Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
The latest update to hermes-memory-installer introduces an auto-repair mechanism for missing embeddings in the gbrain module, selectively rebuilding only affected subsets instead of triggering a full reinstall. This fix …
Cactus Bonsai compresses a 27-billion-parameter AI model to 3.9GB using 1-bit quantization and quantization-aware training, enabling the large language model to run locally on mobile hardware. The technique reduces each …
Simple self-distillation (SSD) improves LLM code generation by fine-tuning models on their own sampled outputs, boosting Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6, with gains concentrated on harde…
Apple researchers developed a personalization system for incremental video search on Apple TV that combines text-based and ID-based embeddings, achieving a 2.99% improvement in NDCG@10 and 3.30% in MRR offline, and stati…
Wang et al. published Agent Workflow Memory on arXiv in 2024 (arXiv:2409.07429), a method enabling agents to induce reusable workflows from past trajectories offline or online without labels, reducing steps to solve new …
EndoSeg, an ovarian-lesion ultrasound segmenter built for the Nebius Serverless AI Builders Challenge, achieves a test Dice of 0.755 but cannot diagnose endometriosis or classify lesion types. The model outputs only a bi…
A developer argues that AI has made writing test scripts easy, shifting the hard part of test automation to building systems that remain useful months later as UIs change, teams grow, and original authors move on. The de…
A new speedrun benchmark for training reinforcement learning agents to solve Sokoban puzzles has been released, with two tracks: an LLM track fine-tuning Qwen3-4B-Instruct-2507 on 8xH100 GPUs to achieve >80% held-out pas…
Researchers have released Rzk, a proof assistant implementing a refinement of Riehl and Shulman's simplicial type theory (RSTT) for synthetic reasoning about ∞-categories. The tool translates RSTT proofs faithfully and c…
Emerging AI, machine learning, and advanced data analytics now offer real-time tools to detect and mitigate insider threats, but organizations must adopt comprehensive strategies to benefit. The convergence of these tech…
Whatnot acquired Shaped, a machine-learning company for real-time recommendations and search, to accelerate live-shopping discovery. Shaped founder Tullie Murrell and nearly a dozen engineers and AI researchers will join…
MLOps tooling continues to improve, but three hard problems remain unsolved at production scale: emission control that prevents unwanted material at the source rather than filtering after the fact, generalizable safe fil…
AI researchers are increasingly studying how babies learn, believing that infant brain architecture could hold the blueprint for the next phase of artificial intelligence. Unlike current AI models like GPT-4 that rely on…
Lattice is an open-source, pure Rust inference and training engine optimized for Apple Silicon, enabling developers to run, quantize, and fine-tune LLMs locally with zero external dependencies. It leverages Metal Shaders…
A developer audited an AI-assisted refactor of the open-source CLI tool YouMindAG and found 46 undefined references across 7 files, silently breaking 7 of 14 commands. The refactor reduced a 1,920-line entrypoint to 410 …
ICF and Snowflake modernized a federal health program's data foundation, reducing ETL processing time by more than 75% and cutting monthly data platform costs by 80% from approximately $30,000 to $6,000. The zero-downtim…
A developer published a field-tested method for auditing Rails + PostgreSQL performance without APM, using pg_stat_statements and pg_stat_activity. The playbook explains why cloud console dashboards lie—due to fingerprin…
A developer's guide explains why edge AI benchmarking on Android requires measuring thermal throttling, memory pressure, and accuracy loss—not just execution time—and recommends using Kotlin 2.x with AICore for productio…
PrismML released Bonsai 27B, a 27-billion-parameter AI model compressed to 3.9 GB that runs on an iPhone 17 Pro Max at 11 tokens per second, the first model at that capability tier to fit on a smartphone. The ternary var…
Brown, a key figure in launching one of Amazon Web Services' earliest offerings and overseeing its compute and machine learning divisions, has left AWS, creating a leadership void in the cloud giant's AI strategy. His te…