Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Qdrant's filtered vector search benchmark on one million deep-image-96 vectors shows that ACORN-1, a query-time algorithm, repairs graph connectivity gaps left by Filterable HNSW's index-time extra edges, with HNSW build…
An engineer has built AI-Autofy, an AI customer-support SaaS using Django, RAG, and self-hosted LLMs. The architecture separates the AI inference layer from the main web application to allow independent scaling, and uses…
A study by researchers testing the Coconut and CODI models found that these latent reasoning models barely use their hidden reasoning steps for logical tasks like PrOntoQA and ProsQA, with performance stemming from train…
A Hacker News user asks what developers would do with a hypothetical CPU-friendly content generator that is orders of magnitude faster than a transformer, runs on an ordinary HP PC without H100s or 64GB RAM, offers long …
Netflix's GenRec, an LLM-backed ranker in the 1B–10B parameter range, beat its mature production recommender system in a four-week A/B test on roughly 10% of live traffic, posting statistically significant gains on both …
Google unveiled its 8th-generation Tensor Processing Unit (TPU) at Cloud Next '26, shipping in two variants for the first time: TPU 8t for training and TPU 8i for inference, both sharing Axion CPUs, liquid cooling, and t…
Hugging Face, Reddit, Kaggle, PromptCube, Discord, Stack Overflow, Dev.to, and GitHub Discussions are the best free AI learning communities for 2025, according to a new guide. Hugging Face is the primary global hub for o…
Pipe has released docs-pipe, a documentation-native RAG module and web dashboard that turns Markdown documentation into a searchable, question-answering AI with heading-aware chunking, hybrid search, cited answers, and a…
Andréas, a full-stack developer and CTO, benchmarked his local AI agent runner and found that the concurrency bottleneck was not the model server or hardware but the scheduling policy. The runner's event loop handled req…
A developer's evaluation of three agent memory frameworks—file-based, structured store, and reinforcement-learning-trained experience—found that the file-based approach, as implemented in OpenClaw's memory-core plugin, o…
DuckDB's official vss extension brings HNSW-based approximate nearest neighbor search directly to its SQL engine, allowing developers to run vector similarity queries without a separate vector database. The extension sup…
A developer's RAG pipeline for MockEvalio failed on its first day when the embedding endpoint returned a 404, prompting a switch from Groq's nomic-embed-text-v1.5 to OpenAI's text-embedding-3-small. The fix also replaced…
A developer spent one day training a machine learning model to predict machine failure but considerably longer turning it into production-ready software, reporting that the model achieved 74.20% accuracy with feature sel…
Z.AI's GLM-5.3 desktop auditor, launched in partnership with Z.AI, offers free unmetered access to the GLM-5.3 API until October 2026, and has autonomously discovered 2,436 zero-day and legacy vulnerabilities across 269 …
Cold Spring Harbor Laboratory researchers propose that the brain's wiring can be understood as an algorithmic problem: a genome-sized program (~1 gigabyte) that a single cell executes to build a brain within a year. They…
Divyakush Punjabi, a full-stack and AI engineer, explains that multimodal AI models do not gain a new visual faculty but instead translate images and audio into the same vector space used for text, allowing attention mec…
The Finbold AI Agent, which aggregates forecasts from DeepSeek Chat, GPT-5.7 Luna, and Grok 4.5, predicts XRP will trade at an average price of $0.97 on September 1, 2026, a decline of nearly 3% from current levels. The …
BDH-CQ, a 150M-parameter model, achieved a 29.5% pass@2 score on the ARC-AGI-1 benchmark at a cost of $0.00070 per task, demonstrating that recurrent latent state reasoning can outperform larger models on abstract reason…
Sankalp placed 12th among 183 entrants with a time of 1,804.779 microseconds on an NVIDIA B200 in GPU Mode's qr_v2 competition, using OpenAI Codex to make over 1,500 submissions over 14 days. His result was roughly 48% s…
Homomorphic encryption has moved from theory to production, enabling private AI inference where servers compute on encrypted data without exposing weights or queries. The CKKS scheme supports encrypted vector operations …