Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Monarch Chrysalis v1, a sparse mixture-of-experts model with latent-space reasoning, has completed its first training run and now runs on a consumer desktop CPU, according to its developers. The 6.93B-parameter model use…
Continuous diffusion language models (CDLMs) are making a comeback in AI research, according to a technical blog post by Sander Dieleman. The post notes that after being largely supplanted by discrete diffusion methods, …
Artur Poniedziałek, an IT project manager, launched Best in IT, a blog focused on practical AI, automation, and developer tools. The blog covers local AI models, RAG systems, AI agents, and automation workflows, with an …
A tutorial by an unnamed author demonstrates that a bigram model with a single embedding table can recover true transition probabilities from 100,000 generated samples, achieving learned probabilities close to the true m…
A Hacker News user posted a survey asking developers about their machine learning practices, including programming languages, model architectures, training methods, and familiarity with technical terms. The post invites …
Tencent's WeChat Vision team released WeMM-Embedding, a family of multimodal embedding models in 2B, 4B, and 9B parameter sizes, supporting text, images, videos, visual documents, and interleaved inputs. The models suppo…
Piqt, a new iOS app from developer Piqt, uses on-device machine learning and Apple's Vision API to curate and rank photos without cloud uploads, staging 'junk' in a separate album and ensuring it cannot delete files dire…
A solo entrepreneur and ML engineer launched AnVendor, a service that reveals which SaaS products any company uses and estimates how much it pays for them, covering about 700 services. The founder says the engine uses no…
A developer added a fourth model, Mistral Small 3.2, midway through a field test of AdversarialDebate, changing the experiment's outcome. The addition revealed that maximum diversity can lead to capitulation cascades, an…
Wayve, a developer of Embodied AI technology for autonomous driving, is hiring a senior Software Engineer for its Data Flywheel Platform in Sunnyvale, California, to build the infrastructure that turns world-scale fleet …
Shrijith Venkatramana, an engineer building LiveReview, explains how the Adam and AdamW optimizers became essential to modern LLM training. The post details how Adam, introduced in 2014 by Diederik Kingma and Jimmy Ba, c…
A developer has released Quipu, an open-source library that implements hybrid encryption by combining classical and post-quantum algorithms. The library uses X25519 with ML-KEM-1024 for key exchange and Ed25519 with ML-D…
AI-driven software testing is transforming quality assurance in 2026, with tools like Testim, Mabl, and Functionize cutting manual test creation time and reducing flaky failures. Companies such as Atlassian and Netflix r…
A new open-source repository, 'social-app-teardown' by user elchrysaki, claims to provide fully sourced research on how Threads, LinkedIn, Instagram, X, and Reddit actually work, including every feature, real ranking alg…
Maneshwar, the developer behind LiveReview, an AI code review tool, explains the differences between CPU, GPU, TPU, NPU, DPU, and QPU, arguing that 'computation' is not a single thing. He details how CPUs handle unpredic…
Revise, a zero-code desktop ML platform, enables health AI teams to clean data, train models, and validate them locally without sending data to the cloud, addressing IRB and HIPAA constraints. Currently used at two top m…
A developer detailed the process of designing a Retrieval-Augmented Generation (RAG) system from first principles, aiming to create a unified knowledge base across scattered sources like Logseq, Gmail, and Notion. The po…
A developer reverse-engineered NVIDIA's DLSS5 binary to infer its model architecture, revealing a heterogeneous, hierarchical, recurrent, single-pass neural renderer with tiled Swin transformer families and a compound 16…
Towards Data Science reports that context engineering guidelines are evolving, with implications for data scientists' daily workflows. The article outlines how practitioners can apply the latest context engineering pract…
Tencent researchers found that non-thinking inference modes in hybrid multimodal AI models increase response failures by up to 48.64%, according to a new paper on arXiv introducing the PatternEval benchmark. The benchmar…