Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
In-flight batching solves the GPU efficiency problem in large language model serving by dynamically adding and removing requests during generation, unlike static batching which wastes resources on finished requests. This…
A new AI safety research team within Arcadia Impact in London has formed over the past four months, growing to eight members who collaborate with the UK AISI alignment team. The team, led by research lead Andrew Draganov…
Google released new Gemma 4 checkpoints optimized with Quantization-Aware Training (QAT) to reduce model memory footprint for local deployment on edge devices and consumer GPUs. The QAT process minimizes quality loss dur…
AI startup Quilty claims its tool can predict a film's box-office success by analyzing an unproduced script and generating a 0-to-100 score for narrative quality, commercial viability, audience resonance, and production …
Quantitative trading firm Jane Street is building and independently financing its own data center, signaling a major escalation in the race for compute power amid surging AI demand. The move follows a $6 billion commitme…
Microsoft AI released a detailed technical report on the development of its first model, MAI-Thinking-1, emphasizing a controlled, reproducible training process built on human-generated data and proprietary scaling laws.…
A GPU training pipeline at a major AI company breached its SLA despite showing 97% GPU utilization across all monitoring tools (Datadog, Grafana, nvidia-smi). Using eBPF kernel tracing, an SRE identified the root cause i…
Large language models (LLMs) frequently exhibit miscalibration, where their reported confidence scores do not match actual accuracy rates, with studies showing mean calibration scores as low as 23.9% in biomedical models…
A developer built a production-grade RAG pipeline that achieves 95.6% retrieval accuracy and 99/100 answer quality by fusing BM25, TF-IDF, and dense vectors with weighted Reciprocal Rank Fusion, addressing the exact-matc…
Artist Trevor Paglen has published a new book, *How to See Like a Machine: Images After AI*, arguing that machine learning has fundamentally altered the function of images, creating an "indexical flip" where any photogra…
Anthropic researchers warned that AI systems could soon improve their own performance faster than humans can supervise them, reviving concerns about the "alignment problem" of ensuring AI reliably pursues human goals. Th…
Researchers have released vLLM, a new inference engine designed to efficiently serve large language models by optimizing memory management and batching. The system achieves up to 24x higher throughput than existing solut…
A developer benchmarked two speech-to-text models on a CPU-only Azure VM and found that using an AI agent to plan and execute the task cut costs by 62%, from $1.96 to $0.74. The agent, Neo, achieved this by researching o…
Developers can optimize spaCy's text processing speed by selectively disabling pipeline components like the dependency parser and lemmatizer when only named entity recognition is needed. The technique reduces processing …
VDF AI's agent networks reduced predicted energy consumption by up to 94.9% compared to a single-model baseline in a controlled benchmark, while maintaining non-inferior output quality within a pre-registered margin. The…
Google Research and Google Cloud introduced a new agentic retrieval-augmented generation (RAG) framework hosted on the Gemini Enterprise Agent Platform, designed to break down complex enterprise queries and iteratively s…
OpenAI's GPT-3.5-Turbo model drops from 90% accuracy to 50% when the correct answer is placed in the middle of a 20,000-token prompt rather than at the start or end, according to research by Liu et al. (2023) published a…
RQM+ and Jordi Labs released Lumo, an in-silico tool that uses machine learning to predict LC-MS and GC-MS response factors for extractables and leachables analysis. The tool routes compounds to chemistry class-specific …
Canonical announced that Ubuntu 26.04 LTS will include single-command installations for NVIDIA CUDA, AMD ROCm, and NVIDIA DOCA-OFED, aiming to reduce deployment friction for AI infrastructure. The company is also partner…
Since 2017, over 300 machine learning papers have adopted the "X Is All You Need" title convention, creating a self-contradictory and growing set of claimed necessities that includes attention, patches, and pretraining o…