Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
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 …
Sebastian Raschka, an AI researcher and author, published a tutorial on building an AI text detector from scratch, including dataset construction, model training, local deployment, and reinforcement learning from verifie…
A technical tutorial demonstrates an end-to-end supervised fine-tuning pipeline for the XYZ-Aquila-SFT dataset using Hugging Face Transformers, PyTorch, and PEFT, culminating in LoRA fine-tuning of Qwen3-0.6B for tool-ca…
Divyakush Punjabi, a full-stack and AI engineer, explains self-attention, the core mechanism behind large language models and modern recommenders, without heavy math. He illustrates the concept with the 'trophy and suitc…
DoorDash's Head of Machine Learning & AI for New Business Verticals, Sudeep Das, presented at QCon AI on shifting from legacy one-shot predictions to an agentic recommendation platform, leveraging language-native consume…
Photoroom, a Y Combinator 2020 cohort company founded in 2019, is hiring a Product ML Engineer in Paris with a salary of €10k/yr, offering a hybrid work model. The role involves owning ML projects end-to-end within produ…
A developer presented a secure and efficient image encryption scheme based on chaotic systems, highlighting the use of chaotic maps for pixel permutation and diffusion. The approach combines sensitivity to initial condit…
NInfer, a from-scratch C++/CUDA inference engine, achieves up to 1,313.8 aggregate decode tokens per second on a single NVIDIA GeForce RTX 5090 for the Qwen3.6-35B-A3B model at concurrency 8, with the Qwen3.6-27B NVFP4 p…
Hugging Face's LeRobot and Strands Agents streamline robotics deployment by using Hugging Face Storage Buckets as a central data hub, eliminating manual file transfers and enabling continuous retraining loops. The integr…
Scaling machine learning models is more complex than simply increasing size, requiring careful engineering to avoid pitfalls, according to a technical essay by Beren Millidge. The author explains that model capacity, tra…
A new blog post from minimallysufficient.com applies the machine learning concept of individual recourse to Pokemon battles, showing how to determine which stats to increase to beat a specific boss. The post builds on a …
A developer is building an AI race engineer in three phases, starting with a pit-wall view from Ergast-compatible F1 data and FastF1 tyre laps, then a baseline classifier to predict pit stops, and finally an explainable …
AWS launched native vector search for Amazon DynamoDB on August 5, 2026, with general availability, enabling developers to store vector embeddings and run approximate nearest neighbor queries directly in DynamoDB. The fe…
Seven researchers posted a preprint on arXiv on 13 August 2026 arguing that reinforcement learning post-training collapses the diversity of a language model's answers, while evolution strategies—optimizing directly in we…
NexusMem, a local memory tool for coding agents, records shell commands, git history, and project docs into a SQLite database, serving ranked, token-budgeted slices on demand. The tool, requiring Node 22 or newer and git…
Moonshot AI's PerceptionBench benchmark shows that no frontier multimodal AI model reaches 60 percent accuracy in visual perception tasks, with GPT-5.6 Sol leading by a narrow margin. The benchmark separates visual perce…
Google DeepMind has developed SL2T, a sign language-to-text translation system that uses a transformer-based architecture to process body keypoints and facial cues, enabling deaf users to sign into phones and receive nat…
Reading time has become a ranking signal in content systems, according to Glad Labs, which found that backfilling word count and reading time was necessary for downstream SEO distribution logic. The company distinguishes…