Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers Zhikai Ding and Ziyi Ye propose Transfer-aware Dynamic Curriculum Sampling (TDCS), a method that adjusts the mix of training examples by difficulty level based on estimated transfer between levels, for large …
An arXiv preprint submitted August 18, 2026, and revised August 19 introduces Co-RL, a cooperative multi-agent reinforcement-learning framework that uses peer-generated rewards instead of ground-truth labels, reporting a…
An eight-author research team proposes Task Specialization Fine-Tuning (TSFT), a method that uses integer linear programming to allocate a limited fine-tuning budget across related task regions, starting from a single pr…
A developer has outlined a DIY method for evaluating AI code review tools, arguing that vendor-published precision and recall numbers are unreliable because they are based on self-selected benchmarks. The method involves…
A paper submitted to arXiv on August 17, 2026, by Kenneth Ge proposes Switching LoRA Adapters as a Tool (SLAaaT), a method that lets AI agents switch between specialized LoRA adapters mid-task. The paper reports that SLA…
Researchers propose DOW-KE, a method for editing a model's learned knowledge by directly optimizing weight updates across multiple layers, reporting stronger results than baselines in five of six model-dataset settings. …
Researchers Minjun Kim and Jong Hak Moon proposed Mr.Dec, a Transformer Decoder-based model that predicts 30-day hospital readmission by representing each admission as a time-aligned sequence of daily electronic health-r…
Researchers introduced Strategic 16K, a 16,000-document corpus of diplomatic cables, and benchmarked six classical and transformer-based models after removing embedded classification markers, in a six-page arXiv paper (a…
PyTorch's built-in GELU function is 20% faster than a hand-rolled version when training GPT-2 small models, according to a developer's benchmark. The same code training the same model on the same data ran at about 21,000…
ServiceNow is hiring a Senior Machine Learning Engineer for its GenAI Search Platform team in Mountain View, California, offering a salary of $161k–274k/yr. The role involves improving search-based question answering sys…
A developer built an AI Pharmacist Assistant that uses OCR to scan drug labels and retrieval-augmented generation (RAG) to detect drug-drug interactions. The system extracts drug names from images, cross-references a SQL…
A new paper on arXiv (2007.13664) proves a universality result for neural network training: if any algorithm can find good weights for a classification task, then an extension of the network can reproduce those weights a…
Empero released Qwen3.8-9B Distill, a full-parameter fine-tune that transfers reasoning from the Qwen3.8 2.4T A95B teacher into the Qwen3.5-9B architecture, trained on roughly 70,000 curated chain-of-thought traces. MMLU…
LM Studio's local LLM performance hinges on whether users run MLX on Apple Silicon Macs or GGUF on Windows GPU laptops, with Apple's unified memory enabling larger models while discrete GPUs like the RTX 5090 offer faste…
Ornith AI, a US open-source lab, released Ornith-1.5, a family of open-weight models in 9B, 35B, and 397B sizes, claiming state-of-the-art open-source performance on coding and agentic tasks. The flagship 397B model scor…
Manticore Search introduced Conversational Search, a feature that combines full-text, vector, and hybrid search with a language model to answer complex product queries, as demonstrated in its Manticore Apparel Shop demo …
A benchmark of Gemma4-E4B on an AMD Ryzen 7040 CPU shows that using 8 threads instead of 16 improves prompt processing from 94.66 to 105.94 tokens per second and token generation from 11.64 to 16.15 tokens per second, in…
A Google Summer of Code 2026 project by Abhay Chaurasiya, mentored by Nina Polshakova and Daneyon Hansen of CNCF, measured the overhead of agentgateway as an inference gateway and found it dramatically outperforms plain …
Researchers from the University of Amsterdam and other institutions propose LINK, a data-level intervention method that improves cross-lingual knowledge transfer in multilingual language models by substituting random wor…
A feature built by Soamee for a SaaS client, costing less than $30 per month to run, reduced support tickets by 40% in the first eight weeks. The system uses basic RAG with OpenAI's text-embedding-3-small and Claude Haik…