Tag
The article reports initial manual results from experiments testing procedural skill transfer in small AI models, providing insights into how skills can be transferred across models.
This paper introduces Narrative-UFET, a method that generates short narratives to provide broader context for ultra-fine entity typing, improving performance on long-tail types compared to sentence-level baselines.
A paper presenting a typed, algebraic approach to parsing, likely from the University of Cambridge.
Introduces a paper accepted at PACT 2025, proposing the ComPilot framework, which uses off-the-shelf LLMs as optimization agents to automatically optimize complex loop nests without fine-tuning, achieving a geometric mean speedup of 3.54x, surpassing the SOTA Pluto.
A study from Southeast University found that GLP-1 drugs like Ozempic reverse depression-like behavior in mice by promoting growth of Lactobacillus delbrueckii, which produces endocannabinoids that reduce stress effects.
Miles Brundage shares a link to an AI policy or research update.
A developer outlines three main prep activities in software development (discuss, research, prototype) and asks why AI's usefulness in these wasn't obvious sooner.
A tweet recommending an article on on-policy distillation published on Hugging Face.
A landmark study finds that screen time for babies and toddlers under two can cause long-term developmental harm, leading researchers to call for revised official guidance and a baby screen-time risk assessment.
A University of Maryland research report challenges the narrative that AI is taking jobs, concluding that systematic evidence does not support such fears.
A new approach called Streamable Gaussian Splatting enables real-time streaming of 3D scenes, with a potentially surprising application hinted at with caution.
An MIT study of over 100,000 GitHub developers finds that AI coding tools increase code volume by up to 300% but only boost shipped software by 30%, highlighting bottlenecks in human review and integration.
Researchers discovered a 'megacluster' of genes in Streptomyces that produces four molecules working together to block a key metabolic pathway in bacteria, offering a new strategy against antibiotic resistance.
Researchers at the Arc Institute identified that STING agonists enter and kill T cells through the SLC7A1 arginine transporter, explaining the toxicity at high doses while still showing anti-tumor effects.
A new analysis of five million Microsoft 365 Copilot conversations reveals how people actually use AI at work, presented by Scott Counts.
A dangerous heat wave in Western Europe is raising concerns about its effects on the brain and cognition, with research showing increased irritability, violence, and cognitive impairment, particularly in people with mental-health disorders. Scientists are studying the mechanisms but find it difficult to measure the direct impact of prolonged heat exposure.
A Twitter user recommends a comprehensive book on generative AI covering language modeling, inference optimization, RL, system scaling, and applied concepts like agentic AI and RAG, also sharing advice to read top-cited papers from Papers With Code.
This paper introduces Qwen3-Instruct SAE, a suite of sparse autoencoders trained on Qwen3 instruction-tuned models, enabling the discovery of millions of interpretable features and demonstrating refusal steering capabilities.
This paper introduces BabelTele, a compressed writing style that uses abbreviations, symbols, and mixed-language fragments to reduce text length by 72.1% while preserving 99.5% semantic fidelity for LLMs, arguing that human readability and machine recoverability are separable.
A study by Emory University and IBM Research introduces a verifiable context governance approach for LLMs, achieving 97% accuracy at one-third the cost.