Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Epoch AI measures AI performance costs falling about 13x per year, while MIT puts annual algorithmic progress at roughly 3x after stripping out hardware gains and competition, according to The Decoder. The report notes t…
Google's MaxText team reproduced AI2's OLMo 3 7B language model from scratch on Google Cloud TPUs, matching AI2's published loss curve over the full ~5.93T-token / 1.41M-step stage-1 pre-training budget and validating th…
An independent experiment fine-tuned a 135M-parameter model on the 10,003-message Banking77 dataset in under 30 minutes on a 16 GB Apple MacBook Air, lifting exact intent accuracy from 1.62% to 83.25% and operational dec…
NVIDIA published a tutorial showing how its Transformer Engine (TE) and BioNeMo MoE recipe cut the overhead of training mixture-of-experts biological foundation models, using GroupedLinear to submit expert work as one gr…
A technical post demonstrates Dynamic Abliteration, a runtime refusal-suppression method that uses PyTorch forward hooks to inject multi-layer contrastive steering vectors into Qwen3-4B's residual stream across layers 12…
SEO practitioner Dejan published a reproduction of Google's query fanout research, training a vector-based diffusion model that replaces text fanouts with embeddings for document retrieval and releasing it on Hugging Fac…
Black Forest Labs released FLUX 3 Action, an open-weight 7B world-action model derived from its multimodal FLUX 3 backbone, which sets a new state-of-the-art success rate on the RoboLab-120 leaderboard at less than half …
A September 23, 2026 arXiv paper introduces ARMS (Always-on Robot in Multi-modal Streams), a streaming policy built on a single pretrained π0.5 backbone plus three lightweight modules that let a dual-arm robot watch live…
A paper submitted to arXiv on 23 Sep 2026 proposes the Agent-Editing World Model (AEWM), which models how reasoning and actions shape future task progress instead of simulating tool responses, addressing task-state conta…
Apple detailed its production use of homomorphic encryption (HE) combined with private information retrieval (PIR) and private nearest neighbor search (PNNS) to power Enhanced Visual Search in Photos, and open-sourced it…
A custom self-attention in-memory computing architecture built on charge-based gain cells reduces attention latency by up to two orders of magnitude and energy consumption by up to four orders of magnitude compared with …
Check Point researchers found that every configuration of TypeSafe AI's Jev decision model could be manipulated via prompt injection, downgrading a high-risk due diligence report on the fictional PonziCorp Financial Serv…
Google announced Retrieve-for-Train-Diffusion (R4T), a three-stage query fan-out framework built on a 53.9M-parameter diffusion model that the company says delivers a 12 to 20 speedup over autoregressive approaches. Goog…
Jacky Kwok announced Contrastive Language Model (CLM), an 8-billion-parameter "System One" model trained with a contrastive learning objective linking states and actions, which delivers up to 9× faster inference than Jev…
Representation Autoencoders (RAEs) let diffusion models operate in the feature spaces of pretrained visual encoders, but many off-the-shelf encoders are not optimized for faithful reconstruction and discard fine-grained …
A user of Japanese-English machine translation rated ChatGPT at roughly 80 out of 100 and DeepL at about 60 for one-shot paragraph translation, arguing the technology is very good but not "solved." The user, posting on L…
Researchers introduced PackLab, a framework for developing, training, and evaluating multimodal large language models (MLLMs) in robotic bin packing, addressing the long-horizon sequential decision-making problem where e…
A developer who built InsNet, a ~21,000-line C++14 deep learning library with 4,828 lines of hand-written CUDA kernels, starting in 2018, describes how its padding-free dynamic batching design anticipated the flat, paddi…
Researchers submitted LensVLM, an inference framework and post-training recipe built on Qwen3.5-9B-Base, to arXiv on 7 May 2026, enabling vision language models to scan compressed rendered text and selectively expand onl…
Researchers developed EMBODIEDSWE-BENCH, a simulation benchmark for coding agents that spans contact-rich manipulation and deformable object tasks in long-horizon, dexterous robotics, to test whether agent-generated solu…