L chip and RDU architec…
Towards Data Science published a vendor-neutral blueprint for building a persistent knowledge layer that refuses to guess, addressing the limitation that RAG retrieves but never remembers. The article includes a complete…
AI systems are now solving geometry problems that previously took humans years, according to a report highlighting the shift from calculation to reasoning. By combining deep learning with formal verification languages li…