Full-text search across 12558 articles. Combine with topic and date filters; results sorted by relevance.
Stanford PhD student Marcel Rød released Gigatoken, a Rust BPE tokenizer that encodes text at 24.53 GB/s on a 144-core AMD EPYC 9565 dual-socket setup, outperforming HuggingFace tokenizers by 989x and OpenAI's tiktoken b…
Unsloth achieves up to 7.3x training speedup over standard Transformers for MoE models like gpt-oss-20B on an NVIDIA B200, according to published benchmarks, while Axolotl delivers up to 1.45x speedup and 30% memory redu…
Poolside released Laguna S 2.1, a 118B-parameter open-weight Mixture-of-Experts model with 8B activated parameters per token, achieving 78.5% on SWE-Bench Multilingual and 70.2% on Terminal-Bench 2.1, topping the publish…
Feyn AI, a YC-backed startup, released SQRL, a family of text-to-SQL models that inspect the database before writing a query. The flagship SQRL-35B-A3B achieves 70.6% execution accuracy on BIRD Dev, outperforming Claude …
Alibaba's Qwen team previewed Qwen3.8-Max-Preview, a 2.4 trillion-parameter multimodal model, on July 19, 2026, at the World AI Conference in Shanghai, claiming it is 'second only to Fable 5' among benchmarked systems. T…
Perplexity AI released WANDR (Wide ANd Deep Research), an open benchmark and evaluation harness built around 500 realistic data-collection tasks for knowledge work. WANDR tests whether research agents can discover a larg…
KDnuggets' weekly roundup for July 13, 2026, features articles on Python design patterns, LLM optimization, SQL portfolio projects, Git worktrees for AI development, structured generation with Outlines, local AI agent or…
Zyphra released ZUNA1.1, an EEG foundation model under the Apache 2.0 license that accepts variable-length inputs from 0.5 to 30 seconds, building on its predecessor ZUNA1. The 380M-parameter masked diffusion autoencoder…
Moonshot AI released Kimi K3, a 2.8-trillion-parameter open Mixture-of-Experts model with native vision and a 1-million-token context window, claiming it as the world's first open 3T-class model. The model introduces Kim…
FastCentNN, an upgrade to Centroid Neural Networks, reduces runtime by up to 16% on synthetic 2D datasets and about 5% on high-dimensional ones while maintaining clustering quality, according to its developers. The algor…
Language models generate text by constructing a categorical probability distribution over their vocabulary at each decoding step, converting raw neural network outputs called logits into valid probabilities using the sof…
The Soofi Consortium, coordinated by KI Bundesverband and funded by the German Federal Ministry for Economic Affairs and Energy, released Soofi S 30B-A3B, an open hybrid Mamba-Transformer Mixture-of-Experts foundation mo…
A study on the 'Granularity Paradox' in time-series forecasting finds that finer temporal disaggregation improves in-sample fit but degrades out-of-sample accuracy due to recursive error compounding. Benchmarking 10 mode…
Researchers at the University of Benghazi are conducting a systematic review comparing interpretable statistical models with explainable AI approaches for predicting breast cancer recurrence. The review, registered in PR…
Researchers introduced D2H-AD, a hybrid anomaly detection framework using Hyperdimensional Computing that outperforms five baselines including One-Class SVM and Autoencoders across benchmark datasets. The lightweight mod…
RamAIn, a YC W26 company building AI agents that automate enterprise workflows by interacting with legacy systems and desktop applications, is hiring its first business hire. The startup, founded by Shourya Vir Jain and …
Jon Hyman, co-founder and CTO of Braze, explained on the *Leaders of Code* podcast how he transformed the company's 300-person engineering organization into an AI-first team within months. He noted that over 60% of Braze…
VisionAId, a new Android app, turns smartphones into visual assistants for the visually impaired using six on-device deep learning models and optional cloud-based Google Gemini Flash. The app features few-shot learning f…
Jürgen Schmidhuber's Gödel Machine, a hypothetical self-improving AI that requires proof before modifying itself, remains unimplemented, but its ideas persist in modern AI research through self-modifying agents, meta-lea…
A technical blog post explains that AdamW with warmup and cosine decay is the optimizer almost everyone uses today to train large language models, tracing the evolution from vanilla SGD through momentum and Adam. The pos…