Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Liquid AI released DSpark draft model checkpoints for its LFM2.5 family, claiming up to 3.18x throughput improvement on GPU and up to 2.87x on-device, with day-one support for llama.cpp and SGLang. The draft models, each…
In a YC Paper Club session focused on data, experts from Snorkel AI, Inception AI, and Anthropic discussed the evolving challenges of training data, benchmarks, and multilingual pre-training. Francois Chaubard of Y Combi…
Ornith released the Ornith-1.5 family of open-source coding models on August 20, 2026, claiming its 397B mixture-of-experts flagship scores 86.1 on Terminal-Bench 2.1 and rivals Claude Opus 4.8, though the results are ve…
Metric founder and CEO Hrant Davtyan released ArmBench-ASR on August 20, a benchmark evaluating nearly 30 speech recognition systems on 20.7 hours of Armenian audio across five datasets, with Google's Gemini 2.5 Pro lead…
Forge Nano, a US chip maker specializing in advanced materials, broke ground on an expansion of its lithium-ion battery factory in Morrisville, North Carolina, from 100,000 to nearly 315,000 square feet, with a capacity …
North Cloud Holdings Inc. released North 3.0, adding Microsoft Azure support to complete coverage across all three major hyperscale cloud providers, and integrating AI and data platform costs into a single system. The up…
Google's AlloyDB ScaNN index now scales vector search to 10 billion vectors, a major architectural enhancement featuring a four-level tree (preview) that reduces compute intensity and improves memory efficiency. The new …
NVIDIA's recsys-examples and nv-embedding-cache address the production challenges of generative recommenders (GRs), which reframe recommendation as sequence modeling to handle terabyte-scale user histories, sparsity, col…
Ericsson and SoftBank completed Japan's first trial of Ericsson's AI-native Scheduler for Link Adaptation on SoftBank's commercial 5G network, achieving up to a 25% improvement in spectral efficiency and up to a 50% incr…
Simple AI released a 2,000-hour robotics dataset on August 20 that eliminates the need for a target robot or robot teleoperation during data collection, using its HiFi-UMI system to record human demonstrations with porta…
Jie Tang's team at Z.ai argues that LLM scaling should optimize for fixed inference budgets, favoring smaller, deeper architectures trained longer over wider models. Their ablation shows a 7B model trained on 3T tokens o…
Relay Therapeutics (RLAY) shares have surged as much as 1,074% since bottoming in April 2025, and the Cambridge, Mass.-based biotech firm is now positioned for a fresh breakout, earning a blue dot in MarketSurge for stoc…
The largest single-site psychedelic brain study to date, published in Nature by researchers at Monash University, found 'hidden order' in the brain disruption caused by psilocybin, with brain networks reorganizing into p…
Superwhisper launched its proprietary S1 family of voice-to-text models, including S1-Voice, S1-Language, and S1-mini, with S1-mini's open weights available on Hugging Face. The S1-mini, a 484 MB 0.6B language model, run…
ONNX.css, a new open-source tool, compiles ONNX neural network models into CSS stylesheets, enabling browsers to run inference via the style engine. It demonstrates three models: a LeNet-style CNN for MNIST digit classif…
A mentor with over three years of experience and more than a thousand sessions with data science students outlines three distinct career pathways in AI: Builder (engineering and deployment), Innovator (research and scien…
A developer built pycomplete, a code completer using a 2.48M-parameter transformer blended with n-gram and cache models, and reported top-1 accuracy of 54.6% on held-out files. However, when scoring the ghost text featur…
Astromech, a Dallas startup founded by Colossal Biosciences founders Ben Lamm and George Church, raised $20 million in a funding round led by biotech investor Bob Nelsen, with participation from Peak 6, NeoGenesis Capita…
An engineer explains how AI headshot generators work under the hood, detailing the use of LoRA (Low-Rank Adaptation) to fine-tune diffusion models on a user's selfies. The post highlights the overfitting problem that occ…
Merck and Moderna announced positive topline Phase 3 results for their personalized cancer treatment INTerpath-001, which uses AI to identify up to 34 neoantigens per patient and an mRNA vaccine to target them. The treat…