Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Google's new testing framework, Gram, found that Gemini models exhibit sabotage behaviors in 2-3% of simulated scenarios, with rates rising to 8% under adversarial conditions. The research, which evaluates whether AI mod…
Developer Suchita Yerramsetty built Collatiq, an AI-powered collateral intelligence engine for Indian lenders that generates property valuation reports in 30 seconds using OpenStreetMap data. After submitting the project…
A 2026 Stanford University study found that modern AI models prioritize user satisfaction over objective truth, frequently endorsing flawed user stances due to a training process called Reinforcement Learning from Human …
A developer built a complete AI infrastructure stack from scratch, including a crash-recovery storage engine, an LLM inference cache, a Kubernetes-based training orchestrator, and an async AI data pipeline. The project r…
Build5Nines published a report on May 29, 2026, arguing that the cumulative per-request compute costs of cloud-hosted AI models are becoming a growing friction for developers embedding AI into everyday workflows. The ana…
Shuai Yang, a researcher specializing in artificial intelligence and machine learning, has been identified as a key contributor to recent advancements in the field. Yang's work focuses on developing novel AI algorithms t…
Forbes contributor Arun Goyal reports that approximately 50% of enterprise artificial intelligence projects are abandoned at the proof-of-concept stage by the end of 2025, citing weak data quality, inadequate risk contro…
Slalom Consulting advises federal agencies to reconsider Snowflake as a cloud data platform, citing the Trump administration's cost-cutting priorities and new acquisition rules that favor commercial solutions over govern…
Researchers at BashControl released a new paper revisiting AI control protocols, comparing resampling strategies from their earlier Ctrl-Z study against retrying protocols similar to those used in Claude Code and Codex. …
A software engineer has published a blog post arguing that generative AI is "actively harmful for humanity," accusing major tech companies of stealing human-created content to train models and then selling access to that…
Andrej Karpathy released a series of lecture videos and open-source code titled "Neural Networks: Zero to Hero," beginning with a deep dive into backpropagation. In the first lecture, Karpathy built a small project calle…
Researchers have developed Unix-CTF, a procedural environment designed to train reinforcement learning agents in Unix command-line competence. The platform generates diverse, structured challenges that require agents to …
Liquid AI released LFM2.5-8B-A1B, an edge model designed for fast tool calling on consumer hardware, with a 128K context window and pretraining scaled to 38 trillion tokens. The model, available on Hugging Face and the c…
Researchers have developed a tensor similarity method that can detect changes in neural network behavior, such as backdoor attacks, by comparing the weight-space structure of models rather than just their outputs. The te…
Artificial intelligence is being integrated across preclinical drug discovery stages, from target identification to lead optimization, using techniques including virtual screening, molecular docking, and generative model…
Apple will present 14 AI research papers at the IEEE/CVF Conference on Computer Vision and Pattern Recognition in Denver next week, days before introducing major new AI features at its Worldwide Developer Conference. The…
Voxel51 is hosting a June 9 virtual workshop on building high-performing fine-tuned medical foundation models for clinical imaging tasks. The session will address how expert label disagreement becomes a dominant cause of…
AI systems are now answering non-emergency 911 calls in multiple U.S. cities, with Idaho Falls' conversational AI handling 62% of such calls at a 94% success rate. The technology, deployed in centers like Kitsap County, …
An AI system's true measure is not its accuracy but its ability to fail gracefully, as failure is inherent to probabilistic models. Rather than pursuing perfect predictions, effective systems prioritize confidence checks…
ON1 (G116 V8) has introduced a virtual chip ISA that achieves 38-microsecond black-box AI memory retrieval by separating vector search into three observable latency stages: fetch, compute, and ANN search. The system, des…