Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Developer 'davidb' released Scrollcase, an open-source tool that packages Python and machine-learning projects into self-contained, verifiable boxes, eliminating the need for Python installs, pip installs, or Docker. The…
Conversationaly, a privacy-first AI meeting assistant for macOS, Windows, and Linux, records meetings, transcribes them live, and generates summaries entirely on-device by default, with no account, cloud round-trip, or t…
DeepReinforce released Ornith-1.5, a family of open models in 397B, 35B, and 9B sizes, extending the self-scaffolding framework into a closed self-improvement loop where the system proposes tasks, generates scaffolds, an…
WhatsApp is testing Scam Alert, an optional on-device machine learning feature that warns users when messages from non-contacts match scam patterns, using confidential computing, differential privacy, and transparency me…
A developer demonstrates how to run fine-tuned large language models on Apple Silicon using MLX Swift bindings, bypassing CoreML's compilation latency. The approach uses lazy evaluation, direct loading of .safetensors we…
Liquid AI released Q4_0 checkpoints for its LFM2.5 language models trained with quantization-aware distillation (QAD), recovering 96.5% to 97.4% of the accuracy lost to quantization across four models. The QAD checkpoint…
Packt has published 'The Ultimate AI Guide for Linux Engineers' by Ezequiel Lanza and Eduardo Spotti, a 300-page, 12-chapter book aimed at seasoned Linux professionals such as sysadmins and DevOps engineers, covering AI/…
Sarvam AI's Saaras V3 speech recognition model achieved a word error rate of approximately 19.3% on the IndicVoices top-10 language benchmarks, outperforming OpenAI's GPT-4o Transcribe, ElevenLabs' Scribe v2, Google's Ge…
Data leakage in machine learning can cause models to achieve 98% accuracy during evaluation but fail in production, according to a technical explainer. The article warns that preprocessing steps like mean imputation and …
TESSRAL, a deeptech product, enables users to train their own image or voice AI models on personal hardware with small datasets, eliminating the need for massive GPU clusters or cloud compute. The product allows users to…
Allora Network upgraded its mainnet to version 0.17 on August 19, enabling decentralized AI predictions to return labeled outcomes instead of single scalar values. The upgrade maintains backward compatibility for existin…
The developers behind the open-source OmnisBench project have released a benchmark designed to measure the true cost savings of LLM routing. Running 364 coding and math tasks across four models, they found that ideal rou…
A new paper from the GDM Amplified Oversight team shows that debate training reduces reward hacking in reinforcement learning from AI feedback (RLAIF), recovering about 45% of the gap between the peak accuracy of trainin…
VIDRAFT, a Korean AI startup, has released Darwin Family, a training-free model merging framework that uses evolutionary algorithms to combine the parameters of two existing models. Their flagship model, Darwin-27B-Opus,…
A developer built Video Dubbing Translator, a local-first, browser-based pipeline for dubbing short videos that keeps source media on-device. The pipeline separates vocals from background, transcribes with WhisperX, tran…
A developer has built an AI-powered stock trading platform inspired by Zerodha, designed to provide personalized guidance and predictive analytics. The platform uses React.js, Node.js, Express.js, MongoDB, Google Cloud A…
A developer outlines a structured approach to AI at work, proposing a small AI organization with layers of limited authority and explicit handoffs, rather than a single autonomous agent. The design includes a human at th…
Exabel's January 2026 survey of 100 fundamental portfolio managers and analysts managing $610 billion found that 71% cite combining data from different sources as the most frustrating part of working with alternative dat…
Hugging Face released Sentence Transformers v6.0, adding a MultiVectorEncoder that natively supports ColBERT-style late interaction and ColPali visual document retrieval checkpoints. The update enables text-to-image docu…
Hugging Face reported that gpt-oss-120b achieved a 16.1 percentage point task completion gain with only 5% more tokens when using selective memory retrieval instead of full guideline injection, while dumping the full mem…