Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A developer proposes decoupling AI features from specific model names by defining service objectives that specify task, quality, latency, and cost constraints. This approach allows an intelligence layer to select the bes…
Researchers from Tsinghua University and Z.ai have proposed IndexCache, a method to reduce the computational cost of DeepSeek Sparse Attention (DSA) in GLM-5.2. IndexCache exploits the observation that adjacent layers in…
AI agents often fail not because they lack intelligence, but because they lack memory, a critical issue that requires distinct solutions. The problem of forgetfulness in AI systems is highlighted as a key challenge that …
A new survey paper on arXiv provides a comprehensive taxonomy of memory systems for AI agents, categorizing memory into forms (token-level, parametric, latent), functions (factual, experiential, working), and dynamics. T…
Drifty, an AI focus agent that automatically tracks and classifies computer activity as focus, neutral, or drift, launched on Hacker News. The macOS app records apps, sites, and sessions in three-minute blocks without ma…
A new video compares artificial intelligence to Clever Hans, the horse that appeared to solve math problems by reading human cues, arguing that AI systems similarly exploit statistical patterns rather than demonstrating …
A developer warns that model collapse—where models trained on their own outputs become blander and lose rare patterns—has become a real engineering constraint in 2026. The degradation is driven by statistical sampling er…
Ford rehired about 300 veteran engineers after AI tools failed to address quality issues in production, leading to the company topping the JD Power 2026 US Initial Quality Study for the first time since 2010. The automak…
DeepReinforce AI released Ornith 1.0, an open-source coding model family that scores 82.4 on SWE-Bench Verified, outperforming Claude Opus 4.7's 80.8, and runs the 35B variant locally on a single RTX 4090. The model uses…
A guide to programming Lisp on the DEC PDP-1, one of the earliest interactive Lisp environments, was published. The tutorial covers bringing up the system, loading and saving programs, and mixing assembly code, highlight…
Independent researcher LIMEN project finds that Transformer models follow a universal dynamic grammar of seven conserved state transition motifs, with ambiguity delaying decisional engagement in modern models like Phi-1.…
A developer built a real-time posture tracking system using MediaPipe Pose and Electron. The system captures webcam frames, processes them through a neural network to identify body landmarks, and sends notifications when…
Jua, a Zurich-based physics AI company, uses ClickHouse Cloud to power its physics foundation model EPT-2, cutting forecast delivery time from one hour to 20 minutes and reducing compute costs by a third. The platform de…
Model monitoring in ML production ensures models remain reliable and precise over time by tracking data quality, performance metrics, drift, and resource utilization. This guide covers key components like data quality ch…
Ray Data 2.56 introduces memory-aware execution and improved prefetching to reduce out-of-memory failures and unnecessary object spilling in AI data pipelines. The update includes automatic batch size tuning and better p…
Together AI presented eight papers at ICML 2026 spanning the full AI stack, from frontier agents to GPU kernels. Their research includes DSGym for standardized data-science agent evaluation, ThunderAgent for improved ser…
Researchers evaluated four state-of-the-art zero-shot cross-lingual voice cloning systems across six languages, finding that Arabic remains particularly challenging. The study, presented at IWSLT 2026, analyzed tradeoffs…
Researchers at Ohio State University developed HydraQE, an end-to-end, reference-free quality estimation system for speech translation that outperforms cascaded text-based baselines. Submitted to the IWSLT 2026 shared ta…
Researchers from KIT submitted a cross-lingual voice cloning system to the IWSLT 2026 track, building on FishAudio-S2-Pro with language tag prompting, reinforcement learning fine-tuning, and reference-conditioned lexical…
Elastic acquired Jina AI in late 2025 and released jina-embeddings-v5-omni, a multimodal embedding model that handles text, images, audio, and video in a single vector space. The model achieves state-of-the-art performan…