Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Absci, a clinical-stage biotechnology company, is hiring a Senior/Staff AI Scientist for Target Discovery with a salary range of $160k–260k/yr for a remote position in the United States. The role involves developing and …
A developer trained a 125M-parameter transformer for real-time MIDI autocomplete, achieving approximately 108 notes per second on an iPhone 15, modeled after GitHub Copilot's code completion approach but for piano perfor…
Khosla Ventures and Radical Ventures co-led an undisclosed initial funding round for Discovery Loop, the AI research startup founded by Jeff Dean and three Google veterans, with participation from Kleiner Perkins, Lights…
Z.ai and NVIDIA engineers trained GLM-5.2, a 744B-parameter mixture-of-experts model quantized to 4-bit NVFP4, with reinforcement learning to play Super Mario Bros., overcoming arithmetic, distributed-systems, and traini…
VirgilFlow, a Python and eBPF-based infrastructure defense tool, fixed a critical performance bug that caused up to 28% telemetry trace drops under high load. The team used Google AI Studio with Gemini 1.5 Pro to refacto…
A developer fine-tuned OpenAI's whisper-small speech recognition model on 7.7 hours of Mozilla Common Voice Spontaneous Speech 4.0 and made it 2.6 times worse, with word error rate rising from 10.8% to 28.0% on held-out …
Waymo LLC, the autonomous driving unit of Alphabet Inc., revealed details of its in-vehicle compute system, including a new purpose-built 5nm ASIC designed to process raw sensor data for its Waymo Driver. The company sai…
North.cloud released v3, adding Azure support, AI spend tracking, and an ML-powered commitment engine called Autobot to unify cloud, AI, and data costs into one financial view. CEO Matt Biringer said the release reflects…
Lila Sciences is hiring a Research Scientist I/II in Computational Organic Electronics in Cambridge, MA, offering $176k–304k/yr, to apply computational methods and AI to accelerate discovery and design of organic electro…
A new method called COPA reduces attack success rates by 6.3× versus the best static baseline and 4.4× on average across lifelong attack streams, while retaining 92% defense on month-old attacks compared to 61% for repla…
Liquid AI released DSpark draft model checkpoints for three LFM2.5 models, enabling speculative decoding that speeds up inference by up to 3.18x on GPUs and 2.87x on-device without changing output quality. The draft mode…
Swayable, a Y Combinator-backed AI and data science platform founded in 2018, is hiring a Senior AI/ML Engineer for its San Francisco office with a salary range of $175k–250k/yr. The role requires 5+ years of experience …
A developer launched a Speculative Decoding Speedup Calculator in OmniTool Hub to help developers configure local inference setups. The technique, which uses a small draft model to speculate tokens and a larger target mo…
Google researchers introduced the Biomarker Discovery Framework, a multi-agent system that prioritizes candidate biomarkers from wearable sensor data with human oversight. Across three cohorts totaling 9,279 participant-…
Amazon Web Services (AWS) introduced a query-aware compression pattern for Amazon Bedrock that reduces input tokens sent to foundation models in Retrieval Augmented Generation (RAG) applications, cutting costs while pres…
Panasonic Avionics Corporation, in collaboration with AWS and the AWS Generative AI Innovation Center, has built an agentic AI system on Amazon Bedrock, Amazon SageMaker, and AWS Glue to reduce aircraft in-flight enterta…
Nari Labs, Meta, and the open-source community have released text-to-speech models that achieve sub-50 ms latency on consumer hardware, making instant voice for games and streams production-ready. A practical pipeline us…
Towards Data Science reports that row-level chunks, where each body row of a table is paired with its column headers, can serve as the unit of retrieval in retrieval-augmented generation (RAG) systems, often returning th…
Lightricks released LTX-2.5, an open-weights video generation model that uses a diffusion-based video decoder to recover fine details from a highly compressed latent space, eliminating the need for a separate upscaling s…
Nari Labs' Qwen3-TTS 1.7B CustomVoice implementation achieves 10 requests per second and sub-50 ms p95 time-to-first-audio on a single NVIDIA H100 SXM, maintaining real-time playback and costing about $2 per 1M character…